Deliver High Quality High Performance HEVC via Intel® Media Server Studio

Deliver High Quality, High 

Performance HEVC via 

Intel® Media Server Studio 

A white paper unlocking new video technology opportunities! 

HEVC is an exciting, cutting edge, highly efficient, new video compression technology enabling 

next generation of digital media applications, products and services. Intel is at the forefront of 

this development, leading the HEVC technology revolution. Intel Media Server Studio 2015 aims to 

offer industry leading, among the best in the class developer focused HEVC solutions with the 

best tradeoff of quality versus performance. This paper introduces the capabilities of Intel’s 

developer HEVC product offering, the Intel Media Server Studio HEVC Encoder and Decoder, 

currently in its R7 release which now packs support for coding of 4:2:0 8-bit, 4:2:0 10 bit, 4:2:2 8- 

bit, and 4:2:2 10-bit content. The paper also identifies the opportunities due to unleashing of a 

range of extremely powerful, higher performance HEVC solutions suited for different 

applications, services, eco-systems, and devices. 

Nov. 11 2015, v1.52 

Author: Atul Puri

LEGAL DISCLAIMER 

INFORMATION IN THIS DOCUMENT IS PROVIDED IN CONNECTION WITH INTEL PRODUCTS. NO LICENSE, 

EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS 

GRANTED BY THIS DOCUMENT. EXCEPT AS PROVIDED IN INTEL'S TERMS AND CONDITIONS OF SALE FOR 

SUCH PRODUCTS, INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR 

IMPLIED WARRANTY, RELATING TO SALE AND/OR USE OF INTEL PRODUCTS INCLUDING LIABILITY OR 

WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR 

INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. 

UNLESS OTHERWISE AGREED IN WRITING BY INTEL, THE INTEL PRODUCTS ARE NOT DESIGNED NOR 

INTENDED FOR ANY APPLICATION IN WHICH THE FAILURE OF THE INTEL PRODUCT COULD CREATE A 

SITUATION WHERE PERSONAL INJURY OR DEATH MAY OCCUR. 

Intel may make changes to specifications and product descriptions at any time, without notice. Designers 

must not rely on the absence or characteristics of any features or instructions marked "reserved" or 

"undefined." Intel reserves these for future definition and shall have no responsibility whatsoever for 

conflicts or incompatibilities arising from future changes to them. The information here is subject to 

change without notice. Do not finalize a design with this information. 

The products described in this document may contain design defects or errors known as errata which may 

cause the product to deviate from published specifications. Current characterized errata are available on 

request. 

Contact your local Intel sales office or your distributor to obtain the latest specifications and before 

placing your product order. 

Copies of documents which have an order number and are referenced in this document, or other Intel 

literature, may be obtained by calling 1-800-548-4725, or by visiting Intel's Web Site. 

MPEG is an international standard for video compression/decompression promoted by ISO. 

Implementations of MPEG CODECs, or MPEG enabled platforms may require licenses from various 

entities, including Intel Corporation. 

Deliver High Quality, High Performance HEVC via Intel® Media Server Studio 

Intel, the Intel logo, Intel Core are trademarks or registered trademarks of Intel Corporation or its 

subsidiaries in the United States and other countries. 

OpenCL and the OpenCL logo are trademarks of Apple Inc. used by permission by Khronos. 

Copyright © 2008-2013, Intel Corporation. All Rights reserved. 

1 

Optimization Notice 

*Other names and brands may be claimed as property of others.

Intel's compilers may or may not optimize to the same degree for non-Intel microprocessors for 

optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and 

SSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or 

effectiveness of any optimization on microprocessors not manufactured by Intel. 

Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. 

Certain optimizations, not specific to Intel microarchitecture, are reserved for Intel microprocessors. 

Please refer to the applicable product User and Reference Guides for more information regarding the 

specific instruction sets covered by this notice. 


Notice revision #20110804 

2

Table of Contents 

Introduction ........................................................................................................................................................ 5 

HEVC Compression Basics .................................................................................................................................. 6 

HEVC Data Hierarchy ........................................................................................................................................... 6 

HEVC Partitioning, Prediction and Coding Technologies ...................................................................................7 

Intel® Media Server Studio 2015 Overview ...................................................................................................... 13 

Intel® Media Server Studio 2015 SDK ................................................................................................................ 15 

Intel® Media Server Studio HEVC Software Encoder Quality Evaluation for 4:2:0 8-bit Content ............... 18 

Quality Evaluation Tests for 4:2:0 8-bit Content ............................................................................................... 18 

Quality Evaluation Results for 4:2:0 8-bit Content ........................................................................................... 20 

Intel® Media Server Studio HEVC Encoder Quality vs Performance Tradeoffs for 4:2:0 8-bit Content ..... 28 

HEVC Software Encoder Performance for 4:2:0 8-bit Content ....................................................................... 29 

Quality vs Performance Tradeoff for TU Modes for 4:2:0 8-bit Content ......................................................... 32 

Intel® Media Server Studio HEVC Software Decoder Performance for 4:2:0 8-bit Content ....................... 34 

Intel® Media Server Studio HEVC Software Encoding Computational Scalability Tests (on E5 Server) ... 36 

Quality Evaluation Results for Computational Scalability Tests for 4:2:0 8-bit .............................................. 36 

Quality vs Performance Tradeoff for TU Modes for Computational Scalability Tests for 4:2:0 8-bit ............ 37 

Intel® Media Server Studio HEVC Software Encoder Quality Evaluation for 4:2:0 10-bit Content ............. 42 

Quality Evaluation Tests for 4:2:0 10-bit Content ............................................................................................. 42 

Quality Evaluation Results for 4:2:0 10-bit Content ......................................................................................... 43 

Intel® Media Server Studio HEVC Encoder Quality vs Performance Tradeoffs for 4:2:0 10-bit Content ... 47 

HEVC Software Encoder Performance for 4:2:0 10-bit Content ...................................................................... 47 

Quality vs Performance Tradeoff for TU Modes for 4:2:0 10-bit Content ...................................................... 49 

Intel® Media Server Studio HEVC Software Encoder Quality Evaluation for 4:2:2 8-bit Content ............... 52 

Quality Evaluation Tests for 4:2:2 8-bit Content ............................................................................................... 52 

Quality Evaluation Results for 4:2:2 8-bit Content ............................................................................................53 

Intel® Media Server Studio HEVC Encoder Quality vs Performance Tradeoffs for 4:2:2 8-bit Content ...... 57 

HEVC Software Encoder Performance for 4:2:2 8-bit Content ......................................................................... 57 

Quality vs Performance Tradeoff for TU Modes for 4:2:2 8-bit Content ........................................................ 60 

Intel® Media Server Studio HEVC Software Encoder Quality Evaluation for 4:2:2 10-bit Content ............. 62 

Quality Evaluation Tests for 4:2:2 10-bit Content ............................................................................................. 62 

Quality Evaluation Results for 4:2:2 10-bit Content.......................................................................................... 62 


3 



Intel® Media Server Studio HEVC Encoder Quality vs Performance Tradeoffs for 4:2:2 10-bit Content .... 65 

HEVC Software Encoder Performance for 4:2:2 10-bit Content ....................................................................... 65 

Quality vs Performance Tradeoff for TU Modes for 4:2:2 10-bit Content ....................................................... 67 

Summary ........................................................................................................................................................... 68 

References ........................................................................................................................................................ 68 

Appendix A......................................................................................................................................................... 69 

A1: Summary of 4:2:0 8-bit Coding Quality and Performance Tests ............................................................ 69 

A2: Summary of Computational Scalability Tests ............................................................................................71 

Appendix B .......................................................................................................................................................... 72 

B1: Summary of 4:2:0 10-bit Coding Quality and Performance Tests ............................................................. 72 

B2: Summary of 4:2:2 8-bit Coding Quality and Performance Tests .............................................................. 73 

B3: Summary of 4:2:2 10-bit Coding Quality and Performance Tests ............................................................. 74 

Appendix C .......................................................................................................................................................... 75 

C1: Example Encoding Directory and Subdirectories structure ..................................................................... 75 

C2: Scripts for 4:2:0 content Encoding with HEVC HM Encoder .................................................................... 76 

C3: Scripts for 4:2:0 content Encoding with Intel® Media Server Studio HEVC Encoder ............................ 80 

C4: Script for HEVC BD rate Computation ...................................................................................................... 84 

4

Deliver High Quality, High 

Performance HEVC via Intel® 

Media Server Studio 

A white paper unlocking new video technology opportunities! 

Introduction 

HEVC (aka, H.265) [1-4] is a new, highly efficient, video compression standard from ISO MPEG that promises 

substantially higher compression over H.264 (aka, AVC) [4,6], its previous generation standard completed around 

10 years ago. In particular HEVC promises roughly a factor of 2 in compression over H.264, that had delivered a 

factor of 2 in compression over MPEG-2, its earlier generation standard. H.264 is currently dominant having 

supplemented or displaced MPEG-2 in nearly all digital video applications, services, products, and eco-systems. 

Over next few years the time seems ripe for HEVC due to its advantages, to supplement or displace H.264 in the 

same manner. Overall MPEG has an excellent history [4] of delivering on video standards that have a wide 

industry following. 

Intel® Media Server Studio SDK (formerly, Media SDK) is a well-known developer product that implements state 

of art standards based highly optimized decoders, corresponding efficient and highly optimized encoders, 

file/stream formatting, and pre- and postprocessing tools supporting efficient coding. Intel® Media Server Studio 

SDK implements many Codec and tools components initially in software, and later as hybrid (of software and 

hardware) or entirely in hardware. The reason for this multi-tier approach is faster time to market for software 

solutions, followed by hybrid solutions that contains partial hardware acceleration, and lastly blazingly fast 

hardware solutions that scale. Intel® Media Server Studio SDK 2015 is available both for Windows and Linux. It 

supports Intel® 4 th and 5 th generation Core and Xeon ® Processor based platforms with Intel Iris Pro and Intel 

HD Graphics. 

Intel® Media Server Studio 2015 is just being released and includes a number of significant tools, technologies, 

and enhancements including improved software implementation of HEVC Encoders and Decoders. Since not all 

HEVC implementations are created equal, this white paper attempts to quantify the quality and performance a 

developer should expect from Intel® Media Server Studio HEVC Codec software implementation. Rest of the 

white paper is organized as per following sections. 

HEVC Compression Basics 

Intel® Media Server Studio Overview 

Intel® Media Server Studio HEVC R7 release (Codec Quality, Encoder Quality vs Performance Tradeoffs, 

and Decoder Performance) results presented in 4 parts as follows. 

o Part 1: 4:2:0 8-bit Coding Quality, Performance, and Computational Scalability Tests 

o Part 2: 4:2:0 10-bit Coding Quality, and Performance Tests 

o Part 3: 4:2:2 8-bit, and Performance Tests 

o Part 4: 4:2:2 10-bit, and Performance Tests 


5 


Appendix A and B summarize quality & performance results. Appendix C discusses test scripts 

HEVC Compression Basics 


HEVC builds on the well-known classical interframe coding framework of block motion compensated transform 

coding. However unlike previous MPEG/ITU-T standards including H.264 instead of using smaller, fixed size 

processing based on macroblocks and blocks for motion compensated prediction and small block transform 

coding, it uses larger, flexible structures that are partitioned for motion compensated prediction, and a range 

of large to small block sizes for transform coding. There are other significant difference as well. 

HEVC Data Hierarchy 

Figure 1 shows high level data structure hierarchy. The terms Video, GOP and Picture as shown are only 

conceptual while Slices and lower layers are actual layers employed by HEVC. 

Video 

Temporal Structure (GOP) 

[pictures grouped for coding] 

Picture 

[1 slice or a sequence of slices. Types I, P, B, or Generalized B] 

Slices 

[1 slice segment or a sequence of slice segments] 

Slice Segment (SS) 

[Dependent or Independent Slice Segments. A sequence of CTUs] 

Coding Tree Unit/Block (CTU/CTB) 

[Starting: 64x64 or 32x32 or 16x16. Recursively QT Split down to 8x8] 

Coding Unit (CU) 

[Sizes: 64x64 (inter only), 32x32, 16x16, 8x8. CU carries intra/inter/skip mode. Inter 

CU’s non-recursive split into 1-4 PUs per one of 8 modes. Intra CUs allow only square 

PUs and go down to 4x4. inter 8x8 CU allow only 8x4 or 4x8 splits but do not support 

bidir pred for these partition sizes. Residual CU recursive QT split up to 4x4 TU] 

6 

Prediction Unit (PU) – Intra partitions: QT – 35 pred dir. Transform Unit (TU) 

Inter partitions: 

[32x32, 16x16, 8x8: DCT, 4x4: DCT & DST] 

Figure 1 HEVC data hierarchy. Video, GOP & Pictures are conceptual, others are actual layers.

HEVC Partitioning, Prediction and Coding Technologies 

We now introduce the significant components of HEVC processing structures as well as discuss actual 

video coding algorithms. Due to significant amount of details only high level concepts are covered. 

Further, a 2 column visual presentation style is used that shows for each topic, a key concept shown in the 

first column and a related illustration in the second column showing. Since this section is a brief overview 

of the standard, the concepts are simplified and not necessarily covered in extreme detail. 

Coding Tree Unit/Block (CTU/CTB) 

Defined at a high level 

A CTU consists of 3 CTBs (1 luma plus 

2 chroma) 

Luma CTB starting size one of 

o 64x64 

o 32x32 

o 16x16 

Corresponding Chroma CTB, half in 

size horizontally and vertically 

Luma CTB split by recursive 

QuadTree down to 8x8 

Coding Unit (CU) 

Always Square 

o Largest CU (LCU) as big as 

size of luma CTB 

o As small as 8x8 

o Sizes: 64x64, 32x32, 16x16, 

8x8 

Traversed in Zig-zag order 

Types: Intra, Inter, Skip 

Intra CU 

o Largest size 32x32 

o Partitioned in to square 

Prediction Units (PU) up to 

4x4 

Inter CU 

o Largest size 64x64 

o 8x8 CU parttioned into 8x4, 

and 4x8 PUs only; no 

bidirectional pred 

Figure 2A Luma CTB starting size options 

Figure 2B Partitioning of a luma CTU into CUs 


7 



Prediction Unit (PU) 

CU partitioning into Prediction 

partitions is nonrecursive 

Intra CUs partitioned into square 

Prediction partitions 

o 32x32 

o 16x16 

o 8x8 

o 4x4 

Inter CU (64x64, 32x32, 16x16, 8x8) 

partitioned into 

1 of 8 Prediction partition modes 

o Partitioned into 1, 2, or 4 

partitions 

o 8x8 PU is partitioned into 

8x4, 4x8 only; 

also no bidirectional 

prediction mode for 8x8 PUs 

Using PU partitions, a residual CU is 

constructed prior to coding 

Transform Unit (TU) 

Residual CU, QuadTree recursively 

split into TUs 

TUs of following sizes (no 64x64 TU) 

o 32x32 

o 16x16 

o 8x8 

o 4x4 

Chroma TU of 1/4 th size of luma TU 

but smallest 

TU for chroma is 4x4 (no 2x2 TUs for 

chroma) 

TU of size 4x4 flagged by coded/not 

coded 

DCT Transform on all TU sizes (32x32, 

16x16, 8x8, 4x4) 

DST Transform on size 4x4 TU 

Figure 2C Intra and Inter PU examples 

Figure 2D Partitioning of a CU into TUs 

8

Intra Coded PU 

Each Intra Coded PU 

o Pred mode for Luma 

o Pred mode for 

Chroma 

All TUs in a PU use the same 

mode 

For Luma candidate choices 

for prediction mode 

o Planar 

o DC 

o 33 Angular Pred 

Directions 

For Chroma candidate choices 

for prediction mode 

o Planar 

o DC 

o Hor 

o Vert 

o Luma pred mode copy 

Inter Coded PU 

Motion Pars specified explicitly or 

implicitly 

o Motion vector 

o Ref Picture Index 

o Picture List Usage Flag 

 

For inter coded CU with 

PredMode=Skip, CU coded with no 

transform coeff, or motion vector, 

and ref picture flag, and ref picture 

list usage obtained by motion merge. 

Figure 2E Intra Luma prediction directions 


 

For inter coded CU with 

PredMode=Inter, either use Motion 

merge or explicit motion pars 

Figure 2F Inter Coded PU Partitionings 

9 



Motion Merge 

Spatial Merge Candidates 

o 5 positions 

o Select 4 Candidates 

o Remove Partition 

Redundancy 

 

 

Temporal Merge Candidates 

o 2 positions 

o Select 1 candidate 

Merge Process 

o Remove duplicates from 

Spatial and Temporal 

Candidates 

o Add combined bi-predictive 

candidates 

o Add nonscaled bi-predictive 

candidates 

o Add zero merge candidates 

o Final merge candidates 

Transforms 

4x4 integer DST approx. Size 4 basis 

matrix shown on right. 

 

 

4x4 integer DCT approx. Size 4 basis 


8x8 integer DCT approx. Size 8 basis 


16x16 integer DCT approx. Size 16 

basis matrix shown on right. 

col_ref 

curr_ref 

tb 

td 

curr_pic 

curr_PU 

B 2 

A0 

(a) second PU of Nx2N 

col_pic 

col_PU 

current 

PU 

B1 

B0 

TL 

B2 

A1 

A0 

C0 

current PU 

(b)second PU of 2NxN 

current PU 

C3 

LCU boundary 

Figure 3A Spatial Merge candidates position, 

position of second PU of Nx2N, and 2NxN, MV 

scaling of temporal merge, coding of spatial merge, 

Temporal Merge candidates C3 and H 

BR 

H 

B0 

10 

32x32 ineger DCT approx. Size 32 

basis matrix not shown. 

