psycho bob: Regurgitating what you read on the Internet doesn't show that you've learned anything. Second, for the love of god, do NOT go to TomsHardware to "learn about AMD". Tom's was notorious for skewing their benchmarks toward Intel. If you want to learn about AMD chips, go to reputable sites like The Tech Report, [H]ardOCP, Anandtech, or Xbit Labs.
You also keep talking about how latencies don't mean jack--until you actually see the difference first hand, you will never understand its significance. Why don't you go find someone with an A64, and have them show you the difference between 3-3-3-8 RAM and 2-2-2-5. Trd2 and cas2 equates to literally a 10-12Mhz increase in memory FSB (e.g I get the same memory bandwidth at 200Mhz with those latencies as I would get at 210-212 with the much looser settings). While memory bandwidth's impact is most significant in gaming, other applications like Divx/MPEG encoding could use it>> it means shaving off seconds, to minutes, to hours even. In PShop it could mean waiting 10-12 less seconds on applying a filter.
Regarding your misunderstanding of FSB, yes, while certain sites would claim that the FX-55 or the FX-series is running at 400Mhz (which means 400 * 8 = 3.2GB/s of memory bandwidth), that couldn't be farther from the truth.
http://www.hardocp.com/article.html?art=Njc1LDI=
B/c of having not only a dual-channel memory controller, but an integrated one, it achieves nearly 6GB/s of memory bandwidth. Running at "only 400Mhz", it far exceed Intel's offering on the 3.46EE, which has an "FSB" of 1066Mhz. Not on the other hand, when I run at "only 400Mhz" or 430Mhz to be exact, my results are much different than what an FX would get. B/c the Bartons do not have an IMC, Bartons are already inherently capped in memory performance and also in CPU performance. What some people forget is that the newer K8 core (Opteron/FX/Athlon64) chips are still essentially a modification and improvement of the Barton (K7) core. Reduced latencies have meant improving anywhere from 50% to 200% of the performance a Barton would have at the same clockspeed. I can tell you that if I had an IMC with my Barton, @ 2.55GHz, it would be in the range of singlehandedly equalling or exceeding dualG5 systems. However, the reality is that w/o that IMC, my Barton is closer to a single G5 (1.8-2.0) range.