IBM 88554RU Installation Guide - Page 25
Explicitly Parallel Instruction Computing EPIC, Level 2 cache, 256 KB
View all IBM 88554RU manuals
Add to My Manuals
Save this manual to your list of manuals |
Page 25 highlights
This is new and the "closest" to the processor, and is used to store micro-operations. These are decoded executable machine instructions. It serves these to the processor at rated speed. This additional level of cache saves decode time on cache hits. Level 2 cache, 256 KB This is equivalent to L1 cache on the Pentium® III Xeon. Level 3 cache 3-6 MB This is equivalent to L2 cache on the Pentium III Xeon or the L3 cache on the Pentium Xeon MP processor. Unlike the design of the original Itanium processor, this L3 cache is now on the processor die, greatly improving performance, up to 2 times greater than that of the original Itanium. The x455 also implements a Level 4 cache as described in "IBM XceL4 Accelerator Cache" on page 12. Intel has also introduced a number of features associated with its Itanium micro-architecture. These are available in the x455, including: 400 MHz frontside bus The Pentium III Xeon processor had a 100 MHz frontside bus that equated to a burst throughput of 800 MBps. With protocols such as TCP/IP, this had been shown to be a bottleneck in high-throughput situations. The Itanium 2 processor improves on this by using a single 200 MHz clock but using both edges of each clock cycle to transmit data. This is shown in Figure 1-6. 200 MHz clock Figure 1-6 Dual-pumped frontside bus This increases the performance of the frontside bus. The end result is an effective burst throughput of 6.4 GBps (128-bit wide data path running at 400 MHz), which can have a substantial impact, especially on TCP/IP-based LAN traffic. This is opposed to the Itanium processor, which had a burst throughput of only 2.1 GBps (64-bit wide data path running at 266 MHz). Explicitly Parallel Instruction Computing (EPIC) EPIC technology, developed by Intel and HP, leads to more efficient, faster processors because it eliminates numerous processing inefficiencies in current processors and attacks the perennial data bottleneck problems by increasing parallelism, rather than simply boosting the raw "clock" speed of the processor. Chapter 1. Technical description 11