Figure 3B Transform basis matrices

Interpolation Filter 

Luma 

o ¼ pel interpolation 

o 7/8 tap filter 

Figure 3C 4-tap DCT/IF Luma Filter, and 8 tap 

 

Chroma 

o 1/8 pel interpolation 

o 4 tap filter 

Deblock Filtering 

Overall Process 

 

 

 

Boundary strength calculation 

o Based on if P or Q is intra, P 

or Q has nonzero coef, Pand 

Q have different ref, P and Q 

have different num of MVs.. 

o 3 levels of strength 0, 1, 2 

Threshold value β and Tc calculation 

from input Q 

Filter on/off Decision 


Figure 3D Deblock filtering in HEVC 

11 


Sample Adaptive Offset (SAO) 

Applied to reconstructed video 

 

SAO Types 


 

Details of how SAO Types work 

o 3 pel patterns for pixel 

classification in Edge 

Offset 

o Pixel Classification Rules 

for Edge Offset 

o Grouping 4 bands and 

Representation 

HEVC Encoder 

Figure 3E Sample Adaptive Offset Types, Edge Classification, 

and Grouping of Bands 

Fig. 4 shows high level block diagram of HEVC Encoder. Input video frames are partitioned recursively from 

CTB’s to CUs and then nonrecursively into PUs. The prediction partition PUs are then combined to generate 

Prediction CUs that are differenced from the original resulting in residual CU’s that are recursively QT split 

into TUs and coded with variable Block Size (VBS) transform of 4x4 (DST or DCT approx), or 8x8, 16x16, and 

32x32 (DCT approx only). CU/PU Partitioner partitions into CU/PU, and the TU partitioner partitions into 

TUs. An Encode Controller controls the degree of partitioning performed which depends on quantizer used 

in transform coding. The CU/PU Assembler and TU Assembler perform the reverse function of partitioner. 

The decoded (every DPCM encoder incorporates a decoder loop) intra/motion compensated difference 

partitions are assembled following inverse DST/DCT to which prediction PUs are added and reconstructed 

signal then Deblock, and SAO Filtered that corespondingly reduce appearance of artifacts and restore 

edges impacted by coding. 

12 

Figure 4 HEVC Encoder

Intel® Media Server Studio Overview 

The Intel® Media Server Studio 2015 suite provides high performance software development tools, 

libraries, and infrastructure needed to help create, develop, debug, test, and deploy next generation 

enterprise grade media solutions on Intel® server processors and graphics. The suite supports multiple 

platforms with cross platform support for Linux and Windows under a single SDK. 

The Intel® Media Server Studio 2015 suite is offered appropriately segmented into bundles in order to 

offer maximum flexibility, cost advantage, and convenience to software developers, data center 

solutions providers, and OEMs to enable faster time to market. 

Communications 

Infrastructure 

Developers 

Encode 

Decode 

Microsoft Windows® 

Server 2012 

Video Software 

Experts and 

Developers 

Intel® HD 

Graphics 

Video 

Processing 

Linux 

Intel® Quick 

Sync Video 

Enterprise Video 

Device 

Manufacturers 

Tune, Test, 

Debug 

Intel® Advanced 

Vector Extension2 

Figure 5A Intel® Media Server Studio stack 

Cloud Based 

Media Services 

Providers 

Intel® Media 

Server Studio 

Cross 

Platform 

Optimized for 

Intel® Processors 

The Intel® Media Server Studio Essentials/Community Edition of the suite refers to a basic bundle of 

features, while the Intel® Media Server Studio Professional Edition refers to an advanced bundle of 

features. Optional high value capabilities are offered via Intel® VideoPro Analyzer, and Intel® Stress 

Bitstreams and Encoder toolkits. 


Side-by-side Features Comparison of the two editions of Intel® Media Server Studio 

Components 

Intel® Media Server Studio 

Professional Edition 

Intel® Media Server 

Studio Essentials Edition 

Intel®Media Server Studio 2015 - 

√ √ 

SDK 

Intel® Media Server Studio 2015 

√ √ 

– Graphics Driver 

13 



Intel® Media Server Studio 2015 - 

√ √ 

Samples 


√ 

– HEVC Decoder & Encoder 


√ 

– Audio Decoder & Encoder 

Intel® VTune Amplifier XE 

Video Quality Caliper Tool 

Premium Telecine and Interlace 

Reverser 

License 

Supported Features 

 

 

 

 

 

 

 

 

√ 

√ 

√ 

Professional HEVC Encoder 

(up to 40 sockets) 

Essential 

Hardware: Intel Xeon® Processor E3-1200 v3, E4-128x v4 product family and 4 th Generation Intel 

Core Processor-based Platforms with Intel Iris Pro and Intel HD Graphics. 

Operating Systems: CentOS*7.1, Ubuntu* 12.04 LTS, SUSE Linux Enterprise Server* (SLES), Windows 

Server 2012. 

Video Compression Codecs: H.264 (AVC), HEVC (H.265) (8-bit software only, and with Intel® Graphics 

Acceleration, and 10-bit software only), MPEG-2, VC-1, MVC, MJPEG (Windows only). 

Video Processing: deitnerlacing, Resizing, Cropping, Composition, Color conversion, Denoising, Framerate 

conversions. 

Audio Compression Codecs: Encode and Decode AAC LC and HE AAC V1, decode AAC LTP, AAC PS, HE- 

AAC v2, and MPEG-Audio (including ‘MP3’) 

Available Samples: Samples for decode, encode, multi-transcode, and video processing. 

Analysis/Improvement Tools: Intel VTune Amplifier XE 2013, Video Pro Analyzer, Visual Quality Caliper 

Tool, Telecine and Interlace Reversal Tool (preview version). 

Stress Bitstreams: HEVC Main-8, VP9 8-bit bitstreams. 

 

Documentation: Detailed documentation, user forum, and technical support to get app up and 

running. 

Related Optional Packages 

14 

Intel® VideoPro Analyzer 2015 offers a comprehensive kit of video analysis software tools for HEVC and VP9 

video coding. It enables visual analysis of bitstreams and the decoding process including statistics 

extraction, debugging and more. It also enables monitoring of visual quality of decoded video. 

Intel® Stress Bitstreams and Encoder 2015 are available for either of the HEVC or VP9 compression 

formats. It contains carefully designed stress bitstreams to remove pain from the validation process,

minimizing the needed development time while maximizing the confidence due to redundant syntax. It 

also provides the capability of being able to develop custom new bitstreams. 

Intel® Media Server Studio 2015 SDK 

The Intel® Media Server Studio SDK processing stack for Windows consists of optimized media libraries 

that are built on top of Microsoft* DirectX*, DirectX Video Acceleration (DVXA) APIs, and platform 

graphics drivers. Intel® Media Server Studio SDK exposes the hardware acceleration features of Intel® 

Quick Sync Video built into 2 nd and all following generations of Intel® Core processors. 

While Intel® Media Server Studio SDK is designed to be a flexible solution for many media workloads, it 

focuses only on media pipeline components which are commonly used and usually the most in need for 

acceleration, such as follows. 

 

 

 

Decoding from video elementary stream formats (H.264, MPEG-2, VC-1, and JPEG/Motion JPEG) to 

uncompressed frames. 

Selected video frame processing operations 

Encoding uncompressed frames to advanced elementary stream formats (MPEG-2, H.264, HEVC) 

Samples, Applications, External Libraries 

Intel® Media Server Studio – 

Optimized Media Library for CPU 

Intel® Media Server Studio API 

Intel® Media Server Studio – 

Optimized Media Library for 

Intel® Processor Graphics 

DXVA/DDI Extensions 

Graphics Drivers 


Figure 5B Intel® Media Server Studio SDK archtecture stack 

In the first quarter of 2014 the following features were added to the SDK. 

 

 

 

HEVC: High Quality HEVC Software Encoder and HEVC Software Decoder 

H.264: (1) Improved Encode quality and BRC, (2) Region of Interest Encoding, 

(3) Adaptive Picture types, Look ahead and others. 

VPP: (1) Frame composite API, (2) Deinterlacer choices: Bob or Advanced. 

15 


System: Splitter/Muxer support for MPEG-2 TS, and MPEG-4. 

Plug-ins/Tools. 


During portion of 2014 and all of 2015 HEVC Codec was improved significantly in quality and speed. 

- The HEVC codec implementation includes a multithreaded HEVC Encoder and HEVC Decoder 

capable of running typically up to 8 threads. The Encoder and Decoder are up to SSE4 and AVX2 

optimized for Intel Core, and Atom platforms. 

- The HEVC Software Decoder is capable of real time decode of HEVC 4K encoded streams on an 

Intel® Core Haswell CPU 2 - 3.5 GHz, 4 Cores. 

- The HEVC Software Encoder supports a range of coding modes with the slowest mode allowing 

quality visually identical to HEVC HM14 and the fastest mode allowing 30-40 frames/sec encoding 

of HD1080p on the aforementioned Haswell system. 

- The HEVC Software Decoder supports HEVC Main Profile 8 bits, Level 1.0 to 6.2. 

A complete list of various components and features of Intel® Media Server Studio SDK is summarized in 

Fig. 5C. 

Figure 5C List of Features of the Intel® Media Server Studio SDK 

Additional flavors of HEVC codecs such as high bit-depth, higher chroma resolution formats and others and 

capabilities for controlling and optimum usage of these codecs, have allowed Intel® Media Server Studio 

SDK to continue to improve its offerings to stay ahead of the pack. 

16

Media Server Studio HEVC R7 results - Part 1 

4:2:0 8-bit Coding Quality, Performance, and 

(on E5) Computational Scalability tests 


17 


Intel® Media Server Studio HEVC Software Encoder Quality 

Evaluation for 4:2:0 8-bit Content 

In this section, we first describe a test methodology employed for evaluating quality and then report on 

relative quality measurements for the Media Server Studio HEVC Codec with respect to MPEG HEVC HM14 

Codec, well known as a high quality reference albeit impractically slow. 


Quality Evaluation Tests for 4:2:0 8-bit Content 

For the purpose of quality evaluation, four test sets, one for each of the 4 main resolutions are selected 

with each test set consisting of 6 publicly available challenging video test sequences. Further for each test 

sequence, 4 well-spaced quantizer (QP) values are selected such that HEVC encoding for each sequence 

generates bit-rates in a wide but moderate range, avoiding extreme values. 

Wherever possible, test sequences that are available in multiple resolutions are selected for each test set 

such that the behavior of a codec over multiple resolutions can be tracked. 

Specifically, Table 0A shows selected Ultra Definition 4K (UHD4K) test set, Table 0B shows High Definition 

1080p (HD1080p) test set, Table 0C shows High Definition 720p (HD720p), and Table 0D shows 

Standard/Extended Standard Definition (SD/XD) test set. 

Table 0A UHD4K Test Set and Quantizers used for Codec RD characteristics measurement 

Quantizers used for RD char. 

UHD4K Test Sequence Resolution frm rate #frm Qp1 Qp2 Qp3 Qp4 

1 Park_joy 3840x2160 50 500 26 29 33 37 

2 Ducks_take_off 3840x2160 50 500 28 31 35 37 

3 CrowdRun 3840x2160 50 500 26 30 34 38 

4 PeopleOnStreet 3840x2160 30 150 22 27 32 37 

5 Traffic 3840x2048 30 300 18 22 26 30 

6 Nebuta_Festival 2560x1600 60 300 30 33 36 39 

Of the UHD4K test sequences referred to in Table 0A, Park_joy, Ducks_take_off and CrowdRun of 

3840x2160 resolution can be obtained from http://media.xiph.org/video/derf/ while PeopleOnStreet of 

3840x2160 resolution, Traffic of 3840x2048 resolution, and Nebuta_Festival of 2560x1600 resolution are 

MPEG HEVC test sequences that can be obtained from MPEG distribution sites such as 

ftp://hvc:US88Hula@ftp.tnt.uni-hannover.de/testsequences. 

18 

Table 0B HD1080p Test Set and Quantizers used for Codec RD characteristics measurement 


HD1080p Test Sequence Resolution frm rate #frm Qp1 Qp2 Qp3 Qp4 

1 Park_joy 1920x1080 50 500 26 29 33 37 


3 CrowdRun 1920x1080 50 500 26 30 34 38 

4 TouchDownPass 1920x1080 30 570 23 26 30 34

5 BQTerrace 1920x1080 60 600 25 27 31 34 

6 ParkScene 1920x1080 24 240 23 26 29 32 

Of the HD1080p test sequences referred to in Table 0B, Park_Joy, Ducks_take_off, CrowdRun and 

TouchDownPass of 1920x1080 resolution can be obtained from http://media.xiph.org/video/derf/ while 

BQTerrace and ParkScene of 1920x1080 resolution are standard MPEG HEVC test sequences that can be 

obtained from MPEG distribution site such as ftp://hvc:US88Hula@ftp.tnt.unihannover.de/testsequences. 

Table 0C HD720p Test Set and Quantizers used for Codec RD characteristics measurement 


HD720p Test Sequence Resolution frm rate #frm Qp1 Qp2 Qp3 Qp4 

1 Park_joy 1280x720 50 500 26 29 33 37 


3 CrowdRun 1280x720 50 500 26 30 34 38 

4 City 1280x720 30 300 22 24 27 29 

5 Crew 1280x720 30 300 22 26 29 32 

6 Sailormen 1280x720 30 300 24 26 28 30 

Table 0C shows HD720p test sequences of which Park_Joy, Ducks_take_off, and CrowdRun of 1280x720 

resolution can be obtained from http://media.xiph.org/video/derf/ while City, Crew, and Sailormen of 

1280x720 resolution are earlier MPEG SVC test sequences with limited public distribution (but can be 

made available on request). 

Table 0D SD/XD Test Set and Quantizers used for Codec RD characteristics measurement 


SD/XD Test Sequence Resolution frm rate #frm Qp1 Qp2 Qp3 Qp4 

1 BasketBallDrillText 832x480 50 500 21 24 27 30 

2 PartyScene 832x480 50 500 24 27 30 33 

3 RaceHorses 832x480 30 300 24 27 30 33 

4 City 704x576 30 300 22 24 26 28 

5 Crew 704x576 30 300 22 25 28 31 

6 Soccer 704x576 30 300 22 25 28 31 


Table 0D shows standard and extended definition test sequences such as BasketBallDrillText, PartyScene, 

and RaceHorses all of 832x480 resolution that are standard MPEG HEVC test sequences that can be 

obtained from ftp://hvc:US88Hula@ftp.tnt.uni-hannover.de/testsequences while City, Crew, and Soccer of 

704x576 resolution can be obtained from MPEG site ftp://ftp.tnt.unihannover.de/pub/svc/testsequences/. 

Now that we have discussed test sequences, we now introduce encoding configuration used. 

HEVC HM encoding is employed in high quality default high delay Random Access configuration but with 

only first frame Intra (other Intra’s can still happen but only due to scene change), pyramid configuration 

of size 8, and 4 Reference Pictures. For each encoding test, the reference quantizer used (which may be 

19 


internally modulated into multiple quantisers as needed for I, P or B (including reference and 

nonreference slices)) is specified in Table 0A – 0D. 


The Media Server Studio (MSS) HEVC software encoder supports several Target Usages (TU) that range 

from TU1 to TU7, and Intel® Graphics optimized version of TU4-TU7 (TU4gacc-TU7gacc) such that TU1 

offers the highest quality, and TU7gacc the highest speed. In our encoding tests, the main four TU modes 

– TU1 (MSS ‘Quality’ mode), TU4 (MSS ‘Balanced’ mode), TU7 (MSS ‘Speed’ mode), and TU7gacc are 

tested. Further for TU1, TU4, TU7, and TU7gacc modes, B-pyramid encoding with pyramid length set to 8 

is used. Likewise, different TU modes use different number of reference pictures such as 4, 3, 2, and 2 

reference pictures respectively for TU1, TU4, TU7, and TU7gacc modes. 

To measure quality of an HEVC codec while many techniques such as peak signal-to-noise ratio (PSNR), 

structural similarity index (SSIM), other objective quality metrics, or even full subjective quality tests can 

be performed, in our own competitive quality assessment tests we have found results of most such 

measures to be fairly consistent with each other as long as codecs are compared in a constant quantizer 

mode. This allows comparison of actual codecs independent of bit rate control (BRC) techniques. 

We now briefly describe a statistically tractable technique to compare video quality produced by the 

codec being tested as compared to a reference codec. Rate Distortion (RD) characteristics for both the 

codecs are computed using each codec’s 4-point PSNR/Bitrate measurements followed by use of MPEG’s 

new BD rate ([5]) curve fitting procedure that generates a continuous RD curve that tightly fits the 

measured points. A single measurement of ‘goodness’ of the codec being tested against the reference 

codec in the form of BD rate is then computed that reflects percentage difference between the codecs. 

The BD rate percentage difference if positive means that the codec being tested is worse in quality, that is 

it costs ‘x’ percentage more bits to generate the same PSNR quality as the reference. The BD rate 

difference measurement procedure allows a straightforward way of computing and independently 

verifying quality of codec with respect to a reference codec. 

