Re: [computer-go] SlugGo approach: GNU vs.Goliath

[David G Doshay <ddoshay@xxxxxxx>]

On 6, Jan 2005, at 4:38 AM, Vincent Diepeveen wrote:

> You guys are making jokes i hope about using floating point in 
> evaluation.
Nope, not joking at all. SlugGo is quite consistent in its use of 
floats.


On 6, Jan 2005, at 7:32 AM, Vincent Diepeveen wrote:

> At 16:11 6-1-2005 +0100, Marco Scheurer wrote:
> > > > It is relevant. It's 2 times slower at least.
> > > Depends on the platform, IMHO. for platforms with separate
> > > floating point boxes ('units') such as the alpha (maybe newer 
> > > pentiums
> > > ?)
> > Apple's G4 and G5 have a really nice vector unit.
> What's the price of a quad g5?
each of the dual G5 cluster nodes cost $2,000.

While you may rise back in horror, for what SlugGo does using the 
standard
GigE between the boxes is fine, and results in the same performance as 
if
all 4 cpus were in the same box. Measured network overhead is 0.00002%
of compute time.

Our cluster has 36 boxes, thus 72 cpus. So far we have never used more
than 18 cpus for SlugGo, and we used 10 for everything we have reported
so far.

Cheers,
David


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/