Originally posted by: DrMrLordX
You assume too much. Even with a theoretical FSB bottleneck, Conroe still kills anything AMD has on the market. What makes me curious are the raw integer and floating-point performance numbers Conroe and Woodcrest have generated versus their real-world performance numbers. If you look at Anandtech's own Woodcrest vs Opteron benchmarks, you'll see that the performance difference in raw integer and fp is staggering. When it comes to "real-world" benchmarks and synthetics, the difference in performance is still decidedly in Conroe's favor, but not to the same extent.
That makes me wonder exactly what it is that prevents Conroe from fully utilizing its incredible computational power. Is it suboptimal code? Is it the cache? FSB or memory subsystem?
My guess is it's the FSB, but I don't know enough to be sure. That's why I'm curious about how well Conroe would perform against itself at the same clock speeds and memory clocks with different FSB speeds. It also makes me wonder whether overclockers will be best served by getting a chip with the highest available multiplier. If you max the chip out before reaching the board's maximum FSB, you may be losing some performance.