Quality Evaluation Results for 4:2:0 8-bit Content 

Before discussing detailed BD rate differences of each Media Server Studio TU mode wrt HM 14, we first 

establish the quality/bitrate tradeoff of HM 14 that will be used as reference. 

First, each video sequence of each test set (such as UHD, HD1080p, HD720p, and SD/XD) is encoded using 

MPEG HEVC HM14 Reference Encoder with each of 4 quantizers as specified in Tables 0A-0D; sample 

encoding scripts used are shown in Appendix C. The overall PSNR (averaged over frames) for each test 

sequence for each Qp for each of luma (Y), and associated chroma components (U and V) is collected 

along with the generated coding bitrate. For instance, Tables 1A-1D show the luma PSNR (chroma PSNR is 

also collected but is not shown to keep tables manageable in size) and total bitrate for each test 

sequence for each of 4 Qps is collected and is used in curve fitting to generate a continuous RD curve for 

HM14 encoding of that test sequence. 

20 

Next, Media Server Studio HEVC Software Encoder in each of its 4 main modes: TU1, TU4, TU7, and 

TU7gacc is utilized to perform encoding of each test sequence with each of 4 quantizers, and PSNRs of Y, 

U, and V, and total bitrates are collected; example encoding scripts are shown in Appendix C.

UHD4K Test 

Sequence 

Table 1A PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on UHD4K Test set 

Qp1 PSNR, Bitrate, Qp2 PSNR, Bitrate, Qp3 PSNR, Bitrate, Qp4 PSNR, Bitrate, 

dB Y kbps dB Y kbps dB Y kbps dB Y kbps 

1 Park_joy 35.46 78618.8 34.05 46580.4 32.11 23364.3 30.28 12681.8 

2 Ducks_take_off 31.43 90261.7 30.50 41146.9 29.49 19223.3 28.98 14004.0 

3 CrowdRun 35.85 54630.9 34.41 27522.1 32.74 15383.6 31.02 9137.1 

4 PeopleOnStreet 39.96 63581.6 37.00 28295.8 34.14 14476.9 31.43 8065.3 

5 Traffic 43.63 51403.5 41.59 22377.6 39.48 9774.2 37.47 4907.8 

6 Nebuta_Festival 30.78 48670.2 28.98 19513.1 28.12 8004.1 27.56 3671.2 

HD1080p Test 

Sequence 

Table 1B PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on HD1080p Test set 


dB Y kbps dB Y kbps dB, Y kbps dB, Y kbps 

1 Park_joy 34.33 35140.4 32.20 21705.7 29.64 10913.1 27.42 5552.4 

2 Ducks_take_off 32.23 28302.7 30.69 16346.3 28.81 7991.4 27.93 5587.9 

3 CrowdRun 35.11 25316.6 32.62 13868.8 30.24 7619.6 28.14 4281.4 

4 TouchdownPass 40.42 5743.1 39.35 2908.6 37.81 1444.7 36.14 790.7 

5 BQTerrace 35.91 12318.7 35.32 6264.8 34.19 2101.4 33.09 1130.2 

6 ParkScene 39.49 5300.4 37.89 3087.3 36.32 1886.6 34.70 1132.8 

HD720p Test 

Sequence 

Table 1C PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on HD720p Test set 



1 Park_joy 33.91 18166.8 31.65 11456.1 28.88 5779.1 26.52 2888.0 

2 Ducks_take_off 32.29 14525.9 30.42 8539.4 28.21 4117.4 27.24 2871.9 

3 CrowdRun 34.49 15194.3 31.62 8438.4 39.04 4599.1 26.83 2518.4 

4 City 38.89 5439.4 37.83 3171.9 36.49 1603.0 35.61 1105.0 

5 Crew 40.76 4477.8 38.98 1975.8 37.72 1206.9 36.37 751.1 

6 Sailormen 37.43 4656.3 36.43 2931.4 35.53 1909.2 34.68 1282.5 

XD/SD Test 

Sequence 

Table 1D PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on XD/SD Test set 



1 BasketBallDrillText 41.07 3753.2 39.13 2405.3 37.19 1562.7 35.35 1023.2 

2 PartyScene 36.74 4478.5 34.62 2738.5 32.65 1702.7 30.75 1038.6 

3 RaceHorses 36.67 3206.6 35.82 1909.6 34.08 1188.9 32.37 747.2 

4 City 38.47 2745.7 37.37 1624.2 36.42 1021.6 35.54 692.4 

5 Crew 39.71 3064.0 38.17 1672.7 36.78 1002.7 35.35 621.5 

6 Soccer 40.06 2754.8 38.30 1600.4 36.61 1001.9 34.87 624.6 


21 



For Media Server Studio HEVC codec in each encoding mode, the 4 collected bitrates and corresponding 4 

PSNRs (each corresponding to one Qp) are used in curve fitting to generate a continuous RD curve for 

each encoding mode. Further, for each test sequence and for each Media Server Studio TU mode, the 

corresponding Media Server Studio RD curve is compared to HM14 RD curve and a BD rate percentage is 

computed that reflects the difference in quality between specific Media Server Studio TU mode with 

respect to TU14 reference. The BD rate provides a measure of departure of Media Server Studio TU 

mode’s quality in percentage bits with respect to the HM14 reference. For instance a BD rate percentage 

of say 4% for Media Server Studio TU1 mode means that Media Server Studio HEVC TU1 mode in order to 

provide the same objective quality as HM14 would require 4% additional bits. The HEVC provided macro 

for BD rate calculation is shown in Appendix C. 

Table 2A-2D show the measured BD rate percentage bitrate for each of luma and chroma components for 

each test sequence of each test set of Table 0A-0D for each of 4 TU modes being evaluated. Fig. 6A-6D 

compares the RD characteristics of two codecs for each test set for TU1 and TU4 modes where the 

difference in quality is the highest and the lowest wrt HM14 reference. 

For instance, Table 2A shows on UHD4K test set, the BD rate percentage difference of Media Server 

Studio HEVC in various TU modes wrt HM14 reference. As can be observed on UHD4K test set, the 

average luma BD rate percentage difference of Media Server Studio HEVC Codec over HM14 (an ideal 

reference), is -0.05%, 17.65%, 33.57%, and 35.22% higher respectively in TU1, TU4, TU7, and TU7gacc modes. 

This means that for UHD4K test set, the Media Server Studio codec as compared to HM14, requires 

around 0% higher bitrate in TU1 mode, 18% higher bitrate in TU4 mode, 34% higher bitrate in TU7 mode, 

and 35% higher bitrate in TU7gacc mode to achieve the same luma PSNR quality as HM14. Further as a 

matter of reference, the HM14 codec is 100 to 3000 times slower (shown in the next section) as compared 

to Media Server Studio HEVC codec depending on the TU setting used. 

Table 2A Relative Quality Evaluation Results on UHD4K test set for Intel® Media Server Studio HEVC Software Codec 

in TU1, TU4, TU7, and TU7gacc modes with respect to MPEG HEVC HM14 Codec 

UHD4K Test 

Sequence 

TU1 BD rate, %age TU4 BD rate, %age TU7 BD rate, %age TU7gacc BD rate, %age 

Y U V Y U V Y U V Y U V 

1 Park_joy -0.18 1.92 -0.63 13.50 7.92 16.47 26.97 25.10 33.01 28.15 25.76 33.61 

2 Ducks_take_off -0.93 6.81 9.79 15.92 -16.84 6.27 28.53 -0.84 23.39 30.00 0.62 25.80 

3 CrowdRun 0.37 2.99 2.14 17.33 18.18 20.35 35.46 46.38 48.19 36.74 47.33 49.31 

4 PeopleOnStreet 0.16 -7.15 -6.21 20.65 21.06 25.66 40.38 52.82 59.84 43.26 55.77 62.65 

5 Traffic 0.31 -9.00 -15.63 20.86 16.15 23.76 36.52 41.70 42.65 37.96 42.62 42.98 

6 Nebuta_Festival 6.91 47.06 42.79 20.02 16.68 5.40 46.90 33.35 16.72 47.68 32.44 16.53 

Average -0.05 -0.89 -2.11 17.65 9.30 18.50 33.57 33.03 41.41 35.22 34.42 42.87 

22 

Further, Table 2A shows on UHD4K test set that Media Server Studio HEVC Codec TU1 mode with respect 

to HM14 reference, results in biggest positive luma BD rate percentage difference (its worst case) on 

Nebuta_Festival 1600p sequence and better than HM14 luma BD rate percentage difference (its best case)

on Ducks_take_off 4K sequence; the corresponding RD curves for luma (Y) for TU1 mode, and HM 

verifying this behaviour are shown in Fig. 6A1. 

32 

Nebuta_Festival Y 

32 

Ducks_take_off Y 

31 

PSNR (dB) 

30 

29 

28 

27 

HEVC HM14.0 

MediaServerStudio TU1 

0 20000 40000 60000 

BitRate (kbps) 

Figure 6A1 RD results of UHD4K scenes with the biggest and the smallest quality difference wrt HM14 for 

Media Server Studio TU1 mode 

Table 2A also shows that Media Server Studio HEVC Codec in TU4 mode with respect to HM14 reference, 

results in the biggest luma BD rate percentage difference on Traffic 4K sequence and the smallest luma 

BD rate percentage difference on Park_Joy 4K sequence; the corresponding RD comparison curves for 

TU4 mode for the two cases are shown in Fig. 6A2. 

PSNR (dB) 

44 

43 

42 

41 

40 

39 

38 

37 

36 

Traffic Y 



0 20000 40000 60000 


PSNR (dB) 

PSNR (dB) 

36 

35 

34 

33 

32 

31 

30 

29 

31 

30 

29 



0 50000 100000 150000 


Park_Joy Y 



0 50000 100000 



Figure 6A2 RD results of UHD4K scenes with the biggest and the smallest quality difference wrt HM14 for 


23 

Next, Table 2B shows BD rate percentage differences of Media Server Studio HEVC wrt HM14 on the 

HD1080p test set. The average luma BD rate percentage difference of Media Server Studio HEVC Codec 


over HM14 reference, is 1.71%, 17.30%, 33.90%, and 37.71% respectively in TU1, TU4, TU7, and TU7gacc 

modes. 


Table 2B Relative Quality Evaluation Results on HD1080p test set for Intel® Media Server Studio HEVC Software Codec 


HD1080p Test 

Sequence 



1 Park_joy 1.13 5.42 1.63 13.40 16.97 27.40 26.40 32.87 42.50 28.41 34.47 44.18 

2 Ducks_take_off 0.55 2.78 7.08 10.49 -3.20 21.87 20.40 11.05 38.23 22.05 12.65 40.62 

3 CrowdRun 1.03 3.31 2.83 16.52 22.22 23.13 35.12 48.29 50.21 36.86 50.23 51.97 

4 TouchDownPass 0.83 -5.35 -4.83 19.31 17.52 17.78 38.34 53.74 50.80 46.23 57.41 54.14 

5 BQTerrace 4.01 -14.71 -22.15 26.65 24.46 32.33 49.85 41.01 61.07 55.94 38.78 61.87 

6 ParkScene 2.69 -6.10 -7.99 17.45 21.20 19.58 33.31 40.70 36.60 36.78 43.46 38.45 

Average 1.71 -2.44 -3.91 17.30 16.53 23.68 33.90 37.94 46.57 37.71 39.50 48.54 

Further, Table 2B shows on HD1080p test set that Media Server Studio HEVC Codec in TU1 mode with 

respect to HM14 reference, results in the biggest luma BD rate percentage difference on BQTerrace 1080p 

sequence and the smallest luma BD rate percentage difference on Ducks_take_off 1080p sequence; the 

corresponding RD comparison curves for TU1 mode for the two cases are shown in Fig. 6B1. 

PSNR (dB) 

37 

36 

35 

34 

33 

BQTerrace Y 



0 5000 10000 15000 20000 


PSNR (dB) 

33 

32 

31 

30 

29 

28 

27 




0 10000 20000 30000 40000 


Figure 6B1 RD results of 1080p scenes with the biggest and the smallest quality difference wrt HM14 for 


24 

Table 2B also shows that Media Server Studio HEVC Codec in TU4 mode with respect to HM14 reference, 

results in the biggest luma BD rate percentage difference on BQTerrace 1080p sequence and the smallest 

BD rate percentage difference for luma on Ducks_take_off 1080p sequence; the corresponding 

comparison curves for TU4 for the two cases are shown in Fig. 6B2.

PSNR (dB) 

37 

36 

35 

34 

33 

32 

BQTerrace Y 



0 5000 10000 15000 


Figure 6B2 RD results of 1080p scenes with the biggest and the smallest quality difference wrt HM14 for 


Next, Table 2C shows BD rate percentage differences of Media Server Studio HEVC wrt HM14 on the 


over HM14 reference is 1.56%, 16.66%, 34.94%, and 39.46% respectively for TU1, TU4, TU7, and TU7gacc 

modes. 

Table 2C Relative Quality Evaluation Results on HD720p test set for Intel® Media Server Studio HEVC Software Codec 


HD720p Test TU1 BD rate, %age TU4 BD rate, %age TU7 BD rate, %age TU7gacc BD rate, %age 

Sequence 


1 Park_joy 1.50 8.52 1.79 13.76 21.34 31.32 25.98 36.64 48.01 28.54 39.09 50.21 

2 Ducks_take_off 0.74 3.17 8.35 9.74 5.09 31.10 19.86 20.60 48.11 21.48 22.18 49.8 

3 CrowdRun 1.76 3.44 3.55 17.36 25.54 26.88 34.09 51.23 54.06 35.86 53.03 056.0 

4 City 0.47 -15.84 -15.61 20.49 14.98 18.35 47.09 54.84 63.71 54.20 57.71 568.8 

5 Crew 2.80 -2.16 -1.90 22.24 12.83 13.89 46.97 57.90 52.16 58.21 64.42 461.4 

6 Sailormen 2.11 -6.27 -11.14 16.38 21.36 27.72 35.62 43.83 43.90 38.49 44.33 045.4 

Average 1.56 -1.53 -2.50 16.66 16.86 24.88 34.94 44.17 51.66 39.46 46.79 55.29 4 

PSNR (dB) 

33 

32 

31 

30 

29 

28 

27 




0 10000 20000 30000 40000 



Further, Table 2C shows that Media Server Studio HEVC Codec in TU1 mode results in the biggest BD rate 

percentage difference for luma on Crew 720p sequence and the smallest BD rate percentage difference in 

luma on Ducks_take_off 720p sequence with respect to HM14 reference; the corresponding RD 

comparison curves for TU1 mode for the two cases are shown in Fig. 6C1. 

25 


42 

Crew Y 

33 


41 

32 


PSNR (dB) 

40 

39 

38 

37 

36 



0 2000 4000 6000 


Figure 6C1 RD results of 720p scenes with the biggest and the smallest quality difference wrt HM14 for 


Table 2C also shows that Media Server Studio HEVC Codec in TU4 mode results in the biggest BD rate 

percentage difference for luma on Crew 720p sequence and the smallest BD rate percentage difference 

for luma on Ducks_take_off 720p sequence with respect to HM14 reference; the corresponding RD 

comparison curves for TU4 mode for the two cases are shown in Fig. 6C2. 

PSNR (dB) 

41 

40 

39 

38 

37 

36 

35 

Crew Y 



0 2000 4000 6000 


PSNR (dB) 

PSNR (dB) 

33 

32 

31 

30 

29 

28 

27 

31 

30 

29 

28 

27 



0 5000 10000 15000 20000 





0 5000 10000 15000 20000 


Figure 6C2 RD results of 720p scenes with the biggest and the smallest quality difference wrt HM14 for 


26 

Next, Table 2D shows BD rate percentage differences of Media Server Studio HEVC wrt HM14 on the 

SD/XD test set. The average luma BD rate percentage difference of Media Server Studio HEVC Codec over 

HM14 reference is 0.07%, 19.44%, 39.33%, and 44.65% respectively for TU1, TU4, TU7, and TU7gacc modes. 

Table 2D Relative Quality Evaluation Results on SD/XD test set for Intel® Media Server Studio HEVC Software Codec in 

TU1, TU4, TU7, and TU7gacc modes with respect to MPEG HEVC HM14 Codec

SD/XD Test 

Sequence 



1 BasketBallDrilText -5.28 -9.95 -8.02 23.65 30.36 33.23 42.58 61.73 65.86 46.58 64.30 68.15 

2 PartyScene 4.15 -1.27 -2.32 20.65 22.38 23.82 35.33 40.91 42.96 38.62 43.68 46.47 

3 RaceHorses 1.12 -1.18 -2.40 21.41 21.25 27.44 45.05 54.40 65.82 51.79 60.87 72.86 

4 City 0.49 -15.36 -16.88 18.37 11.15 15.15 41.57 38.53 48.10 46.92 41.37 50.18 

5 Crew 1.83 -2.89 -3.22 17.58 10.88 13.20 38.77 46.39 48.64 46.0 53.00 56.13 

6 Soccer -1.88 -6.68 -5.98 15.00 5.65 11.86 32.67 29.39 40.96 38.01 32.88 43.63 

Average 0.07 -6.22 -6.47 19.44 16.95 20.78 39.33 45.22 52.06 44.65 49.35 56.24 

Also, Table 2D shows that Media Server Studio HEVC Codec in TU1 mode results in the biggest BD rate 

percentage difference for luma on PartyScene XD sequence and the smallest BD rate percentage 

difference in luma on BasketballDrillText SD sequence with respect to HM14 reference; the corresponding 

RD comparison curves for TU1 mode for the two cases are shown in Fig. 6D1. 

