AMD Radeon R9 290X and Radeon R9 290 AMD’s new Hawaii GPU makes an appropriately volcanic debut HAWAII Powerful GPU; competitive pricing; great 4K performance SUNBURN Runs very hot; high power consumption; loud stock cooler HOW MUCH? R9 290X Price £428 inc VAT Supplier www.scan.co.uk Manufacturer www.amd.com IN DETAIL R9 290X Graphics processor AMD Radeon R9 290X, 1GHz Pipeline 2,816 stream processors, 64 ROPs Memory 4GB GDDR5, 5GHz effective Bandwidth 320GB/sec Compatibility DirectX 11.2, OpenGL 4.3, AMD Mantle Outputs/inputs 2 x DVI-D (Dual-Link), 1 x DisplayPort, 1 x HDMI Power connections 1 x 6-pin, 1 x 8-pin, top-mounted Size 278mm long, dual-slot iven that AMD’s R-series debut last month G was effectively a 7000-series rebrand, many were left a little disappointed. However, now AMD has finally unveiled two graphics cards based on its new Hawaii GPU: the Radeon R9 290X 4GB and the Radeon R9 2904GB. When it first launched, the R9 290X cost £50 less than Nvidia’s GTX 780, but the latter’s price has since dropped from £500 to £400, indicating how the R9 290X performs. This makes room for Nvidia’s GTX 780 Ti, which is intended to supersede the GTX 780 in the green team’s product stack in the near future. The R9 290X now costs £428, but AMD was unable to provide a UK retail price for the R9 290 at the time of writing, so we won’t be able to give it a final score yet. The new GPU features AMD’s Graphics Core Next architecture, so both cards are fully compatible with AMD’s upcoming Mantle API. Similarly, like the R7 260X, the Hawaii GPU has a separate non-graphics piece of silicon called TrueAudio. This is a programmable audio pipeline, designed to offload audio processing from the CPU for an enhanced and consistent audio experience, regardless of the system used. However, like Mantle, it’s yet to be supported by a current game, and requires developer implementation to function. Hawaii is built on a 28nm process and contains 6.2 billion transistors. It measures 438mm² and its die is thus 24 per cent larger than the R9 280X’s Tahiti GPU. Workloads are now divided between four Shader Engines (SEs). Each SE has a geometry processor, doubling the number found in Tahiti. The geometry processors contain a geometry assembler, vertex assembler and tessellator, and are fed by the graphics command processor, as well as eight asynchronous compute engines (ACEs), all of which can operate simultaneously. This is a fourfold increase in the ACE-count of the R9 280X, so there are potential benefits for compute performance too. Hawaii’s main processing power comes from AMD’s Compute Units (CUs), which are unchanged. The R9 290X has 11 CUs per SE for a total of 44 – a 37.5 per cent increase over the R9 280X. The R9 290, meanwhile, has a slightly reduced count of 40. Each CU is made up of four SIMD engines of 16 stream processors each, bringing the total counts for the 290X and 290 to 2,816 and 2,560 respectively. A CU also houses four texture units, so the new cards have 176 and 160 of these respectively. The only other area where the cards differ is in core clock speeds, which are 1GHz for the top model, and 947MHz for the 290. A rasteriser in every SE means Hawaii doubles Tahiti’s rendering specification here too, and this is complemented by a doubling of the GPU’s render back-end units. Each of these houses four ROPs, so both cards feature 64 ROPs in total. These allow the R9 290X to theoretically pump out up to 64Gpixels/sec. AMD is pushing 4K and multi-monitor setups hard with Hawaii’s launch, and while the pixel fill rate is only one piece of the pie at such high resolutions, it should still prove beneficial. HAWAII IS BUILT ON A 28NM PROCESS AND CONTAINS 6.2 BILLION TRANSISTORS Hawaii’s memory interface has also been bolstered to 512-bit, as it now has eight 64-bit memory controllers compared to the previous six. There’s also an extra 1GB of GDDR5 (4GB total) memory, which will again be beneficial at high resolutions and detail settings. Effective memory clock speeds are down to 5GHz from Tahiti’s 6GHz, however, which is due to a shrinking of the die area that the memory interface occupies. Nevertheless, total memory bandwidth is still up by 11 per cent at 320GB/sec. Another hardware change relates to CrossFire, which no longer requires an external bridge. Instead, cards communicate solely via PCI-E, thanks to a new CrossFire XDMA engine in the compositing block. AMD claims that no performance has been lost in the transition. ZeroCore, which all but turns off the GPU to save power when the system enters its long-idle state, returns for the 290-series as well. 20 January 2014
