[Beowulf] I wanna know the performance difference between Intel Nocona and AMD Opteron

Mark Hahn hahn at physics.mcmaster.ca
Wed Apr 6 08:14:18 PDT 2005


> hardware platform to choose. One is Intel nocona , the other is AMD
> Opteron.
> 
> Can anybody send me some articles related to :
> 
> 1. performance difference between those two platforms on 32-bit
> computing;
> 2. performance difference between those two platforms on 64-bit
> computing;

this question doesn't make sense.  the performance differences between
P4 and K8 architectures have less to do with which mode they're in 
than Intel and AMD's very different approaches.

Intel uses a shared-FSB, so processors share/compete for memory bandwidth.
AMD attaches memory directly to CPUs, and provides a fast, point-to-point 
link between CPUs (no bus, really).  so AMD is slightly NUMA, for instance,
though even remote memory is very fast and AMD unambiguously spanks Intel
in that domain (especially for more CPUs.)  of course, if your code is 
in-cache, this is irrelevant.

there are some other differences, such as details of cache size/latency/etc,
pipeline length, etc.  AMD spanks Intel for traditional (x87) FP performance,
but Intel seems to usually be slightly ahead on MMX/SSE/etc stuff.  most of 
these factors seem minor enough that they could easily be epiphenomena
(of the fact that Intel agressively tunes their own compiler, for instance.)

I find that AMD and Intel hold themselves fairly close in performance,
if you normalize by price (and ignore outliers).  so much so that power 
dissipation becomes quite relevant in choosing...

regards, mark hahn.





More information about the Beowulf mailing list