PSNR (dB) 

38 

37 

36 

35 

34 

33 

32 

31 

30 

PartyScene Y 



0 2000 4000 6000 


PSNR (dB) 

Figure 6D1 RD results of SD/XD scenes with the biggest and the smallest quality difference wrt HM14 for 


Further, Table 2D shows that Media Server Studio HEVC Codec in TU4 mode results in the biggest BD rate 

percentage difference for luma on BasketballDrillText XD sequence and the smallest BD rate percentage 

difference for luma on Soccer SD sequence with respect to HM14 reference; the corresponding RD 

comparison curves for TU4 mode for the two cases are shown in Fig. 6D2. 

43 

42 

41 

40 

39 

38 

37 

36 

35 

BasketballDrillText Y 



0 2000 4000 6000 



27 



PSNR (dB) 

Figure 6D2 RD results of SD/XD scenes with the biggest and the smallest quality difference wrt HM14 for 


To summarize, the luma BD rate percentage bitrate difference of the Media Server Studio HEVC in various 

modes with respect to an ideal (but slow) HM14 reference is as follows. 

 

 

 

 

42 

41 

40 

39 

38 

37 

36 

35 

34 

BasketballDrillText Y 



0 2000 4000 6000 


PSNR (dB) 

TU1 mode on average is only 0 - 2% worse in BD rate (while around 125 times faster, as shown later) as 

compared to HM14 on all test sets. This is highest quality mode in Media Server Studio. 

TU4 mode on average is only 16 - 19% worse in BD rate (while around 950 times faster, as shown later) 

as compared to HM14. This mode represents excellent tradeoff of quality vs speed. 

TU7 mode on average is 33 - 39% worse in BD rate (while around 2450 times faster, as shown later) as 

compared to HM14. This is the fastest software only mode in Media Server Studio. 

TU7gacc mode on average is also 35 - 44% worse in BD rate rate (while being around 3150 times faster, 

as shown later) as compared to HM14 on all test sets. This is the fastest mode in Media Server Studio 

and combines TU7 software mode with Intel® Graphics Acceleration. 

Now that we have completed quality analysis of various TU modes of Media Server Studio, the next 

obvious step is to perform analysis of encoding speed offered by each of these modes; this issue is 

discussed at length in the next section. 

41 

40 

39 

38 

37 

36 

35 

34 

Soccer Y 



0 1000 2000 3000 4000 


Intel® Media Server Studio HEVC Encoder Quality vs Performance 

Tradeoffs for 4:2:0 8-bit Content 

28 

For measurement of encoding speed (fps) and speed vs quality tradeoffs, a recently released reference 

PC Platform (Intel® Core i7-4770K CPU @ 3.5 GHz – 4 Cores/8Threads) is employed.

HEVC Software Encoder Performance for 4:2:0 8-bit Content 

We first measure encoding speed (fps) of Media Server Studio HEVC Software Encoder in TU1 (highest 

quality) mode and MPEG HEVC HM14 on different resolution test sets. Results of these measurements 

comparing the two speeds are shown in Fig. 7A. 

Encoding Speed (fps) 

Figure 7A Average encoding speed comparison of Media Server Studio TU1 mode with HM14 

From Figure 7A it can be seen that the encoding speed of the TU1 mode is 100 to 141 times (multithreaded 

on 4 cores) the speed of HM14 (single threaded/1 core) for the 4 resolution test sets. For 1080p content, 

TU1 is 141 times faster than HM14. Earlier we had shown that the quality of TU1 mode was nearly identical 

(only around 0-2% less in BD rate) with respect to HM14. 

Next in Table 3A-3D we show measurement of encoding speed of Media Server Studio HEVC Software 

Encoder on each sequence of all four resolution test sets in each of four TU1, TU4, TU7, and TU7gacc 

modes. 

Table 3A Encoding Speed of UHD4K test set on Media Server Studio HEVC Software Codec in TU1, TU4, TU7, and 

TU7gacc modes. For reference the average Speed of HM14 Encoder for this test set is .004 fps. 

UHD4K Test 

Sequence 

7.0 

6.0 

5.0 

4.0 

3.0 

2.0 

1.0 

0.0 

Performance Comparison of MediaServerStudio TU1 Mode and HM14.0 

5.78 (141x) 

3.04 (117x) 

2.09 (139x) 

0.40 (100x) 

0.041 0.026 0.015 0.004 

SD/XD HD720p HD1080p UHD4K 

Resolution 



TU1 Enc Speed TU4 Enc Speed TU7 Enc Speed TU7gacc Enc Speed 

fps fps fps fps 

1 Park_joy 0.36 2.70 7.86 10.97 

2 Ducks_take_off 0.27 2.11 6.58 9.43 

3 CrowdRun 0.38 3.40 9.28 12.45 

4 PeopleOnStreet 0.33 3.03 7.59 9.73 

5 Traffic 0.65 4.42 11.02 14.49 

6 Nebuta_Festival* 1.05 7.66 19.29 24.30 

Average 0.40 3.13 8.47 11.41 


29 


* For average encoder speed calculation, for all TU’s Nebuta_Festival is excluded as its size is 2560x1600 while others are of 

size 4K (3840x2160). 


Table 3B Encoding Speed of HD1080p test set on Media Server Studio HEVC Software Codec in TU1, TU4, TU7, and TU7gacc 

modes. For reference the average speed of HM14 Encoder for this test set is .015 fps. 

HD1080p Test 

Sequence 



1 Park_joy 1.50 10.02 25.06 34.59 

2 Ducks_take_off 1.18 8.65 24.40 34.80 

3 CrowdRun 1.33 10.53 27.34 38.20 

4 TouchDownPass 2.09 17.92 46.05 60.02 

5 BQTerrace 4.08 21.01 56.20 67.22 

6 ParkScene 2.36 16.59 42.51 54.18 

Average 2.09 14.12 36.93 48.17 

Table 3C Encoding Speed of HD720p test set on Media Server Studio HEVC Software Codec in TU1, TU4, TU7, and TU7gacc 


HD720p Test 

Sequence 

TU1 Enc Speed TU4 Enc Speed TU7 Enc Speed TU7gaccEnc Speed 


1 Park_joy 3.22 20.68 49.64 68.33 

2 Ducks_take_off 2.75 19.17 49.55 67.45 

3 CrowdRun 2.67 19.75 48.35 68.46 

4 City 3.92 30.87 69.16 89.06 

5 Crew 2.40 25.46 65.17 89.73 

6 Sailormen 3.30 24.75 67.05 86.73 

Average 3.04 23.45 58.15 78.29 

Table 3D Encoding Speed of SD/XD test set on Media Server Studio HEVC software Codec in TU1, TU4, TU7, and TU7gacc 


SD/XD Test 

Sequence 

TU1 Enc Speed TU4 Enc Speed TU7 Enc Speed TU7gaac Enc Speed 


1 BasketBallDrillText 6.42 50.41 132.21 153.24 

2 Party_Scene 7.52 54.83 133.19 148.70 

3 RaceHorses 4.72 40.10 103.49 126.06 

4 City 6.90 54.04 127.11 150.20 

5 Crew 3.82 38.80 103.84 133.84 

6 Soccer 5.29 45.45 120.86 149.45 

Average 5.78 47.27 120.12 143.58 

30 

As can be seen from Table 3A-3D the encoding speed of TU4 mode is in the range of 6.8 to 8.2 that of TU1 

mode, and that the encoding speed of TU7 mode is 2.5 to 2.7 times that of TU4 mode, and TU7gacc mode 

is around 32% faster, ie, 1.2 to 1.34 times the speed of TU7. In other words TU7gacc mode reflects an

overall speed of 24 to 29 times that of TU1 mode. Also, since TU1 for 4 core muli-threaded coding is 

typically on average 125 times the speed of HM (range of 100 to 141 times), TU7gacc reflects a typical 

speedup factor of 2850 to 3500 over HM14! 

A side-by-side comparison of encoding speed of different modes and for the 4 different resolution test 

sets is shown by the bar graphs of Fig. 7B. Table 3A-3D also show that the TU7gacc mode is able to reach 

encoding speed of 9.5 to 14.5 fps for UHD4K content, 35 to 67 fps for HD1080p content, around 67 to 90 

fps for HD720p, and 126 to 153 fps for SD/XD content. 


140 

120 

100 

80 

60 

40 

20 

0 

Encoding Performance 


Resolution 




MediaServerStudio TU7gacc 

Figure 7B Average encoding speed of all test sets on Media Server Studio HEVC Software Encoder 

Fig. 8 shows another benefit of TU7gacc as compared to TU7 in terms of its capability of reducing the load 

on the CPU. Thus TU7gaac is currently able to reduce the load on CPU by around 5% from around 95% to 

around 90% enabling CPU to be used by other applications. This is in addition to 32% encoding speedup 

that TU7gacc provides over TU7 while achieving identical video quality. 


31 


CPU load Comparison between MediaServerStudio TU7 and TU7gacc 

110 

100 




CPU Load (%) 

90 

80 

70 

60 

50 


Resolution 

Figure 8 CPU Load Difference between TU7 and TU7gacc modes 

Quality vs Performance Tradeoff for TU Modes for 4:2:0 8-bit Content 

We now discuss results of Codec Quality vs Encoding Performance tradeoffs for all modes for Media 

Server Studio HEVC Encoder for all 4 resolutions tested. 

Fig. 9 shows a comparison of Quality (negative BD rate percentage difference wrt HEVC HM14) vs 

Encoding Performance (fps) for each of the four resolution test sets at each of the four TU1, TU4, TU7, 

and TU7gacc encoding modes. The y-axis basically shows the quality difference in terms of loss of BD rate 

percentage difference in the process of increasing speed up of the encoder in going from TU1 to TU4 to 

TU7 to TU7gacc operating points. 

To summarize, encoding performance-wise Media Server Studio HEVC codec in different modes achieves 

the following speedup of HEVC encoding on 4 core Reference PC platform specified earlier. 

32

Quality vs Performance Y 

5 

0 20 40 60 80 100 120 140 160 

Figure 9 Quality vs Encoding Speed Tradeoffs of Media Server Studio HEVC Encoding modes 

 

 

BDRate %age dif wrt HM14.0 

0 TU1 TU1 TU1 

-5 

-10 

-15 

-20 

-25 

-30 

-35 

-40 

-45 

-50 

TU4 

TU7 

TU4 

TU7gacc 

TU4 

TU4 

TU7 TU7gacc 

TU7 

TU7gacc 


For UHD4K test set, Media Server Studio encoder in TU1, TU4, TU7 and TU7gacc, correspondingly on 

average provides 0.40 fps, 3.1 fps, 8.5 fps and 11.4 fps. This reflects corresponding speedup factors 

wrt HM14 (single threaded) of 100, 783, 2118, and 2850. 

For HD1080p test set, Media Server Studio encoder in TU1, TU4, TU7 and TU7gacc, correspondingly on 



TU7 

SD/XD 

HD720p 

HD1080p 

UHD4K 

TU7gacc 


 

 

For HD720p test set, Media Server Studio encoder in TU1, TU4, TU7 and TU7gacc, correspondingly on 



For SD/XD test set, Media Server Studio encoder in TU1, TU4, TU7 and TU7gacc, correspondingly on 



33 


Intel® Media Server Studio HEVC Software Decoder Performance for 

4:2:0 8-bit Content 

In this section we describe results of decoding speed measurement of Intel® Media Server Studio HEVC 

Decoder. For measurement of decoding speed (fps), the same reference PC Platform (Intel® Core i7- 

4770K CPU @ 3.5 GHz – 4 Cores/8 Threads) used for encoding speed measurement is employed. 


34 

The Media Server Studio HEVC Software Decoder is able to achieve very high threading throughput 

consuming over 90% of resources on the noted machine. 

For measurement of decoder performance, longer bitstreams of typically around 1000 or more frames are 

necessary to obtain a stable measurement. Thus, each of the video sequences of Table 0A-0D since they 

are relatively short were extended by palindromic repetition (so as not to introduce sudden scene 

changes that might introduce an unnatural behavior in the measurement) to 900 – 1200 frames long and 

compressed with HEVC using the same Qp quantizers as in Table 0A-0D. These longer compressed 

streams were then used for decoder performance measurment. 

Tables 4A-4D show average speed of decoding bitstreams of each sequence of each of the 4 test sets as 

well as an overall average deoding speed for each category. 

Table 4A Decoding Speed of UHD4K on Intel® Media Server Studio HEVC Codec in TU1, TU4, TU7 modes 

UHD4K Test Sequence 

TU1 Dec Speed TU4 Dec Speed TU7 Dec Speed 

fps fps fps 

1 Park_joy 70.98 62.31 67.98 

2 Ducks_take_off 72.83 66.19 61.57 

3 CrowdRun 73.07 67.90 74.39 

4 PeopleOnStreet 65.06 59.67 65.50 

5 Traffic 54.70 46.48 49.75 

6 Nebuta_Festival* 172.78 144.69 159.11 

Average 67.33 60.51 63.84 

* For average calculation, Nebuta_Festival is excluded as its size is 2560x1600 while others are 3840x2160. 

Table 4B Decoding Speed of HD1080p on Intel® Media Server Studio HEVC Codec in TU1, TU4, TU7 modes 

HD1080p Test Sequence 


fps fps fps 

1 Park_joy 183.16 205.53 232.34 

2 Ducks_take_off 198.84 242.18 233.12 

3 CrowdRun 135.32 212.90 244.08 

4 TouchDownPass 221.87 312.77 346.70 

5 BQTerrace 255.18 329.90 368.03 

6 ParkScene 203.97 291.99 326.28 

Average 199.72 265.88 291.76

Table 4C Decoding Speed of HD720p on Intel® Media Server Studio HEVC Codec in TU1, TU4, TU7 modes 

HD720p Test Sequence 


fps fps fps 

1 Park_joy 273.47 412.55 477.47 

2 Ducks_take_off 300.95 461.09 476.58 

3 CrowdRun 220.06 383.81 463.36 

4 City 300.66 512.35 564.14 

5 Crew 267.24 494.03 538.53 

6 Sailormen 283.76 504.51 561.94 

Average 274.36 461.39 513.67 

Table 4D Decoding Speed of SD/XD on Intel® Media Server Studio HEVC Codec in TU1, TU4, and TU7 modes 

SD/XD Test Sequence 


fps fps fps 

1 BasketBallDrillText 504.41 942.65 1067.21 

2 PartyScene 476.69 866.48 1031.10 

3 RaceHorses 408.82 687.30 908.16 

4 City 509.83 1011.16 1134.78 

5 Crew 402.83 804.19 905.82 

6 Soccer 475.78 964.00 1099.34 

Average 463.06 879.30 1024.40 

Decoding Speed (fps) 

1200 

1000 

800 

600 

400 

200 

Decoding Performance 

TU1 

TU4 

TU7 


0 


Resolution 

Figure 10 Media Server Studio HEVC Software Decoding speed for all resolutions 

As can be see from Table 4A-4D, as expected, the decoding speed is inversely proportional to the spatial 

resolution of video sequence being decoded. Average decoding speed for 1080p resolution content is 

35 


oughly 4 times that of 4K resolution content (Table 4B vs Table 4A), average decoding speed of 720p 

resolution content is almost a factor of 1.6 of 1080p content (Table 4C vs Table 4B), and that of SD/XD 

resolution content is close to a factor of 2 as compared to decoding speed of HD720p content (Table 4D 

vs Table 4C). 


36 

Further, Fig. 10 shows a combined graph of variation of Media Server Studio HEVC Software Decoder 

performance (fps) across various resolutions test sets. On the Haswell platform, including the rendering 

overhead, one 60 fps UHD4K stream or two 30 fps UHD4K streams, or up to eight HD1080p streams can 

be easily decoded in realtime. 

Intel® Media Server Studio HEVC Software Encoding Computational 

Scalability Tests (on E5 Server) 

For measurement of computational scalability beyond 4 core i7-4770K system used in tests thus far, an E5 

Server (Intel® Xeon® Processor E5-2697v2, 12 Core, 2.7 GHz, 2 processors per board, total of 24 physical 

cores) is employed. Currently such servers do not inherently support Intel® graphics acceleration. The 

purpose of this test is to demonstrate threading scalability of the Intel Media Server Studio HEVC 

software encoder when many more cores are available. This also addresses an application where very 

high resolution video (4K and above) is needed to be HEVC encoded in realtime in software on many 

available cores. 

Quality Evaluation Results for Computational Scalability Tests for 4:2:0 8-bit 

Using the test set of 4:2:0 8-bit content as discussed earlier via Tables 0A-0B and test procedure establish 

earlier, we first evaluate video quality of the Intel® Media Server Studio HEVC for the three main TU modes, ie, 

TU1, TU4, and TU7, for the encoding configuration used to provide high degree of scalability. 

The results of quality tests are shown in Tables 5A and 5B for UHD4K and HD1080p test sets respectively. 

