AMD Opteron vs. Intel Xeon in Linux

From: Uncle Al (UncleAl0_at_hate.spam.net)
Date: 10/13/03 

Date: Mon, 13 Oct 2003 14:06:00 -0700

AMD Opteron vs. Intel Xeon in Linux

We just finished crunching a 3.3 quadrillion atom calculation as
detailed in the "Need for Severe Speed" thread. A 2.8 GHz Intel Xeon
was loaded with the 32-bit code. 18 hours later a 1.8 GHz AMD Opteron
was loaded with the identical program and parameters, both under
Linux. Based on 2 GHz AMD consumer PC benchmarks for the same
calculation, we expected

   1) 8 CPU-weeks in the PC (forget it)
   2) 5.7 CPU weeks in the Xeon (maybe 3-4 given "hyperthreading")
   3) (8)(2/1.8)(1/2) = 4.4 weeks in the Opteron given double
bandwdith for the 64 bit-chip running a 32 bit program.

It looked like a good horserace. As of this morning, less than one
week after starting, the Opteron had output the results which graphed
and had stats exactly as expected. The Xeon is still cooking. AMD is
correct: a well-designed CPU can outperform a much faster Intel chip.
(Windows execution is 40% slower than Linux in benchmarks.)

It looks like a 64-bit optimized port of the 32-bit program in four
Opteron CPU-weeks can deliver the same crunch as a 168-Xeon cluster
over a 3.5 day holiday weekend. That is incredible - especially if it
runs in less time. We're gonna try the Opteron.

No more "Intel Inside" for Uncle Al. 

--
Uncle Al
http://www.mazepath.com/uncleal/eotvos.htm
 (Do something naughty to physics)

Relevant Pages