INTEL XEON vs AMD ATHLON

[Mark Hahn]

> that Athlon MPX goes head by head with Xeon in terms of CPU speed for
> heavy numerics, but it is considerably less expensive.

that is fact.  they're radically different designs, with the P4
being relatively dumb, but clock-scalable.  the Athlon is much faster
than a clock-equating comparison would suggest.  that is, AMD's
athMP/2000 does indeed compare favorably versus a Xeon 2.0,
both in real performance and price.  not by a huge margin, but then
again, AMD and Intel stay a month or two out of phase in repricing
(and AMD is due.)

> particular, Xeon is only 5-10% more expensive, but it is more powerful

pricewatch says athmp/2.0 is 209 and xeon/2.0 is 251 - that's 20%.
depending on what kind of system configuration you're looking at,
that's simply noise compared to the MB (e7500's seem to start at 
just under $500, vs mpx boards at $200.)

> since it is currently running at 2.4 GHz, has 400 MHz front bus plus 512
> Kb of secondary cash as compared to 1.76 GHz, 266 MHz and 384 Kb for
> Athlon.

2.4 vs 1.76 is simply a bogus comparison.  the P4 family does definitely
provide a faster FSB, but in some sense, it needs it more.  in the end,
the systems are quite competitive.

> Also, the newest linux kernel will provide multithreading on Xeons
> that could also result in substantial boost for numerical calculations. 

don't get your hopes up.  HT wins when the system has large amounts
of idle resources that can profitably be interleaved.  that's normally
something that the programmer (and chip designers!) try to avoid, 
since it's waste.  since Prestonia doesn't have more resources,
most code goes half as fast when running HT, for no net gain.  it's 
entirely possible to have a massively "imbalanced" program (pointer
chasing, perhaps) that might show a noticable speedup, but I haven't
seen anything like that with "normal" codes.

> say: Is it really true that the new Xeon is superior than Athlon as far as
> heavy numerical calculations in cluster environment are concerned?

"heavy numeric" is ill-defined, of course.  lots of random fetches
from dram?  cache-friendly?  predictable branches?  prefetchable?

in short, AMD remains competitive, modulo month-by-month variance
in product cycles.  they have a serious advantage in chip die area,
but are clearly stretching the K7 until the K8 kicks in - 
but the much rumored jump to 512K cache might provide a worthwhile 
late-race kicker...

regards, mark hahn.