Table 5A Relative Quality Evaluation Results on UHD4K test set for Intel® Media Server Studio HEVC Software Codec 

in TU1, TU4, and TU7 modes with respect to MPEG HEVC HM14 Codec 

UHD4K Test 

Sequence 

TU1 BD rate, %age TU4 BD rate, %age TU7 BD rate, %age 

Y U V Y U V Y U V 

1 Park_joy -0.18 1.92 -0.63 13.50 7.92 16.47 26.97 25.10 33.01 

2 Ducks_take_off -0.93 6.81 9.79 15.92 -16.84 6.27 28.53 -0.84 23.39 

3 CrowdRun 0.37 2.99 2.14 17.33 18.18 20.35 35.46 46.38 48.19 

4 PeopleOnStreet 0.15 -7.15 -6.22 20.65 21.06 25.66 40.38 52.82 59.84 

5 Traffic 0.32 -9.01 -15.64 20.86 16.15 23.76 36.52 41.70 42.65 

6 Nebuta_Festival 6.91 47.07 42.79 20.02 16.68 5.40 46.90 33.35 16.72 

Average -0.05 -0.89 -2.11 17.65 9.30 18.50 33.57 33.03 41.41

Table 5B Relative Quality Evaluation Results on HD1080p test set for Intel® Media Server Studio HEVC Software Codec 

in TU1, TU4, and TU7 modes with respect to MPEG HEVC HM14 Codec 

HD1080p Test 

Sequence 



1 Park_joy 1.13 5.41 1.63 13.42 16.96 27.48 26.40 32.88 42.50 

2 Ducks_take_off 0.55 2.78 7.08 10.49 -3.20 21.87 20.40 11.05 38.23 

3 CrowdRun 1.03 3.32 2.83 16.51 22.24 23.07 35.12 48.29 50.21 

4 TouchDownPass 0.83 -5.35 -4.83 19.31 17.53 17.80 38.34 53.74 50.80 

5 BQTerrace 4.06 -14.83 -22.44 26.64 24.92 31.83 49.85 41.01 61.06 

6 ParkScene 2.69 -6.04 -7.99 17.45 21.24 19.67 33.31 40.70 36.60 

Average 1.71 -2.45 -3.96 17.30 16.61 23.62 33.90 37.94 46.57 

From Table 5A and 5B we can see that results on encoding server configuration used for E5 for all three TU’s 

provide identical high video quality to the configuration tested earlier for i7 processor 4770K, 4-core used for 

all other tests in this white paper. 

Encoder Quality vs Performance Tradeoff for TU Modes for Computational 

Scalability Tests for 4:2:0 8-bit 

Next, we perform performance analysis tests on the aforementioned E5 server configuration. Fig. 11A shows 

results of this measurement which when compared to earlier tests on i7 4770k configuration for TU1 quality 

demonstrates a speedup factor of 3.5x for UHD content and 2.1x for HD content. This also shows the 

benefits of high throughput server configuration for higher resolution (UHD) content. The E5 server 

configuration also shows much reduced load on the cores, allowing additional similar encodes to run on the 

server - this item is discussed further a little later via Figure 12. 


5.0 

4.5 

4.0 

3.5 

3.0 

2.5 

2.0 

1.5 

1.0 

0.5 

0.0 


4.34 (289x) 

HD1080p 

1.39 (347x) 

0.015 0.004 

Resolution 



UHD4K 



37 

Next, Table 6A and 6B show encoding speed measurements for each sequence of UHD4K and HD1080p test sets. 



Table 6A Encoding Speed of UHD4K test set on Media Server Studio HEVC Software Codec in TU1, TU4, and 

TU7 modes. For reference the average Speed of HM14 Encoder for this test set is .004 fps. 

UHD4K Test 

Sequence 

TU1 Enc Speed TU4 Enc Speed TU7 Enc Speed 

fps fps fps 

1 Park_joy 1.24 8.80 24.65 

2 Ducks_take_off 1.01 7.23 23.19 

3 CrowdRun 1.37 11.39 27.28 

4 PeopleOnStreet 1.38 11.68 22.72 

5 Traffic 1.93 13.00 27.87 

6 Nebuta_Festival* 2.83 24.23 56.40 

Average 1.39 10.42 25.14 

* For average encoder speed calculation, for all TU’s Nebuta_Festival is excluded as its size is 2560x1600 while 

others are of size 4K (3840x2160). 

Table 6B Encoding Speed of HD1080p test set on Media Server Studio HEVC Software Codec in TU1, TU4, and 

TU7 modes. For reference the average speed of HM14 Encoder for this test set is .015 fps. 

HD1080p Test 

Sequence 


fps fps fps 

1 Park_joy 3.24 20.59 64.65 

2 Ducks_take_off 2.56 17.61 66.53 

3 CrowdRun 2.92 21.42 70.82 

4 TouchDownPass 4.71 43.83 106.82 

5 BQTerrace 7.52 48.70 118.64 

6 ParkScene 5.08 37.17 98.57 

Average 4.34 31.55 87.67 

In comparing these results on E5 compared to i7 4770k configuration for which measurements were 

shown earlier in Table 3A and 3B, we observe that for UHD4K test set, the speed gain factor of the E5 

configuration over the i7 4770k configuration is 3.5, 3.3, and 3.0 for TU1, TU4, and TU7 respectively. The 

same comparison for HD1080p test set shows that the speed gain factor of the E5 configuration over the 

i7 4770k configuration is 2.1, 2.2, and 2.4 for TU1, TU4, and TU7 respectively. 

A side-by-side comparison on E5 of actual encoding speed in fps of TU1, TU4, TU7, both for UHD4K and 

HD1080p test sets is shown in Figure 11B. 

38



140 

120 

100 

80 

60 

40 

20 

0 

Figure 11B Average encoding speed of all test sets on Media Server Studio HEVC Software Encoder 

Fig. 12 shows the actual CPU load on E5 configuration which appears to be in 45-70% range as compared to 

around 95% observed for the i7 4770k configuration as per Figure 8, discussed earlier. 

CPU Load (%) 

80 

70 

60 

50 

40 

30 

20 

10 

0 

HD1080p 

Resolution 




UHD4K 

CPU load on E5 for MediaServerStudio TU7 

HD1080p 

Resolution 


UHD4K 


Figure 12 CPU load on E5 for Media Server Studio TU7 (fastest cpu only mode) 

39 





5 

0 

-5 

-10 

-15 

-20 

-25 

-30 

-35 

-40 

0 10 20 30 40 50 60 70 80 90 100 

TU1 

TU4 

TU7 

TU4 


HD1080p 

UHD4K 

Figure 13 Quality vs Encoding Speed Tradeoffs of Media Server Studio HEVC Encoding modes 

Finally, Figure 13 shows quality vs performance tradeoffs of the E5 configuration showing that it can almost 

reach real-time encoding of 4:2:0 8-bit UHD4K content. 

TU7 

40


4:2:0 10-bit Coding Quality, and Performance 

tests 


41 









For the purpose of quality evaluation, two test sets for 4:2:0 10-bit content, one for 1080p, and the other 

for 1600p were employed. Each test set is composed of 3 publicly available challenging video test 

sequences of 10-bit bit-depth. Further for each test sequence, 4 well-spaced quantizer (QP) values are 

selected such that HEVC encoding for each sequence generates bit-rates in a wide but moderate range, 

avoiding extreme values. 

Specifically, Table 7A shows selected Ultra Definition 1600p (UHD1600p) test set, and Table 7B shows 

High Definition 1080p (HD1080p) test set. 

Table 7A UHD1600p Test Set for 4:2:0 10-bit and Quantizers used for Codec RD characteristics measurement 


UHD1600p Test Sequence Resolution frm rate #frm Qp1 Qp2 Qp3 Qp4 

1 Nebuta_Festival 2560x1600 60 300 22 27 32 37 

2 SteamLocomotiveTrain 2560x1600 60 300 22 27 32 37 

3 ReadySetGo 2560x1600 60 300 22 27 32 37 

From UHD1600p test sequences referred to in Table 7A, Nebuta_Festival, and SteamLocomotiveTrain are 

MPEG HEVC test sequences of 2560x1600 resolution that can be obtained from MPEG distribution sites 

such as ftp://hvc:US88Hula@ftp.tnt.uni-hannover.de/testsequences. Further, the last sequence, 

ReadySetGo, of this test set is also of size 2560x1600 but obtained by windowing into an original 

3840x2160 resolution video using an offset of (764, 560) for top left hand corner - this original video can 

be found at http://ultravideo.cs.tut.fi/#testsequences. 

Table 7B HD1080p Test Set for 4:2:0 10 bit and Quantizers used for Codec RD characteristics measurement 

HD1080p Test Sequence Resolution frm rate #frm 


Qp1 Qp2 Qp3 Qp4 

1 CrowdRun 1920x1080 50 500 26 30 34 38 

2 ParkScene 1920x1080 24 240 23 26 29 32 

3 Seeking 1920x1080 50 500 22 27 32 37 

42 

From HD1080p test sequences referred to in Table 7B, all three sequences CrowdRun, ParkScene, and 

Seeking are of 1920x1080 resolution and can be obtained from http://media.xiph.org/video/derf/.


Before discussing detailed BD rate differences of each Media Server Studio TU mode wrt HM14, we first 


UHD1600p Test 

Sequence 

First, each video sequence of each test set (such as UHD1600p, HD1080p) is encoded using MPEG HEVC 

HM14 Reference Encoder with each of 4 quantizers as specified in Tables 7A-7B. The overall PSNR 

(averaged over frames) for each test sequence for each Qp for each of luma (Y), and associated chroma 

components (U and V) is collected along with the generated coding bitrate. For instance, Tables 8A-8B 

show the luma PSNR (chroma PSNR is also collected but is not shown to keep tables manageable in size) 

and total bitrate for each test sequence for each of 4 Qps is collected and is used in curve fitting to 

generate a continuous RD curve for HM14 encoding of that test sequence. 

Next, Media Server Studio HEVC Software Encoder in each of its 3 main modes: TU1, TU4, and TU7 is 

utilized to perform encoding of each test sequence with each of 4 quantizers (of Table 7A-7B) used in HM 

14 encoding, and PSNRs of Y, U, and V, and total bitrates are collected for each test. 

Table 8A PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on 4:2:0 10-bit UHD1600p Test set 



1 Nebuta_Festival 30.77 48676.1 28.96 19640.4 28.11 8011.4 27.55 3678.2 

2 SteamLocomotive 41.53 22431.0 40.04 5523.4 38.84 2232.8 37.07 1066.4 

3 ReadySetGo 41.78 24443.7 40.04 11730.2 37.80 6343.4 35.31 3663.6 

HD1080p Test 

Sequence 

Table 8B PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on 4:2:0 10-bit HD1080p Test set 



1 CrowdRun 35.13 25614.2 32.62 13979.3 30.24 7652.9 28.13 4301.7 

2 ParkScene 39.59 6432.6 37.98 3467.1 36.37 2064.3 34.73 1195.5 

3 Seeking 37.29 40340.7 34.92 12963.8 32.60 5528.0 30.20 2556.7 






respect to HM14 reference. The BD rate provides a measure of departure of Media Server Studio TU 

mode’s quality in percentage bits with respect to the HM14 reference. 


Table 8A-8B show the measured BD rate percentage bitrate for each of luma and chroma components for 

each test sequence of each test set of Table 7A-7B for each of 3 TU modes being evaluated. Fig. 14A1-14B2 

compares the RD characteristics of two codecs for each test set for TU1 and TU4 modes where the 


43 


For instance, Table 9A shows on UHD1600p test set, the BD rate percentage difference of Media Server 

Studio HEVC in various TU modes wrt HM14 reference. As can be observed on UHD1600p test set, the 


reference), is 5.38%, 23.29%, and 44.69%, higher respectively in TU1, TU4, and TU7 modes. This means that 

for UHD1600p test set, the Media Server Studio codec as compared to HM14, requires around 5% higher 

bitrate in TU1 mode, 23% higher bitrate in TU4 mode, and 45% higher bitrate in TU7 mode to achieve the 

same luma PSNR quality as HM14. 


Table 9A Relative Quality Evaluation Results on 4:2:0 10-bit UHD1600p test set for Intel® Media Server Studio HEVC 

Software Codec in TU1, TU4, and TU7 modes with respect to MPEG HEVC HM14 Codec 

UHD1600p Test 

Sequence 



1 Nebuta_Festival 8.02 23.01 22.22 22.61 6.06 -0.65 45.00 26.50 14.37 

2 SteamLocomotive 2.96 -17.32 -13.10 22.63 6.49 7.87 45.07 27.01 25.16 

3 ReadySetGo 5.17 3.70 2.91 24.65 17.14 19.46 43.98 47.26 49.43 

Average 5.38 3.13 4.01 23.29 9.90 8.89 44.69 33.59 29.65 

Further, Table 9A shows on UHD1600p test set that Media Server Studio HEVC Codec TU1 mode with 

respect to HM14 reference, results in the biggest luma BD rate percentage difference on Nebuta_Festival 

sequence and the smallest luma BD rate percentage difference on SteamLocomotive sequence; the 

corresponding RD comparison curves for TU1 mode for the two cases shown in Fig. 14A1. 

PSNR (dB) 

31 

30 

29 

28 




PSNR (dB) 

42 

41 

40 

39 

38 

Steam_Locomotive Y 



27 

0 20000 40000 60000 


37 

0 10000 20000 30000 


Figure 14A1 RD results of 4:2:0 10-bit UHD1600p scenes with the biggest and the smallest quality 

difference wrt HM14 for Media Server Studio TU1 mode 

44 

Table 9A also shows that Media Server Studio HEVC Codec in TU4 mode with respect to HM14 reference, 

results in the biggest luma BD rate percentage difference on ReadySetGo sequence and the smallest luma

BD rate percentage difference on sequence Nebuta_Festival the corresponding RD comparison curves for 

TU4 mode for the two cases are shown in Fig. 14A2. 

PSNR (dB) 

43 

42 

41 

40 

39 

38 

37 

36 

35 

34 

Figure 14A2 RD results of 4:2:0 10-bit UHD1600p scenes with the biggest and the smallest quality 

difference wrt HM14 for Media Server Studio TU4 mode 

Next, Table 9B shows BD rate percentage differences of Media Server Studio HEVC wrt HM14 on the 


over HM14 reference, is 2.9%, 9.8%, and 33.7% respectively in TU1, TU4, and TU7 modes. 

Table 9B Relative Quality Evaluation Results on 4:2:0 10-bit HD1080p test set for Intel® Media Server Studio HEVC 


HD1080p Test 

Sequence 

ReadySetGo Y 



0 10000 20000 30000 




1 CrowdRun 2.19 1.21 0.72 15.88 19.59 21.04 34.75 45.79 48.36 

2 ParkScene 3.08 1.80 8.67 15.88 20.53 30.33 30.80 41.00 47.29 

3 Seeking 3.26 -0.79 -1.38 -2.31 -1.49 -0.43 35.42 39.99 40.70 

Average 2.85 0.74 2.67 9.82 12.87 16.98 33.66 42.26 45.45 

PSNR (dB) 

31 

30 

29 

28 

27 




0 20000 40000 60000 



Further, Table 9B shows on HD1080p test set that Media Server Studio HEVC Codec in TU1 mode with 

respect to HM14 reference, results in the biggest luma BD rate percentage difference on Seeking 1080p 

sequence and the smallest luma BD rate percentage difference on CrowdRun 1080p sequence; the 


45 



PSNR (dB) 

38 

37 

36 

35 

34 

33 

32 

31 

30 

0 20000 40000 60000 


Figure 14B1 RD results of 4:2:0 10-bit HD1080p scenes with the biggest and the smallest quality difference 

wrt HM14 for Media Server Studio TU1 mode 

Table 9B also shows that Media Server Studio HEVC Codec in TU4 mode with respect to HM14 reference, 

results in the biggest luma BD rate percentage difference on CrowdRun sequence and the smallest BD 

rate percentage difference for luma on Seeking sequence; the corresponding comparison curves for TU4 

for the two cases are shown in Fig. 14B2. 

PSNR (dB) 

36 

35 

34 

33 

32 

31 

30 

29 

28 

27 

Seeking Y 



CrowdRun Y 



0 10000 20000 30000 


PSNR (dB) 

PSNR (dB) 

36 

35 

34 

33 

32 

31 

30 

29 

28 

38 

37 

36 

35 

34 

33 

32 

31 

30 

29 

CrowdRun Y 


0 10000 20000 30000 


Seeking Y 



0 20000 40000 60000 




46 

To summarize, for 4:2:2 8-bit content the luma BD rate percentage bitrate difference of the Media Server 

Studio HEVC in various modes with respect to an ideal (but slow) HM14 reference is as follows. 

 

TU1 mode on average is only 2 to 4% worse in BD rate as compared to HM14 on all test sets. This is 

highest quality mode in Media Server Studio.

TU4 mode is -2 - 16% worse in BD rate as compared to HM14 0. As shown later, this mode provides an 

excellent tradeoff of quality vs speed. 

TU7 mode is 30 - 36% worse in BD rate as compared to HM14. As shown later, this is the fastest 

software only mode in Media Server Studio. 







PC Platform (Intel® Core i7-4770K CPU @ 3.5 GHz – 4 Cores/8Threads) is employed. 






1.6 

1.4 

1.2 

1.0 

0.8 

0.6 

0.4 

0.2 

0.0 


1.38 (98x) 

HD1080p 

0.89 (81x) 

0.014 0.011 

Resolution 



UHD1600p 



From Fig. 15A it can be seen that the encoding speed of the TU1 mode is 80 to 100 times (multithreaded 

on 4 cores) the speed of HM14 (single threaded/1 core) for the two resolution test sets. 

