Anyone have information on latest LSU beowulf?

[Patrick Geoffray]

On Wed, 2002-10-09 at 12:53, Daniel Kidger wrote:

> So if in ATLAS's SUMMARY.LOG, if I get
> <cut> 
>       The best matmul kernel was ATL_dmm_sse2_80.c, written by Peter
> Soendergaard
>       This gave performance of 3623.65MFLOPS (200.9227777752460f apparent
> peak)
>       mmNN   : ma=0, lat=3, nb=48, mu=4, nu=1 ku=48, ff=1, if=5, nf=1
> <cut>
> 
> does this mean that I have a 48x48 or an 80x80 dgemm kernel ?

I would read 48.

> (For this node using xhpl+atlas, 80 and 160 give better performance than 48
> or 96)

I don't know why it would behave that way. Maybe another resource is the
effective contention. Have you tried the "tuning" part of the Atlas web
site, to find the cache edge for example ?

This rule of thumb to take a multiple of ATLAS dgemm block size to use
in HPL was given to me by Clint when he was working on Atlas at UTK.
Maybe something changed with the new kernels in Atlas, but I doubt it. 

For me, 160 (4x40) was always the best choice on PIII. Never tried on P4
(SSE2).

Patrick

-- 
----------------------------------------------------------
|   Patrick Geoffray, Ph.D.      patrick at myri.com 
|   Myricom, Inc.                http://www.myri.com
|   Cell:  865-389-8852          685 Emory Valley Rd (B)
|   Phone: 626-821-5555          Oak Ridge, TN 37830
----------------------------------------------------------