Milo, Re: "NW isn't beating Athlon with greater Memory Bandwidth. It has the same Bandwidth (AKA throughput) it's always had. Is your head in the clouds?"
You misunderstood what I said. You said that the Athlon was kicking Pentium 4 butt, while only having PC2100 memory, as if that's supposed to de-emphasize the need for memory bandwidth. I responded that it is the Pentium 4 who is outperforming the Athlon, and the reason it does is only partially due to having greater memory bandwidth.
You are bringing up L2 cache in order to establish your point that latency is better than bandwidth. In that case, you are talking about overall average latency due to data locality, which is slightly off-topic from our previous discussion, which was based strictly on memory latency. Adding more cache will improve the performance of a processor overall, but it does nothing for the latency in the memory itself. AMD hopes to tackle that problem with an integrated memory controller, and I do believe it will buy them a significant one-time benefit. However, I also believe there are inherent trade offs with this approach, and what AMD makes up for in their one time performance gain, they will lose out in other areas.
wbmw |