47 

Next in Table 10A-10B we show measurement of encoding speed of Media Server Studio HEVC Software 

Encoder on each sequence of the two resolution test sets in each of the three TU1, TU4, and TU7 modes. 


Table 10A Encoding Speed of 4:2:0 10-bit UHD1600p test set on Media Server Studio HEVC Software Codec in TU1, TU4, and 

TU7 modes. For reference the average Speed of HM14 Encoder for this test set is .011 fps. 

UHD1600p Test 

Sequence 


fps fps fps 

1 Nebuta_Festival 0.85 5.97 14.91 

2 SteamLocomotive 1.18 7.68 20.75 

3 ReadySetGo 0.64 5.76 14.92 

Average 0.89 6.47 16.86 


Table 10B Encoding Speed of 4:2:0 10-bit HD1080p test set on Media Server Studio HEVC Software Codec in TU1, TU4, and 

TU7 modes. For reference the average speed of HM14 Encoder for this test set is .014 fps. 

HD1080p Test 

Sequence 


fps fps fps 

1 CrowdRun 1.12 8.72 22.32 

2 ParkScene 1.91 12.14 31.32 

3 Seeking 1.10 8.70 23.00 

Average 1.38 9.85 25.55 

As can be seen from Table 10A-10B the encoding speed of TU4 mode is around 7.2x of that of TU1 mode, 

and that the encoding speed of TU7 mode is 2.6 times that of TU4 mode. In other words TU7 mode 

reflects an overall speed of around 19 times that of TU1 mode. Also, since TU1 (for 4 core multithreaded 

coding) is typically around 98 times the speed of HM, TU7 reflects a typical speedup factor of around 1700 

over HM14! 

A side-by-side comparison of encoding speed of different TU modes and for the 2 different resolution test 

sets is shown next by the bar graphs of Fig. 15B. 

48


30 

25 

20 

15 

10 

5 

0 

HD1080p 


Resolution 




UHD1600p 

Figure 15B Average encoding speed of all 4:2:0 10-bit test sets on Media Server Studio HEVC Software Encoder 




Figure 16 shows a comparison of Quality (negative BD rate percentage difference wrt HEVC HM14) vs 

Encoding Performance (fps) for each of the two resolution test sets at each of the three TU1, TU4, and 

TU7 encoding modes. The y-axis basically shows the quality difference in terms of loss of BD rate 


TU7 operating points. 

To summarize, for 4:2:0 10-bit content, encoding performance-wise Media Server Studio HEVC codec in 

different modes achieves the following speedup of HEVC encoding on 4 core Reference PC platform 

specified earlier. 


49 



0 

0 5 10 15 20 25 30 

TU1 


 

 


-5 

-10 

-15 

-20 

-25 

-30 

-35 

-40 

-45 

-50 

TU4 

TU4 

Figure 16 Quality vs Encoding Speed Tradeoffs for 4:2:0 10-bit content for Media Server Studio HEVC Encoding TU 

modes 

For 4:2:0 10-bit UHD1600p test set, Media Server Studio encoder in TU1, TU4, and TU7 modes, 

correspondingly on average provides 0.89 fps, 6.47 fps, and 16.86 fps. This reflects corresponding 

speedup factors wrt HM14 (single threaded) of 81, 588, and 1533. 

For 4:2:0 10-bit HD1080p test set, Media Server Studio encoder in TU1, TU4, and TU7 modes, 



TU7 


HD1080p 

UHD1600p 

TU7 

50



tests 


51 









For the purpose of quality evaluation, two test sets of 4:2:2 8-bit content, one for 1080p, and the other 

one for 720p resolution are used with each test set consisting of 3 publicly available challenging video test 

sequences. Further for each test sequence, 4 well-spaced quantizer (QP) values are selected such that 

HEVC encoding for each sequence generates bit-rates in a wide but moderate range, avoiding extreme 

values. 

Specifically, Table 11A shows selected High Definition (HD1080p) test set, and Table 11B shows 

Intermediate Definition 720p (720p) test set. 

Table 11A HD1080p Test Set and Quantizers used for Codec RD characteristics measurement 

HD1080p Test Sequence Resolution frm rate #frm 



1 RushFieldCuts 1920x1080 30 570 22 27 32 37 

2 Red_Kayak 1920x1080 30 570 22 27 32 37 

3 TouchDownPass 1920x1080 30 570 23 26 30 34 

From 4:2:2 8-bit 1080p test sequences referred to in Table 11A, all three sequences, ie, RushFieldCuts, 

Red_Kayak, and TouchDownPass can be obtained from http://media.xiph.org/video/derf/. 

Table 11B 720p Test Set and Quantizers used for Codec RD characteristics measurement 

720p Test Sequence Resolution frm rate #frm 



1 CrowdRun 1280x720 50 497 26 30 34 38 


3 Park_joy 1280x720 50 500 26 29 33 37 

52 

From 4:2:2 8-bit 720p test sequences referred to in Table 11B, all three sequences CrowdRun, 

Ducks_take_off, and Park_joy can be obtained from http://media.xiph.org/video/derf/.




HD1080p Test 

Sequence 

First, each video sequence of each test set (such as HD1080p, 720p) is encoded using MPEG HEVC HM14 

Reference Encoder with each of 4 quantizers as specified in Tables 11A-11B. The overall PSNR (averaged 

over frames) for each test sequence for each Qp for each of luma (Y), and associated chroma 

components (U and V) is collected along with the generated coding bitrate. For instance, Tables 12A-12B 

show the luma PSNR (chroma PSNR is also collected but is not shown to keep tables manageable in size) 

and total bitrate for each test sequence for each of 4 Qps is collected and is used in curve fitting to 

generate a continuous RD curve for HM14 encoding of that test sequence. 

Next, Media Server Studio HEVC Software Encoder in each of its 4 main modes: TU1, TU4, TU7, and 

TU7gacc is utilized to perform encoding of each test sequence with each of 4 quantizers (of Table 11A-11B) 

used in HM 14 encoding, and PSNRs of Y, U, and V, and total bitrates are collected for each test. 

Table 12A PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on 4:2:2 8-bit HD1080p Test set 



1 RushFieldCuts 40.02 19450.9 37.25 7191.2 34.70 3209.9 32.16 1589.2 

2 Red_Kayak 41.89 16506.7 38.76 7987.6 35.62 3685.4 32.88 1653.7 

3 TouchDownPass 40.42 5995.0 39.34 2984.7 37.80 1464.8 36.13 791.4 

720p Test 

Sequence 

Table 12B PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on 4:2:2 8-bit 720p Test set 



1 CrowdRun 34.50 16213.2 31.63 8760.9 29.05 4684.5 26.84 2492.8 

2 Ducks_take_off 32.29 15641.9 30.42 8978.0 28.21 4205.4 27.24 2867.8 

3 Park_joy 33.90 19524.1 31.64 11946.9 28.87 5927.4 26.52 2861.1 









Table 13A-13B show the measured BD rate percentage bitrate for each of luma and chroma components 

for each test sequence of each test set of Table 11A-11B for each of 4 TU modes being evaluated. Fig. 17A1- 

17B2 compares the RD characteristics of two codecs for each test set for TU1 and TU4 modes where the 


53 


For instance, Table 13A shows on HD1080p test set, the BD rate percentage difference of Media Server 

Studio HEVC in various TU modes wrt HM14 reference. As can be observed on HD1080p test set, the 


reference), is 5.1%, 16.9%, 33.7%, and 37.8%, higher respectively in TU1, TU4, TU7, and TU7gacc modes. This 

means that for HD1080p test set, the Media Server Studio codec as compared to HM14, requires around 

5% higher bitrate in TU1 mode, 17% higher bitrate in TU4 mode, 34% higher bitrate in TU7 mode, and 38% 

higher bitrate in TU7gacc mode to achieve the same luma PSNR quality as HM14. 


Table 13A Relative Quality Evaluation Results on 4:2:2 8-bit HD1080p test set for Intel® Media Server Studio HEVC 

Software Codec in TU1, TU4, TU7, and TU7gacc modes with respect to MPEG HEVC HM14 Codec 

HD1080p Test 

Sequence 



1 RushFieldCuts 5.97 -8.21 -7.79 17.62 13.59 17.81 35.61 41.99 51.07 38.21 43.48 52.77 

2 Red_Kayak 3.19 -17.06 -13.96 12.44 -9.92 3.45 25.06 8.92 30.58 26.46 11.50 31.31 

3 TouchDownPass 6.22 -4.51 -4.29 20.67 18.93 17.65 40.45 57.78 54.72 48.83 61.28 58.16 

Average 5.12 -9.92 -8.68 16.91 7.53 12.97 33.71 36.23 45.46 37.83 38.75 47.41 

Further, Table 13A shows on 4:2:2 8-bit HD1080p test set that Media Server Studio HEVC Codec TU1 mode 

with respect to HM14 reference, results in the biggest luma BD rate percentage difference on 

TouchdownPass sequence and the smallest luma BD rate percentage difference on Red_Kayak sequence; 

the corresponding RD comparison curves for TU1 mode for 

the two cases shown in Fig. 17A1. 

PSNR (dB) 

41 

40 

39 

38 

37 

36 

TouchdownPass Y 



PSNR (dB) 

44 

42 

40 

38 

36 

34 

32 

Red_Kayak Y 



35 

0 2000 4000 6000 8000 


30 

0 5000 10000 15000 20000 


Figure 17A1 RD results of 4:2:2 8-bit HD1080p scenes with the biggest and the smallest quality difference 


54 

Table 13A also shows that on 4:2:2 8-bit HD1080p test set Media Server Studio HEVC Codec in TU4 mode 


Touchdown_Pass sequence and the smallest luma BD rate percentage difference on Red_Kayak sequence; 

the corresponding RD comparison curves for TU4 mode for the two cases are shown in Fig. 17A2.

720p Test 

Sequence 

PSNR (dB) 

41 

40 

39 

38 

37 

36 

35 

TouchdownPass Y 



0 2000 4000 6000 8000 


Figure 17A2 RD results of 4:2:2 8-bit 1080p scenes with the biggest and the smallest quality difference wrt 

HM14 for Media Server Studio TU4 mode 

Next, Table 13B shows BD rate percentage differences of Media Server Studio HEVC wrt HM14 on 4:2:2 8- 

bit 720p test set. The average luma BD rate percentage difference of Media Server Studio HEVC Codec 

over HM14 reference, is 4.7%, 14.6%, 28.7%, and 30.8% respectively in TU1, TU4, TU7, and TU7 gacc modes. 

Table 13B Relative Quality Evaluation Results on 4:2:2 8-bit 720p test set for Intel® Media Server Studio HEVC 

Software Codec in TU1, TU4, TU7, and TU7gacc modes with respect to MPEG HEVC HM14 Codec 



1 CrowdRun 4.61 -7.55 -7.60 17.88 17.30 18.73 36.66 37.36 40.45 38.59 38.98 41.99 

2 Ducks_take_off 3.78 -0.97 -3.67 10.83 5.80 29.07 21.43 16.91 41.06 23.07 18.75 43.14 

3 Park_joy 5.80 -4.18 -8.04 14.93 15.20 25.00 28.10 27.30 39.22 30.87 29.72 41.60 

Average 4.73 -4.24 -6.43 14.55 12.77 24.27 28.73 27.19 40.24 30.84 29.15 42.24 

PSNR (dB) 

Further, Table 13B shows that on 4:2:2 8-bit 720p test set, Media Server Studio HEVC Codec in TU1 mode 

with respect to HM14 reference, results in the biggest luma BD rate percentage difference on Park_Joy 

sequence and the smallest luma BD rate percentage difference on Ducks_take_off sequence; the 


44 

42 

40 

38 

36 

34 

32 

30 

Red_Kayak Y 



0 5000 10000 15000 20000 



55 



PSNR (dB) 



Table 13B also shows that on 4:2:2 8-bit 720p test set, Media Server Studio HEVC Codec in TU4 mode with 

respect to HM14 reference, results in the biggest luma BD rate percentage difference on CrowdRun 

sequence and the smallest BD rate percentage difference for luma on Ducks_take_off sequence; the 

corresponding comparison curves for TU4 for the two cases are shown in Fig. 17B2. 

PSNR (dB) 

35 

34 

33 

32 

31 

30 

29 

28 

27 

26 

35 

34 

33 

32 

31 

30 

29 

28 

27 

26 

park_joy Y 



0 10000 20000 30000 


CrowdRun Y 



0 5000 10000 15000 20000 


PSNR (dB) 

PSNR (dB) 

33 

32 

31 

30 

29 

28 

27 

33 

32 

31 

30 

29 

28 

27 

26 

ducks_Take_off Y 



0 5000 10000 15000 20000 


ducks_Take_off Y 



0 5000 10000 15000 20000 


Figure 17B2 RD results of 4:2:2 8-bit 720p scenes with the biggest and the smallest quality difference wrt 


To summarize, for 4:2:2 8-bit content the luma BD rate percentage bitrate difference of the Media Server 

Studio HEVC in various modes with respect to an ideal (but slow) HM14 reference is as follows. 

56 

 

 

TU1 mode is only around 5% worse in BD rate as compared to HM14 on all test sets. This is highest 

quality mode in Media Server Studio. 

TU4 mode is 14.5 - 17% worse in BD rate as compared to HM14. As shown later, this mode provides an 

excellent tradeoff of quality vs speed.

TU7 mode is 21 - 37% worse in BD rate as compared to HM14. As shown later, this is the fastest 


TU7gacc mode is 31 - 38% worse in BD rate as compared to HM14. As shown later, this is the fastest 

mode in Media Server Studio. 













3.0 

2.5 

2.0 

1.5 

1.0 

0.5 

0.0 


2.56 (64x) 

HD720p 

1.25 (63x) 

0.040 0.020 

Resolution 



HD1080p 


Figure 18A Average encoding speed comparison of 4:2:2 8-bit content on Media Server Studio TU1 mode 

with HM14 

From Fig. 18A it can be seen that the encoding speed of the TU1 mode is around 61 times (multithreaded 

on 4 cores) the speed of HM14 (single threaded/1 core) for the two resolution test sets. 

57 


Table 14A Encoding Speed of 4:2:2 8-bit HD1080p test set on Media Server Studio HEVC Software Codec in TU1, TU4, and TU7 

modes. For reference the average Speed of HM14 Encoder for this test set is 0.020 fps. 

HD1080p Test 

Sequence 



1 RushFieldCuts 1.18 11.89 31.10 41.23 

2 Red_Kayak 0.83 10.20 27.36 38.73 

3 TouchDownPass 1.74 16.33 43.50 55.16 

Average 1.25 12.81 33.99 45.04 


Table 14B Encoding Speed of 4:2:2 8-bit 720p test set on Media Server Studio HEVC Software Codec in TU1, TU4, and TU7 

modes. For reference the average speed of HM14 Encoder for this test set is 0.040 fps. 

720p Test 

Sequence 

TU1 Enc Speed TU4 Enc Speed TU7 Enc Speed TU7gaccEnc Speed 


1 CrowdRun 2.39 18.67 47.68 65.51 

2 Ducks_take_off 2.40 17.52 48.19 63.99 

3 Park_joy 2.87 19.37 47.56 64.31 

Average 2.56 18.52 47.81 64.60 

As can be seen from Table 14A-14B the encoding speed of TU4 mode is in around 8.7 times that of TU1 

mode, encoding speed of TU7 mode is 2.4 times that of TU4 mode, and encoding speed of TU7gacc mode 

is around 1.3 times that of TU7 mode. In other words TU7gacc mode reflects an overall speed of around 

28 times that of TU1 mode. Also, since TU1 for 4 core muli-threaded coding is typically around 63 times the 

speed of HM, TU7gacc reflects a typical speedup factor of around 1800 over HM14! 

A side-by-side comparison of encoding speed of different TU modes and for the 2 different resolution test 

sets is shown next by the bar graphs of Fig. 18B. 

58


Figure 18B Average encoding speed of all 4:2:2 8-bit test sets on Media Server Studio HEVC Software 

Encoder 

Fig. 19 shows results of TU7gacc as compared to TU7 in terms of its capability of reducing the load on the 

CPU. For 1080p content TU7gcac is currently able to reduce the load on CPU by a few percent from 

enabling CPU to be used by other applications. This is in addition to around 32% encoding speedup that 

TU7gacc provides over TU7 while achieving identical video quality. 

CPU Load (%) 

70 

60 

50 

40 

30 

20 

10 

0 

110 

100 

90 

80 

70 

HD720p 


Resolution 





HD1080p 

CPU load Comparison between MediaServerStudio TU7 and TU7gacc 




60 

50 

HD720p 

Resolution 

HD1080p 

59 

Figure 19 CPU load difference between TU7 and TU7 gacc modes 







Encoding Performance (fps) for each of the two resolution test sets at each of the four TU1, TU4, TU7, 

and TU7gacc encoding modes. The y-axis basically shows the quality difference in terms of loss of BD rate 


TU7 to TU7 gacc operating points. 





