Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Multi-Hydra Computer Feasible in Future?

Author: Vincent Diepeveen

Date: 09:15:04 02/16/04

Go up one level in this thread


On February 15, 2004 at 14:34:08, Bob Durrett wrote:

>
>I envision a standard equipment rack with 32 or 64 Hydra cards, a power supply,
>possibly a conventional computer for orchestration, and fans.

You mean a SGI altix3000 node. That's a box from $500k with 16 bricks (4
processors a brick and each brick has 2 mainboards with interconnects) so in
total say 64 hydra cards. each card say $3000.

>Feasible?

Donninger has more interesting things to do than to get 64 hydra cards to work.

>If so, what performance could be expected?

Quite a bit worse than in paderborn, i am very sure that they will have major
problems to get to work 64 cards at a cluster. I doubt they ever will manage to
get to work more than 16 nodes (currently that would be 32 cards with 2 cards a
node).

>Note that it might not be necessary to combine the Hydras in the most efficient
>way possible.  Maybe not SMA.

It works at cluster currently with slow MPI calls transferring a lot of data
using broadcasts.

In short, they will not scale easily to more than 4 dual xeon nodes.

>Bob D.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.