GRAPHICS CARDS Also, AMD’s PowerTune technology has seen some upgrades, with the company’s own second-generation Serial VID (SVI2) VR controller now at the heart of the cards’ power management systems. It can access cess 255 different voltage steps that it can select approximately every ten microseconds, based on load demands and other data. Under load, it will always aim to go up a step on the voltage/clock scale to boost performance, provided it stays within the card’s thermal, power and fan speed limits. As before, increasing the card’s power limit tells the controller it can draw more power if necessary. However, through the Catalyst Control Center (CCC), you can now set your card’s target temperature. Rather than having a basic fan curve, the card dwill actively aim for the set temperature by running its fans as slowly as possible to maintain it. By default, the target temperature is 95°C for both cards, the highest permitted, which is very hot. AMD promises that the GPU is safe to run at this temperature for its entire life, but that doesn’t alleviate our concerns about its potential effect on nearby components. You also set a maximum fan speed here, and if the card reaches this speed, but is still set to go over its target temperature, it will then reduce clock speeds. PowerTune is thus a dynamic trade-off between power, noise and thermals. The new CCC makes this fairly clear, but its user interface is annoying to use, particularly as clock speeds are now adjusted in percentages, rather than absolute megahertz values. Both cards also boast a dual BIOS switch, although only on the R9 290X is it loaded with different profiles; Über Mode and Quiet Mode. The only setting affected, however, is the maximum fan speed – 55 per cent and40 per cent respectively (for the R9 290, it’s 47 per cent regardless). Either way, it’s largely a pointless feature,as you can’t override the default profile settings, and changing the maximum fan speed is much quicker through software. The card itself is a standard AMD affair with a thin red and black plastic cover and radial fan, which blows air through the aluminium heatsink fins directly out of the rear I/O. This heatsink sits directly on top of the massive copper base plate, which draws heat away from the GPU as well as the memory and power circuitry via a black metal contact plate. It’s powered by a 6-pin and 8-pin PCI-E socket combo, and the juice is routed through a typical 5+1 phase power design. PERFORMANCE In all three games tested, and at every resolution, the R9 290X has a marginal but repeatable performance advantage over Nvidia’s GTX 780. In the non-4K tests, the two cards are very closely matched, however, as the R9 290X only twice exceeds the GTX 780’s minimum frame rate by more than 2fps. The GTX 780 does have a 7 per cent lead in Unigine Valley, but given IN DETAIL R9 290 Graphics processor AMD Radeon R9 290X, 947MHz Pipeline 2,560 stream processors, 64 ROPs Memory 4GB GDDR5, 5GHz effective Bandwidth 320GB/sec Compatibility DirectX 11.2, OpenGL 4.3, AMD Mantle Outputs/inputs 2 x DVI-D (Dual-Link), 1 x DisplayPort, 1 x HDMI Power connections 1 x 6-pin, 1 x 8-pin, topmounted Size 278mm long, dual-slot DIRECTX 11.2 AMD’s new GPUs support DirectX 11.2, exclusive to Windows 8.1 and the Xbox One. Its most relevant feature for enthusiasts comes from the updated Direct3D API. It’s called Tiled Resources, and it’s designed to reduce the GPU memory footprint of high-resolution textures in games. It doesn’t require any new hardware capabilities, so existing DirectX 11 cards, including AMD’s HD 7000-series and Nvidia’s Kepler and Fermi GPUs, will all support the key gaming features of DirectX 11.2 too. High-resolution textures look great, but they also take up lots of video memory. This memory can also be wasted if you don’t look at the whole texture at once, as only a portion of it actually needs to be rendered in such detail. Tiled Resources attempts to address this problem by allowing textures to be split into tiles on mipmaps of various detail levels. Through the use of a programmable hardware page table on the GPU memory, it only loads the tiles appropriate to the level of detail the player can see at any one time, dynamically swapping tiles in and out of the memory. It can also be applied to effects such as shadows. The end result is high-resolution resources using less memory without sacrificing detail, and with less memory overhead than software-based tiling approaches. For example, Microsoft demoed it using just a 16MB tile pool but working with 3GB of textures in total. While this is impressive, it’s unlikely to be a compelling reason for existing Windows 7 users to make the jump (Windows 8.1 is free for Windows 8 users), though, as it will require developer time and effort for it to become a real performance advantage. 22 January 2014 21
Gigabyte is the one of the renowned company which manufactures multiple products of computer such as Graphic Cards, Mother Boards, Mouse, Keyboard, Speaker, USB Charger, Tablet, Laptops, Accessories and More. Here we provide you Gigabyte Customer Care Number for those customers who are using Gigabyte products.