Figure 13: Pentium 4 2.4GHz with 533MHz and 800MHz FSB, 8.0 compiler.
Figure 14 shows the Pentium 4 percentage performance gain from 400MHz to 533MHz FSB, but with 3.2GB/sec memory bandwidth on both systems. It is interesting to observe that increasing the FSB bandwidth but not the memory bandwidth still increases performance in the Pentium 4 system by 4% overall.
Figure 14: Pentium 4 performance gain from 400 to 533MHz FSB with PC-800.
Figure 15 shows the percentage performance gain in by increasing the memory bandwidth from 3.2 to 4.3GB/sec, where both systems have a 533MHz FSB, with a overall gain of 5.0%. It does not seem unreasonable to conclude that the performance gain in the Pentium 4 systems realized by increasing both FSB and memory bandwidth from 3.2GB/sec to 4.3GB/sec is 9%.
Figure 15: Pentium 4 2.4GHz/533MHz FSB from PC-800 to PC-1066.
Figure 16 shows a 6.8% relative performance gain on the increase in memory bandwidth from 4.3 to 6.4GB/sec. Some applications show nearly zero performance gain on memory bandwidth. The mcf application scales best with bandwidth, but only poorly with processor frequency. This is consistent with an application that has a working set larger than cache and has predictable large block memory accesses. An application with either unpredictable memory access patterns or small block accesses may not benefit much from memory bandwidth if the round-trip latency for memory accesses is more limiting.
Figure 16: Pentium 4 2.4GHz from 533MHz to 800MHz FSB.
Figure 17 compares the Pentium 4 3.0GHz/512K, Athlon 2.2GHz/512K and Opteron 2.0GHz/1M processors. The Opteron processor has a larger cache (1M) than the Athlon (512K) an integrated memory controller, which reduces memory access latency by 20-25ns, and a wider path to memory (128-bit versus 64-bit). The Opteron processor also adds 64-bit addressing capability, but this feature by itself is not a directly related to performance. A reasonable conclusion between the Athlon and Opteron performance results is that there is no significant performance improvement in processor core. The areas where Opteron show performance improvements over the Athlon are applications known to benefit from either cache size or memory bandwidth.
Figure 17: Pentium 4, Athlon and Opteron processors.
Figure 18, below, shows the performance of recent high-end microprocessors. The Pentium 4 is still the reigning champion. So much for the predictions of the demise of the X86 architecture. The one application where Itanium2 has an edge on Pentium 4 is the mcf application, which is known to benefit from cache size and memory bandwidth, areas the Itanium2 is very well equipped.
Figure 18: Recent high-end microprocessor performance.