0 

-5 

-10 

-15 

-20 

-25 

-30 

-35 

-40 

TU1 

TU1 

TU4 


0 10 20 30 40 50 60 70 

TU4 

TU7 


TU7gacc 

TU7 

HD720p 

HD1080p 

TU7gacc 


modes 

60 

 

 

For 4:2:2 8-bit 1080p test set, Media Server Studio encoder in TU1, TU4, TU7, and TU7 gacc modes, 

correspondingly on average provides 1.25 fps, 12.8 fps, 34 fps, and 45 fps. This reflects corresponding 

speedup factors wrt HM14 (single threaded) of around 62, 640, 1699, and 2252. 

For 4:2:2 8-bit 720p test set, Media Server Studio encoder in TU1, TU4, and TU7 modes, 

correspondingly on average provides 2.55 fps, 18.5 fps, 47.8 fps, and 64 fps. This reflects 

corresponding speedup factors wrt HM14 (single threaded) of 64, 563, 1195, and 1615.



tests 


61 









For the purpose of quality evaluation, four test sets, one for each of the 4 main resolutions are selected 

with each test set consisting of 3 publicly available video test sequences. Further for each test sequence, 

4 well-spaced standard quantizer (QP) values used by MPEG for RD analysis for comparing HEVC encoders 

are used. 

Specifically, Table 15 shows the selected 4:2:2 10-bit Ultra High Definition 4K (UHD4K) test set. 

Table 15 4:2:2 10-bit UHD4K Test Set and Quantizers used for Codec RD characteristics measurement 

UHD4K Test Sequence Resolution frm rate #frm 



1 FastWalk 4096x2304 50 300 22 27 32 37 

2 Footbridge 4096x2304 24 300 22 27 32 37 

2 Venetian_crossing 4096x2304 60 300 22 27 32 37 

Since there is relatively little, public, free test content test for 4:2:2 10-bit UHD4K (or other resolution), the 

above noted (non-free) content was employed. The content is described at http://www.testvid.com/alltest-material/4k-media/84-test-4k-quad-hd-usa-europe-asia-video-t2v041 

and can be purchased from 

http://www.testvid.com/. 




62 

First, each video sequence of the 4:2:2 10-bit UHD4K is encoded using MPEG HEVC HM14 Reference 

Encoder with each of 4 quantizers as specified in Table 15. The overall PSNR (averaged over frames) for 

each test sequence for each Qp for each of luma (Y), and associated chroma components (U and V) is 

collected along with the generated coding bitrate. For instance, Tables 16 shows the luma PSNR (chroma 

PSNR is also collected but is not shown to keep tables manageable in size) and total bitrate for each test 

sequence for each of 4 Qps is collected and is used in curve fitting to generate a continuous RD curve for 

HM14 encoding of that test sequence.

Next, Media Server Studio HEVC Software Encoder in each of its 3 main modes: TU1, TU4, and TU7 is 

utilized to perform encoding of each test sequence with each of 4 quantizers (of Table 15) used in HM 14 

encoding, and PSNRs of Y, U, and V, and total bitrates are collected for each test. 

UHD4k Test 

Sequence 

Table 16 PSNR Y, and Total Bitrate for each of 4 QPs by HM 14 Codec on 4:2:2 10-bit UHD4K Test set 



1 FastWalk 48.11 38231.4 44.71 16403.4 41.61 7585.5 38.80 3938.2 

2 Footbridge 47.90 8496.9 44.86 3429.7 41.69 1542.9 38.65 740.6 

3 Venetian_Crossing 44.91 36959.6 42.25 10349.2 39.64 4177.3 36.78 2076.0 








Table 17 shows the measured BD rate percentage bitrate for each of luma and chroma components for 

each test sequence of Table 16 for each of 3 TU modes being evaluated. Fig. 21A-21B compares the RD 

characteristics of two codecs on the UHD4K test set for TU1 and TU4 modes where the difference in 

quality is the highest and the lowest wrt HM14 reference. 

For instance, Table 17 shows on UHD4K test set, the BD rate percentage difference of Media Server 

Studio HEVC in various TU modes wrt HM14 reference. As can be observed on UHD4K test set, the 


reference), is 8.3%, 18.2%, and 28.6%, higher respectively in TU1, TU4, and TU7 modes. This means that for 

UHD4K test set, the Media Server Studio codec as compared to HM14, requires around 8% higher bitrate 

in TU1 mode, 18% higher bitrate in TU4 mode, and 29% higher bitrate in TU7 mode to achieve the same 

luma PSNR quality as HM14. 

Table 17 Relative Quality Evaluation Results on 4:2:2 10-bit UHD4K test set for Intel® Media Server Studio HEVC 


UHD4K Test 

Sequence 



1 FastWalk 9.32 -4.87 -2.44 20.31 1.97 3.97 30.58 32.59 29.85 

2 Footbridge 5.62 -2.87 -3.66 16.80 20.63 14.05 27.63 43.79 47.56 

3 Venetian_crossing 10.0 5.31 5.05 17.38 11.92 10.95 27.54 36.61 35.94 

Average 98.34 -0.81 -0.35 18.17 11.51 9.66 28.58 37.66 37.78 


63 


Further, Table 17 shows that on 4:2:2 10-bit UHD4K test set, Media Server Studio HEVC Codec TU1 mode 


Venetian_Crossing sequence and the smallest luma BD rate percentage difference on Footbridge 

sequence; the corresponding RD comparison curves for TU1 mode for the two cases are shown in Fig. 21A. 


PSNR (dB) 

46 

45 

44 

43 

42 

41 

40 

39 

38 

37 

36 

Venetian_Crossing Y 



0 20000 40000 60000 


Figure 21A RD results of 4:2:2 10-bit UHD4K scenes with the biggest and the smallest quality difference wrt 


Table 17 also shows that on 4:2:2 10-bit UHD4K test set, Media Server Studio HEVC Codec in TU4 mode 

with respect to HM14 reference, results in the biggest luma BD rate percentage difference on FastWalk 

sequence and the smallest luma BD rate percentage difference on Footbridge sequence; the 

corresponding RD comparison curves for TU4 mode for the two cases are shown in Fig. 21B. 

PSNR (dB) 

49 

47 

45 

43 

41 

39 

37 

FastWalk Y 



0 20000 40000 60000 


PSNR (dB) 

PSNR (dB) 

50 

48 

46 

44 

42 

40 

38 

49 

47 

45 

43 

41 

39 

37 

Footbridge Y 



0 5000 10000 


Footbridge Y 



0 5000 10000 


64 

Figure 21B RD results of 4:2:2 10-bit UHD4K scenes with the biggest and the smallest quality difference wrt 


To summarize, the luma BD rate percentage bitrate difference of the Media Server Studio HEVC in various 

modes with respect to an ideal (but slow) HM14 reference is as follows.

TU1 mode on average is 8% worse in BD rate as compared to HM14. This is highest quality mode in 

Media Server Studio. 

TU4 mode on average is 18% worse in BD rate as compared to HM14. As shown later, this mode 

provides an excellent tradeoff of quality vs speed. 

TU7 mode on average is 29% worse in BD rate as compared to HM14. As shown later, this is the fastest 











quality) mode and MPEG HEVC HM14 on 4:2:2 10-bit UHD test set. Results of these measurements 



0.6 

0.5 

0.4 

0.3 

0.2 

0.1 

0.0 


0.55 (138x) 

UHD4K 

Resolution 

0.004 




Figure 22A Average encoding speed comparison of 4:2:2 10-bit content on Media Server Studio TU1 mode 

with HM14 

65 


From Fig. 22A it can be seen that the encoding speed of the TU1 mode is over 120 times (multithreaded on 4 

cores) the speed of HM14 (single threaded/1 core) for the 4:2:2 10-bit UHD4K test set. 

Next in Table 18 we show measurement of encoding speed of Media Server Studio HEVC Software 

Encoder on each sequence on each of the three TU1, TU4, and TU7 modes. 


Table 18 Encoding Speed of 4:2:2 10-bit UHD4K test set on Media Server Studio HEVC Software Codec in TU1, TU4, and TU7 

modes. For reference the average Speed of HM14 Encoder for this test set is .004 fps. 

UHD4K Test 

Sequence 


fps fps fps 

1 FastWalk 0.24 2.71 7.14 

2 Footbridge 0.68 3.94 12.11 

3 Venetian_crossing 0.72 3.90 9.92 

Average 0.55 3.52 9.72 

As can be seen from Table 18 for the 4:2:2 10-bit UHD4K test set, the encoding speed of TU4 mode is 6.4 

times that of TU1 mode, and that the encoding speed of TU7 mode is 2.8 times that of TU4 mode. In 

other words TU7 mode reflects an overall speed of 17.7 times that of TU1 mode. Also, since TU1 for 4 core 

multithreaded coding is 137 times the speed of HM, TU7 reflects a typical speedup factor of 2430 over 

HM14! 

A side-by-side comparison of encoding speed of different TU modes and for the 4:2:2 10-bit UHD4K 

resolution test sets is shown next by the bar graphs of Fig. 22B. 


12 

10 

8 

6 

4 

2 





0 

UHD4K 

Resolution 

66 

Figure 22B Average encoding speed of 4:2:2 10-bit UHD4K test set on Media Server Studio HEVC Software 

Encoder



Server Studio HEVC Encoder for the UHD4K test set tested. 


Encoding Performance (fps) for the test set at each of the three TU1, TU4, and TU7 encoding modes. The 

y-axis basically shows the quality difference in terms of loss of BD rate percentage difference in the 

process of increasing speed up of the encoder in going from TU1 to TU4 to TU7 operating points. 





0 

-5 

-10 

-15 

-20 

-25 

-30 


0 2 4 6 8 10 12 

TU4 

TU7 

UHD4K 


-35 



modes 

 

For 4:2:2 10-bit UHD4K test set, Media Server Studio encoder in TU1, TU4, and TU7 modes, 



67 


Summary 


In this white paper we first presented an overview of the new MPEG HEVC compression standard, and 

then introduced Intel® Media Server Studio, a developer product. Next we presented a test methodology 

for quality evaluation of HEVC Codecs and applied the methodology to evaluate quality of Intel® Media 

Server Studio HEVC Software Codec. Encoder Quality versus performance tradeoffs of various modes of 

Intel Media Server Studio HEVC Software Encoder were discussed next. This was then followed by 

evaluation of performance of HEVC Software Decoder of Intel® Media Server Studio. Appendix A 

summarizes the quality, encode performance and decode performance achieved. 

As can be seen from results of thorough testing on variety of video resolutions, bit-depth, and chroma 

resolutions, Media Server Studio HEVC Software Codec delivers impressive encoding and decoding 

performance while achieving excellent tradeoffs in quality to meet the needs of highly demanding 

communication infrastructure developers, video software experts and developers, enterprise video 

device manufacturers, cloud based media services providers, and others. 

References 

[1] G. Sullivan, J.-R. Ohm, W.-J. Han, T. Wiegand, “Overview of the High Efficiency Video Coding (HEVC) 

Standard,” IEEE Transactions on Circuits and Systems for Video Technology, Vol. 22, No. 12, pp 1649-1668, 

Dec 2012. 

[2] JCT-VC Video Subgroup Editors, “ISO/IEC JTC1/SC29/WG11 MPEG 2012/N13154: HM9 High Efficiency 

Video Coding (HEVC) Test Model 9 Encoder Description,” Oct. 2012, Shanghai. 

[3] Joint Collaborative Team on Video Coding, “JCTVC-L1003_v34: High Efficiency Coding Video Coding 

(HEVC) Text specification draft 10,” Jan. 2013, Geneva. 

[4] Leonardo Chiariglione (Ed.), “The MPEG Representation of Digital Media,” Springer, ISBN 978-1-4419- 

6183-9, 2012. 

[5] J. Wang, X. Yu, D. He, “JCTVC-F270: On BD-Rate Calculation,” Jul 2011, Torino. Also based on ...J. Zhao, 

Y. Su, A. Segall, “ITU-T COM16-C-404: On the Calculation of PSNR and bit-rate differences for the SVT test 

data,” April 2008. 

[6] A. Puri, X. Chen, A. Luthra, “Video Coding using the H.264/MPEG-4 AVC Compression Standard,” Signal 

Processing Image Communication, pp. 793-849, 2004. 

68

Appendix A 

A1: Summary of 4:2:0 8-bit Coding Quality and Performance 

Tests of Intel® Media Server Studio HEVC Codec 

Resolution 

Quality difference with respect to HEVC HM14 Ref. Codec 

Media Server Media Server 

Studio TU4 Studio TU7 

Luma BD rate Luma BD rate 

Media Server 

Studio TU1 

Luma BD rate 


Studio TU7gacc 

Luma BD rate 

Standard/Extended Definition (SD/XD) 0.07% 19.44% 39.33% 44.65% 

High Definition 720p (HD720p) 1.56% 16.66% 34.94% 39.46% 


Ultra High Definition 4K (UHD4K) -0.05% 17.65% 33.57% 35.22% 

Resolution 

Encode Speed (Speed factor wrt HM14 @ ) on Haswell α - 4 cores 



Speed, fps 

(Speed 

Factor) 



Speed, fps 

(Speed Factor) 



Speed, fps 




Speed, fps 

(Speed Factor ) 

Standard/Extended Definition (SD/XD) 5.78 (141x) 47.27 (1153x) 120.12 (2930x) 143.58 (3501x) 

High Definition 720p (HD720p) 3.04 ( 117x) 23.45 (902x) 58.15 (2237x) 78.29 (3011x) 

High Definition 1080p (HD1080p) 2.09 (139x) 14.12 (941x) 36.93 (2462x) 48.17 (3211x) 

Ultra High Definition 4K (UHD4K) 0.4 (100x) 3.13 (783x) 8.47 (2118x) 11.41 (2850x) 

Resolution 



Speed, fps 

Decode Speed on Haswell α - 4 cores 



Speed, fps 



Speed, fps 

Standard/Extended Definition (SD/XD) 463 879 1024 

High Definition 720p (HD720p) 274 461 514 

High Definition 1080p (HD1080p) 200 266 292 

Ultra High Definition 4K (UHD4K) 67 61 79 


α Intel® Core-i7 Processor 4770k: 4 Core, 3.5 GHz. 

@ 

HM14 runs single threaded only. 

69 



70

A2: Summary of Computational Scalability Tests (E5) of 4:2:0 

8-bit HD and UHD Coding by Intel® Media Server Studio HEVC 

Codec 

Resolution 



TU1 Luma BD rate 





High Definition 1080p (HD1080p) 1.71% 17.30% 33.90% 

Ultra High Definition 4K (UHD4K) -0.05% 17.65% 33.57% 

Resolution 

Encode Speed (Speed factor wrt HM14 @ ) on E5 # - 12 cores 


TU1 Speed, fps 








High Definition 1080p (HD1080p) 4.34 (289x) 31.55 (2103x) 87.67 (5845x) 

Ultra High Definition 4K (UHD4K) 1.39 (347x) 10.42 (2605x) 25.14 (6285x) 


71 


Appendix B 

B1: Summary of 4:2:0 10-bit Coding Quality and Performance of 

Intel® Media Server Studio HEVC Codec 


Resolution 








High Definition 1080p (HD1080p) 2.85% 9.82% 33.66% 

Ultra High Definition 1600p (UHD1600p) 5.38% 23.29% 44.69% 

Resolution 











High Definition 1080p (HD1080p) 1.38 (98x) 9.85 (704x) 25.55 (1825x) 

Ultra High Definition 1600p (UHD1600p) 0.89 (81x) 6.47 (588x) 16.86 (1533x) 

72

B2: Summary of 4:2:2 8-bit Coding Quality and Performance of 

Intel® Media Server Studio HEVC Codec 

Resolution 


Media Server Media Server 

Studio TU4 Studio TU7 

Luma BD rate Luma BD rate 



Luma BD rate 



Luma BD rate 



Resolution 




Speed, fps 

(Speed 

Factor) 



Speed, fps 




Speed, fps 




Speed, fps 

(Speed Factor ) 




73 


B3: Summary of 4:2:2 10-bit Coding Quality and Performance 

Tests of Intel® Media Server Studio HEVC Codec 


Resolution 








Ultra High Definition 4K (UHD4K) 8.34% 18.17% 28.58% 

Resolution 











Ultra High Definition 4K (UHD4K) 0.55 (137x) 3.52 (880x) 9.72 (2430x) 

α Intel® Core-i7 Processor 4770k: 4 Core, 3.5 GHz. 

# Intel® Xeon® Processor E5-2697v2: 12 Core, 2.7 GHz. 2 processors per board, total of 24 physical cores. 

@ 

HM14 runs single threaded only. 

74

Appendix C 

C1: Example Encoding Directory and Subdirectories structure 

Codec 

cfg 

encoder_random_access 

_main.cfg 

TAppEncoder.exe 

bin 

TAppDecoder.exe 

HM14 

MSS 

PSNR.exe 

script_par_rdpf_pd_hevc.bat 

script_qp_hevc.bat 

win_x64 

mss_enc_params.cmd 

mss_qp_hev.cmd 

bin 

2fca99749fdb49.. 

1fdd936825ad47.. 

libmfxhw64.dll 

mfxplugin64_he 

vce_sw.dll 

plugin.cfg 

mfxplugin64_he 

vcd_sw.dll 

plugin.cfg 

libmfxsw64.dll 

Park_joy_1920x1080_50.yuv 

.. 

sample_decode.exe 

sample_encode.exe 

HM14_Bitstreams 

MSS_Bitstreams 

Content 

Output 

run_HM14.bat 


75 


C2: Scripts for 4:2:0 content Encoding with HEVC HM 

Encoder 

We now provide example scripts (windows batch file and the other windows batch files it calls) for 

encoding using HEVC HM 14 encoder. 


