Scientia (from AMDzone) posted a great post about some trends we have being seeing from benchmarks on conroe:
Read the whole thread here:
www.amdzone.com
...Thus far, I have been completely unable to find any SuperPi scores that show comparable 2MB C2D, 4MB C2D, and FX. Just attempting to judge from some high scores the advantage is at least 15%. One website ran two copies of SuperPi to evenly stress both cores. They used this to show that C2D used less power than K8. Stressing both cores is also a problem since running two copies of the same benchmark would allow C2D to have only a single copy in L2 and maximize its large cache. While the claim is made frequently and loudly that an inexpensive C2D will beat an FX, I have yet to see any review that ran a separate benchmark on each core simultaneously to stress the cores and cache. This would be the best test of real world use but I haven't seen it done. The procedure really isn't that difficult. You run a continuous application on one core and a second timed application on the second core. Then you compare the scores with a single instance and nothing running on the second core. Not difficult. Of course, C2D would take a hit because of cache dividing and another one if the second application accessed memory because of lower memory bandwidth and cache thrashing. The question is how much of a drop. Perhaps a cheat hasn't been found to prevent this yet.
Other tricks are common such as using 4-4-4 timing for AMD because C2D doesn't benefit from the faster timing. Other tricks include praising C2D's higher IPC and then making sure that the C2D chip is clocked higher. These cheats are a continuation from earlier when Yonah chips were overclocked via the FSB which made both memory and processor clock faster and these were compared with FX's which clocked the cpu higher but had lower proportionate memory increases...
Read the whole thread here:
www.amdzone.com