Hi Jim - didn't mean to aim the benchmark mess at you particularly.
However: the guys on Anand's forums were jumping to conclusions off their memory, I think. Go back and look at the new thread there and you will see the same numbers I looked at, except they haven't loaded the Firing Squad Kryotech 800 data yet. On most numbers the Athlon 650 does NOT exceed the oc'd 700 CUMine, more to the point, if you try to adjust the numbers to represent 733/RDRAM/i820 vs. 700/100, I think you will come to the same conclusion I did.
It is a notable fact that HardOCP's numbers for the 800 CUMine seem about linearly better than the PIII 500 he had, however, as one of the Anand boys also notes, processor performance does not usually scale linearly over that kind of range, so the 800 CUm is likely somewhat faster than an 800 PIII would be.
More to the point, we don't have to try to extrapolate up and back that far. The Kryotech 800 Athlon system is, for performance purposes, pretty much what an Athlon 800 wi. 100 MHz mem. would be if AMD were shipping it today on a .25um process. And the overclocked 800 CUm on 133 MHz mem. is pretty much like what a real 800/133 CUm would be, except that the bus overclocking would speed up other components that might have an effect on the benchs. That is why I think the synthetic CPU-intensive benchs are more meaningful for this system than, for instance, games. Elmer and others please note, most of these benchs are already using SSE or 3DNow. With all those caveats, if you compare those two systems, you see that the CUm is VERY competitive on everything except the SiS Drystone Int test. And if you then look at the 810/830/850 oc'd Athlons, you see considerable improvements in the same kinds of benchs, even at only 810 vs. 800. Thus I think the 125 mem speed is proving very helpful vs the 100 of the Kryotech system.
Finally, other tests of the i820/RDRAM vs bx/133 have shown that the i820 can boost perf several %. Put that together with what can be extrapolated from the oc'd 800 CUm, and I stand by my original assertion.
Needless to say, I would really like to see controlled, well-understood, head-to-head comparisons. But this is interesting info, nonetheless.
Vince |