76 

run_HM14.bat 

REM Script Dir Sequence FPS 

FRM Res qp1 qp2 qp3 qp4 

call Codec\HM14\script_qp_hevc.bat Content park_joy_3840x2160_50_8bit 

50 500 2160 26 29 33 37 

call Codec\HM14\script_qp_hevc.bat Content ducks_take_off_3840x2160_50_8bit 

50 500 2160 28 31 35 37 

call Codec\HM14\script_qp_hevc.bat Content crowdRun_3840x2160_50_8bit 

50 500 2160 26 30 34 38 

call Codec\HM14\script_qp_hevc.bat Content PeopleOnStreet_3840x2160_30_8bit 

30 150 2160 22 27 30 33 

call Codec\HM14\script_qp_hevc.bat Content Traffic_3840x2048_30_8bit 

30 300 2048 18 22 26 30 

call Codec\HM14\script_qp_hevc.bat Content Nebuta_Festival_2560x1600_60_8bit 

60 300 1600 30 33 36 39 

call Codec\HM14\script_qp_hevc.bat Content park_joy_1920x1080_50 

50 500 1080 26 29 33 37 

call Codec\HM14\script_qp_hevc.bat Content ducks_take_off_1920x1080_50 

50 500 1080 28 31 35 37 

call Codec\HM14\script_qp_hevc.bat Content crowdRun_1920x1080_50 

50 500 1080 26 30 34 38 

call Codec\HM14\script_qp_hevc.bat Content touchdown_pass_1920x1080_30 

30 570 1080 23 26 30 34 

call Codec\HM14\script_qp_hevc.bat Content BQTerrace_1920x1080_60 

60 600 1080 25 27 31 34 

call Codec\HM14\script_qp_hevc.bat Content ParkScene_1920x1080_24 

24 240 1080 23 26 29 32

call Codec\HM14\script_qp_hevc.bat Content park_joy_1280x720_50 

50 500 720 26 29 33 37 

call Codec\HM14\script_qp_hevc.bat Content ducks_take_off_1280x720_50 

50 500 720 28 31 35 37 

call Codec\HM14\script_qp_hevc.bat Content crowdRun_1280x720_50 

50 500 720 26 30 34 38 

call Codec\HM14\script_qp_hevc.bat Content City_1280x720_30 

30 300 720 22 24 27 29 

call Codec\HM14\script_qp_hevc.bat Content Crew_1280x720_30 

30 300 720 22 26 29 32 

call Codec\HM14\script_qp_hevc.bat Content Sailormen_1280x720_30 

30 300 720 24 26 28 30 

call Codec\HM14\script_qp_hevc.bat Content BasketballDrillText_832x480_50 

50 500 480 21 24 27 30 

call Codec\HM14\script_qp_hevc.bat Content PartyScene_832x480_50 

50 500 480 24 27 30 33 

call Codec\HM14\script_qp_hevc.bat Content RaceHorses_832x480_30 

30 300 480 24 27 30 33 

call Codec\HM14\script_qp_hevc.bat Content City_704x576_30 

30 300 576 22 24 26 28 

call Codec\HM14\script_qp_hevc.bat C Content Crew_704x576_30 

30 300 576 22 25 28 31 

call Codec\HM14\script_qp_hevc.bat Content Soccer_704x576_30 

30 300 576 22 25 28 31 

pause 

script_qp_hevc.bat 


@echo off 

REM ENCODE_SCRIPT DIR FILE RES QP FPS FRM 

start Codec\HM14\script_par_rdpf_pd_hevc.bat %1 %2 %5 %9 %3 %4 



call Codec\HM14\script_par_rdpf_pd_hevc.bat %1 %2 %5 %6 %3 %4 

77 


script_par_rdpf_pd_hevc.bat 


@echo on 

set KI=-1 

set INPUTDIR=%1 

set SEQUENCE=%2 

set RES=%3 

set QPVAL=%4 

set FRAMERATE=%5 

set FRAMES=%6 

call :ResolutionSetup %RES% W H 

if %KI% gtr -1 ( 

set ENCMODE=RA 

) else ( 

set ENCMODE=1i 

) 

set OUTPUTDIR=HEVC_%SEQUENCE%_%QPVAL%_%ENCMODE% 

if NOT EXIST Output\HM14_Bitstreams\%OUTPUTDIR% ( 

mkdir Output\HM14_Bitstreams\%OUTPUTDIR% 

Codec\HM14\bin\TAppEncoder.exe -c Codec\HM14\bin\cfg\encoder_randomaccess_main.cfg - 

i %INPUTDIR%\%SEQUENCE%.yuv -wdt %W% -hgt %H% -q %QPVAL% -b 

Output\HM14_Bitstreams\%OUTPUTDIR%\HEVC_%SEQUENCE%_%QPVAL%.bin -o 

Output\HM14_Bitstreams\%OUTPUTDIR%\HEVC_%SEQUENCE%_%QPVAL%_recon.yuv -f %FRAMES% - 

fr %FRAMERATE% -ip %KI% > 

Output\HM14_Bitstreams\Logs\HEVC_%SEQUENCE%_%QPVAL%_%ENCMODE%.log 

) 

78 

:ResolutionSetup 

( 

IF %~1 EQU 1600 ( 

set %~2=2560

) 

) else ( 

) 

set %~3=1600 

IF %~1 EQU 2160 ( 

set %~2=3840 

set %~3=2160 

) else ( 

IF %~1 EQU 2048 ( 

set %~2=3840 

set %~3=2048 

) else ( 

IF %~1 EQU 1080 ( 

set %~2=1920 

set %~3=1080 

) else ( 

IF %~1 EQU 720 ( 

set %~2=1280 

set %~3=720 

) else ( 

IF %~1 EQU 576 ( 

set %~2=704 

set %~3=576 

) else ( 

IF %~1 EQU 480 ( 

set %~2=832 

set %~3=480 

) else ( 

echo Unknown resolution. 

exit /b 

) 

) 

) 

) 

) 

) 


79 


C3: Scripts for 4:2:0 content Encoding with Intel® Media 

Server Studio HEVC Encoder 

We now provide example scripts (windows batch file and the other windows batch files it calls) for 

encoding using the MSS HEVC encoder. 


80 

run_MSS_HEVC.bat 

REM Script Sequence FPS qp1 

qp2 qp3 qp4 TU Resolution 

FOR %%G in (speed, balanced, quality) do ( 

call Codec\MSS\mss_qp_hevc.cmd park_joy_3840x2160_50_8bit 50 26 

29 33 37 %%G 2160 

call Codec\MSS\mss_qp_hevc.cmd ducks_take_off_3840x2160_50_8bit 50 28 

31 35 37 %%G 2160 

call Codec\MSS\mss_qp_hevc.cmd crowdRun_3840x2160_50_8bit 50 26 

30 34 38 %%G 2160 

call Codec\MSS\mss_qp_hevc.cmd PeopleOnStreet_3840x2160_30_8bit 30 22 

27 30 33 %%G 2160 

call Codec\MSS\mss_qp_hevc.cmd Traffic_3840x2048_30_8bit 30 18 

22 26 30 %%G 2048 

call Codec\MSS\mss_qp_hevc.cmd Nebuta_Festival_2560x1600_60_8bit 60 30 

33 36 39 %%G 1600 

call Codec\MSS\mss_qp_hevc.cmd park_joy_1920x1080_50 50 26 

29 33 37 %%G 1080 

call Codec\MSS\mss_qp_hevc.cmd ducks_take_off_1920x1080_50 50 28 

31 35 37 %%G 1080 

call Codec\MSS\mss_qp_hevc.cmd crowdRun_1920x1080_50 50 26 

30 34 38 %%G 1080 

call Codec\MSS\mss_qp_hevc.cmd touchdown_pass_1920x1080_30 30 23 

26 30 34 %%G 1080 

call Codec\MSS\mss_qp_hevc.cmd BQTerrace_1920x1080_60 60 25 

27 31 34 %%G 1080 

call Codec\MSS\mss_qp_hevc.cmd ParkScene_1920x1080_24 24 23 

26 29 32 %%G 1080

call Codec\MSS\mss_qp_hevc.cmd park_joy_1280x720_50 50 26 

29 33 37 %%G 720 

call Codec\MSS\mss_qp_hevc.cmd ducks_take_off_1280x720_50 50 28 

31 35 37 %%G 720 

call Codec\MSS\mss_qp_hevc.cmd crowdRun_1280x720_50 50 26 

30 34 38 %%G 720 

call Codec\MSS\mss_qp_hevc.cmd City_1280x720_30 30 22 

24 27 29 %%G 720 

call Codec\MSS\mss_qp_hevc.cmd Crew_1280x720_30 30 22 

26 29 32 %%G 720 

call Codec\MSS\mss_qp_hevc.cmd Sailormen_1280x720_30 30 24 

26 28 30 %%G 720 

call Codec\MSS\mss_qp_hevc.cmd BasketballDrillText_832x480_50 50 21 

24 27 30 %%G 480 

call Codec\MSS\mss_qp_hevc.cmd PartyScene_832x480_50 50 24 

27 30 33 %%G 480 

call Codec\MSS\mss_qp_hevc.cmd RaceHorses_832x480_30 30 24 

27 30 33 %%G 480 

call Codec\MSS\mss_qp_hevc.cmd City_704x576_30 30 22 

24 26 28 %%G 576 

call Codec\MSS\mss_qp_hevc.cmd Crew_704x576_30 30 22 

25 28 31 %%G 576 

call Codec\MSS\mss_qp_hevc.cmd Soccer_704x576_30 30 22 

25 28 31 %%G 576 

) 

move summary.txt Output\MSS_Bitstreams 


mss_qp_hevc.cmd 

@echo off 

REM ENCODE_SCRIPT SEQUENCE QP TU RESOL FPS 

call Codec\MSS\mss_enc_params.cmd %1 %3 %7 %8 %2 




81 


mss_enc_params.cmd 


@echo off 

set SEQ=%1 

set QPI=%2 

set /a QPB=%QPI%+1 

set /a QPP=%QPI%+1 

set OP=%3 

set DIR=Content 

set ODIR=Output\MSS_Bitstreams 

set RES=%4 

set FRATE=%5 

call :ResolutionSetup %RES% W H 

set "startTime=%time%" 

Codec\MSS\win_x64\bin\sample_encode.exe h265 -sw -f %FRATE% -u %OP% -async 2 -cqp -qpi %QPI% - 

qpp %QPP% -qpb %QPB% -w %W% -h %H% -i %DIR%\%SEQ%.yuv - 

o %ODIR%\msdkHEVC_%SEQ%_%OP%_qp%QPI%.h265 

set "stopTime=%time%" 

call :timeDiff diff startTime stopTime 

Codec\MSS\win_x64\bin\sample_decode.exe h265 -i %ODIR%\msdkHEVC_%SEQ%_%OP%_qp%QPI%.h265 - 

o %ODIR%\msdkHEVC_%SEQ%_%OP%_qp%QPI%_dec.yuv -p 15dd936825ad475ea34e35f3f54217a6 

Codec\PSNR.exe %DIR%\%SEQ%.yuv %ODIR%\msdkHEVC_%SEQ%_%OP%_qp%QPI%_dec.yuv %W% %H% 

echo Encoding time: %diff% msec. >> summary.txt 

goto :eof 

82 

:timeDiff 

setlocal 

call :timeToMS time1 "%~2" 

call :timeToMS time2 "%~3" 

set /a diff=time2-time1

( 

ENDLOCAL 

set "%~1=%diff%" 

goto :eof 

) 

:timeToMS 

::### WARNING, enclose the time in " ", because it can contain comma separators 

SETLOCAL EnableDelayedExpansion 

FOR /F "tokens=1,2,3,4 delims=:,.^ " %%a IN ("!%~2!") DO ( 

set /a "ms=(((30%%a%%100)*60+7%%b)*60+3%%c-42300)*1000+(1%%d0 %% 1000)" 

) 

( 

ENDLOCAL 

set %~1=%ms% 

goto :eof 

) 

:ResolutionSetup 

( 

IF %~1 EQU 1600 ( 

set %~2=2560 

set %~3=1600 

) else ( 

IF %~1 EQU 2160 ( 

set %~2=3840 

set %~3=2160 

) else ( 

IF %~1 EQU 2048 ( 

set %~2=3840 

set %~3=2048 

) else ( 

IF %~1 EQU 1080 ( 

set %~2=1920 

set %~3=1080 

) else ( 

IF %~1 EQU 720 ( 

set %~2=1280 


83 



) 

) 

) 

) 

) 

) else ( 

) 

set %~3=720 

IF %~1 EQU 576 ( 

) else ( 

) 

set %~2=704 

set %~3=576 

IF %~1 EQU 480 ( 

) else ( 

) 

set %~2=832 

set %~3=480 

echo Unknown resolution. 

exit /b 

C4: Script for HEVC BD rate Computation 

We now provide example script (Visual Basic macro) for BD rate calculation provided by MPEG HEVC 

committee. This macro needs to be installed in an excel document, and upon entering the computed 

encoding Bitrates generated by HM14 and MSS HEVC Encoder’s, and PSNRs generated by comparing 

decoded yuv files with respect to original yuv files and entering them into the same excel document, BD 

rate difference between HM14 and MSS HEVC encoder can be computed. 

BDrate.bas 

84 

Attribute VB_Name = "Module1" 

Public Function pchipend(h1 As Double, h2 As Double, del1 As Double, del2 As Double) As Double 

Dim d As Double 

d = ((2 * h1 + h2) * del1 - h1 * del2) / (h1 + h2)

If (d * del1 < 0) Then 

d = 0 

ElseIf ((del1 * del2 < 0) And (Abs(d) > Abs(3 * del1))) Then 

d = 3 * del1 

End If 

pchipend = d 

End Function 

Public Function bdrint(rate As Range, dist As Range, low As Double, high As Double) As Double 

Dim log_rate(1 To 4) As Double 

Dim log_dist(1 To 4) As Double 

Dim i As Long 

For i = 1 To 4 

log_rate(i) = WorksheetFunction.Log(rate(5 - i), 10) 

log_dist(i) = dist(5 - i) 

Next i 

Dim H(1 To 3) As Double 

Dim delta(1 To 3) As Double 


H(i) = log_dist(i + 1) - log_dist(i) 

delta(i) = (log_rate(i + 1) - log_rate(i)) / H(i) 

Next i 

Dim d(1 To 4) As Double 

d(1) = pchipend(H(1), H(2), delta(1), delta(2)) 


d(i) = (3 * H(i - 1) + 3 * H(i)) / ((2 * H(i) + H(i - 1)) / delta(i - 1) + (H(i) + 2 * H(i - 1)) / delta(i)) 

Next i 


d(4) = pchipend(H(3), H(2), delta(3), delta(2)) 

Dim c(1 To 3) As Double 

Dim b(1 To 3) As Double 


c(i) = (3 * delta(i) - 2 * d(i) - d(i + 1)) / H(i) 

b(i) = (d(i) - 2 * delta(i) + d(i + 1)) / (H(i) * H(i)) 

85 


Next i 

' cubic function is rate(i) + s*(d(i) + s*(c(i) + s*(b(i))) where s = x - dist(i) 

' or rate(i) + s*d(i) + s*s*c(i) + s*s*s*b(i) 

' primitive is s*rate(i) + s*s*d(i)/2 + s*s*s*c(i)/3 + s*s*s*s*b(i)/4 


Dim s0 As Double 

Dim s1 As Double 

Dim result As Double 

result = 0 


s0 = log_dist(i) 

s1 = log_dist(i + 1) 

' clip s0 to valid range 

s0 = WorksheetFunction.Max(s0, low) 

s0 = WorksheetFunction.Min(s0, high) 

' clip s1 to valid range 

s1 = WorksheetFunction.Max(s1, low) 

s1 = WorksheetFunction.Min(s1, high) 

s0 = s0 - log_dist(i) 

s1 = s1 - log_dist(i) 

If (s1 > s0) Then 

result = result + (s1 - s0) * log_rate(i) 

result = result + (s1 * s1 - s0 * s0) * d(i) / 2 

result = result + (s1 * s1 * s1 - s0 * s0 * s0) * c(i) / 3 

result = result + (s1 * s1 * s1 * s1 - s0 * s0 * s0 * s0) * b(i) / 4 

End If 

Next i 

bdrint = result 

End Function 

86 

Public Function bdrate(rateA As Range, distA As Range, rateB As Range, distB As Range) As Double 

Dim minPSNR As Double 

Dim maxPSNR As Double

minPSNR = WorksheetFunction.Max(WorksheetFunction.Min(distA), WorksheetFunction.Min(distB)) 

maxPSNR = WorksheetFunction.Min(WorksheetFunction.Max(distA), WorksheetFunction.Max(distB)) 

Dim vA As Double 

Dim vB As Double 

vA = bdrint(rateA, distA, minPSNR, maxPSNR) 

vB = bdrint(rateB, distB, minPSNR, maxPSNR) 

Dim avg As Double 

avg = (vB - vA) / (maxPSNR - minPSNR) 

bdrate = WorksheetFunction.Power(10, avg) - 1 

End Function 


87

Deliver High Quality High Performance HEVC via Intel® Media Server Studio

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?