rettihSlluB
Distinguished
This topic has been talked in the amdzone forums and this is a good explanation about the whole situation involving vectorized code (call it SSE):wouldnt be too quick to jump to conclusions. sisoft doesnt use x87 code afaik, which is where K8 reigns supreme, instead it uses either integer or SSE. There are quite a few workloads, especially SSE code, that even with proper optimization will fun faster on a high clocked P4 than on a K8. Now how representative mandlbrot creation is for anything else than well.. creating madelbrots is another issue all together.
If I'm not wrong, anandtech, arstechnica and aceshardware posted a review that back up the above quote about Opteron/Athlon 64 performance in floating point calculations using legacy x87 code AND vectorized code.The issue of SSE code gets very confusing when trying to benchmark the performance because most of the time you will have to compare two different benches.
Intel very thoughtfully included a check in its compiler that prevents it from creating SSE code on AMD processors (see the link in my previous post). So, the benchmark for Intel will be created using the Intel compiler and will include SSE. In contrast, the benchmark for AMD will either be created without SSE support or will be created with a different compiler. This greatly reduces the validity of the benchmark.
In tests that I have seen where Intel and AMD were compiled on Linux systems with the same compiler the Intel processor was a bit slower. In tests that I have seen where the Intel compiler was hacked to fool it into generating SSE code for AMD the Intel processor was still a bit slower. It has been claimed more than once that Intel Compiler SSE code runs faster on AMD than on Intel.
It has been shown again and again that Intel maintains an advantage in graphics and mp3 coding benchmarks only when good code is generated for Intel and poor code is generated for AMD. Whenever the AMD code was improved it has always beaten the Intel code.
As far as I can tell when the benches are not artificially skewed by comparing SSE for Intel with non-SSE for AMD or by artificially tuning the code for a larger cache, Intel will win only about 10% of the benches.
The funniest thing of all, is that Athlons benefit more from the Intel compiler than intel's own offerings. 😉
My Beloved Rig:
ATHLON 64 FX 55
2X1024 CORSAIR XMX XPERT MODULES
MSI K8N DIAMOND (SLI)
2 MSI 6800 ULTRA (SLI MODE)
OCZ POWERSTREAM 600W PSU