[Beowulf] AMD Opteron Linpack Benchmark

[Greg Lindahl]

On Fri, Sep 22, 2006 at 02:55:28PM +0800, Lai Dragonfly wrote:

> maybe i need to try Pathscale or PGI compiler.

While we love it when people try our compiler, it doesn't make any
difference on HPL when you're using a good library like Goto or ACML.

I think AMD's "Emerald" cluster with dual-core Opterons and InfiniPath
managed to get 83%. Of course the interconnect is a rather weak effect
with HPL, the bigger effect is using all the memory you've got.

-- greg