Dear Tench:
It is the later. AMD properly spends its resources on delivering high quality CPUs not buggy ones (at least 40 errata including MOVSB(What?)). When IFC 4.5 only seems to produce runnable code for SPECmark runs, I suspect the IFC compiler series is for highly restricted environments (SPEC only). P3, and for that matter P4, do not do as well with other compilers that have a much bigger environment. When these other compilers get SSE2 extensions, can we really tell how much of that is reproducable in real world applications. Those benchmarks that are highly optimized for P3 and Athlon, P3 gets beat quite easily. P4 does do much better with some of those benchmarks but, few are truly P4 optimal. The problem for Intel is that very few P4 optimized apps are available and it does not do well with ones optimized for P3 in many cases. Finding benchmarks that both of us can agree on that are highly optimized for P4, P3, and Athlon is very, very small. Although, with the number of Athlons beginning to rise real fast now, this may not be true much longer.
One thing is clear, when the dataset is less than 380K, Athlon seems to outrun P4, above 500K to about 1GB, P4 can beat it in some applications (those with a low flops per data byte ratio), and between 1GB and 4GB, Athlon would be many times quicker (hard drives are so much slower than RAM). This is better than with P3 as Athlon is much quicker across the whole data set size.
Pete
Well it will take some time for those to show up. |
