I think that the AMD will do well. So far it does not look good as a workstation but it is still way to early in the game. P4 looked like a piece of junk back when it first hit the street. It got the first couple of rounds under its belt and came out swinging with higher clock speeds, a better FSB, and tweaked out toys.
The benches we see today are only the start of it all for the Hammer. Take STREAM 5.1 for example. STREAM was never written to handle extra bandwidth through the X-Bar and Hypertransport. Notice how it makes even the dual Xeon systems look bad compared to a single P4 3.0 GHz. Hammer systems can use info from all of the other DIMMs in the entire system. If it looks only to the DIMMs attached to the immediate MCH then of course you are going to see poor results on the memory benches. The bandwidth is much higher than what was shown. HT would allow for an additional 19.2 GB/s on top of the memory in the DIMMs under the immediate MCH. Take a request from CPU1 that draws from the memory as a whole. It can pull info from the DIMMs attached to CPU2. Limit it as much as you want. Even if there were two CPUs with two channels a piece.
DDR333 = 500/3 or 166.667 MHz ~ 2666.667 MB/s or 2.67 GB/s
Take that and multiply it by 4 (two sets of dual channels) = 10666.667 MB/s or 10.67 GB/s
The benches returned only 1.7 GB/s.
Part of that could be because of the motherboard. There is no way to tell how the memory buses were allocated on the motherboard. Maybe it only used two channels for both CPU's? That would be a possible solution, in theory, since you could have CPU1's MCH run the DIMMs and CPU2's be disabled. "That is a SPECULATION."
The bandwidth is much better than 1.7GB/s and when a single P4 got a score of over 700 MB/s better there has to be a problem with the benchmark. Just to show haw bad some of these benches are look at the <A HREF="http://www.tomshardware.com/cpu/20030422/opteron-27.html" target="_new">Sisoft Sandra 2003 Memory bench</A>. The single Xeon did better than a dual Xeon.
I think one reason there is some improvement in the scores of the Dual Opteron verse the single XP 3000+ is the simplicity of incorporating the MCH into the die. No FSB bottleneck between the MCH and the CPU. Both are operating at the same clock frequency.
There are a lot of things that need to change in the world of benchmarks. There has never been a system like this before and now that it is here there needs to be a way to rate the two against each other.
There are my two cents.
<font color=blue><b>
Why is it a penny for your thoughts but you have to put your two cents in? Somebody is making a penny. - Steven Wright
</b></font color=blue>
<A HREF="http://www.millionmanlan.com/MMLDefault.asp" target="_new">Million Man LAN 2 is June 25-29, 2003 in Louisville Kentucky... Be there!</A>