Figure 7 compares the Pentium 4 2.0GHz processors with 256K and 512K L2 cache. The 2.0GHz processors use the 850 chipset and Intel 6.0 compiler.
Figure 7: Pentium 4 2.0GHz processors with 256K and 512K L2 cache.
Figure 8 compares the Pentium 4 3.2GHz processors with 512K L2 cache and with 2M L3 cache on top of the 512K L2 cache. The 3.2GHz processors use the 875 chipset and 8.0 compiler.
Figure 8: Pentium 4 3.2GHz processors with 512K and 2M (L3).
Figure 9 shows the percentage performance gain on the Pentium 4 2.0GHz processor with 256K to 512K L2 cache. The Pentium 4 gains 11% from 256K to 512K cache compared to the 22% gained for the same cache size change on the Pentium III.
Figure 9: Pentium 4 2.0GHz performance gain from 256K to 512K L2 cache.
Figure 10 shows the percentage performance gain on Pentium 4 3.2GHz processor from 512K L2 cache to 2M L3 cache. The gzip application showed 3.7% performance gain from 256K to 512K cache, but nearly no gain to the 2M L3 cache. One possible reason for this is that the 512K cache was sufficient to bring most of the working set into cache, and further increases in size have not continued benefit. Note that the gzip application scales very well with frequency, which support this explanation. Another application, eon, has essentially zero performance gain at either cache size changes. This could indicate either that the application was already in cache or that the working set is too large for any of the cache sizes to have meaningful influences. Since eon scales well with frequency, the first explanation is more likely.
Figure 10: P4 3.2GHz performance gain from 512K to 2M(L3) cache.
Front-Side Bus and Memory Bandwidth
The Pentium 4 performance from the 400MHz bus to the 533MHz bus is somewhat more complicate to decipher. Figure 11 shows the Pentium 4 2.4GHz performance with the 400MHz and 533MHz FSB using the 5.0.1 compiler. Both systems, however, actually have the same memory bandwidth (3.2GB/sec) using PC-800 RDRAM memory. So the 533MHz FSB bandwidth is 4.3GB/sec while the memory bandwidth is 3.2GB/sec.
Figure 11: Pentium 4 2.4GHz with 400MHz and 533MHz FSB, 5.0.1 compiler.
Figure 12 shows the Pentium 4 2.4GHz performance where both systems have the 533MHz FSB, the first system use PC-800 RDRAM and the second system uses PC-1066 memory. Both systems use the 6.0 compiler, which has some performance improvements over version 5.0.1.
Figure 12: Pentium 4 2.4GHz/533MHz FSB, PC-1066 memory, 6.0 compiler.
Figure 13 shows the Pentium 2.4GHz performance with 533MHz and 800MHz FSB. Both systems use the 8.0 compiler. The 533MHz uses PC-1066 RDRAM memory and the 850 chipset, while the 800MHz system using DDR-400 memory and the 875 chipset.