The IBM eServer BladeCenter JS20 - IBM Redbooks
The IBM eServer BladeCenter JS20 - IBM Redbooks
The IBM eServer BladeCenter JS20 - IBM Redbooks
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>The</strong> major features of the PowerPC 970 include:<br />
► Speculative, superscalar RISC processor core with vector extensions<br />
► On-chip level-1 and level-2 caches<br />
► Deeply pipelined design up to 25 stages long<br />
► Aggressive branch prediction<br />
► In order dispatch of up to five operations per cycle<br />
► Out of order issue of up to ten operations per cycle<br />
► In order completion of up to five operations per cycle<br />
► Register renaming<br />
► Up to 215 instructions in-flight<br />
► Fast selective flush of incorrect speculative instructions and results<br />
► Hardware prefetching of up to eight streams including four VMX streams<br />
We elaborate on some of these features in the following sections.<br />
Cache structure<br />
<strong>The</strong> PowerPC 970 includes several on-chip caches to reduce memory latency<br />
when fetching instructions and performing data load and store operations.<br />
Understanding the structure of these caches is sometimes useful when doing<br />
low-level assembly language programming.<br />
<strong>The</strong> on-chip caches include:<br />
► 64 KB, direct-mapped instruction cache<br />
► 32 KB, 2-way set associative data cache<br />
► 128-entry, instruction effective to real address translation (ERAT) cache<br />
instructions<br />
► 128-entry, data ERAT cache<br />
► 64-entry, fully associative segment lookaside buffer (SLB)<br />
► 1024-entry, 4-way set associative translation lookaside buffer (TLB)<br />
► 512 KB, 8-way set associative, level-2 cache<br />
Speculative superscalar features<br />
<strong>The</strong> design of the PowerPC 970 uses a variety of techniques to enable<br />
superscalar operation, where multiple instructions are executed during each<br />
processor clock cycle. This capability is enabled by the exploitation of multiple<br />
execution units within the processor core.<br />
44 <strong>The</strong> <strong>IBM</strong> Eserver <strong>BladeCenter</strong> <strong>JS20</strong>