top500 list (was: opteron VS Itanium 2)

Craig Tierney ctierney at hpti.com
Mon Nov 17 10:06:40 EST 2003


On Mon, 2003-11-17 at 07:33, Mark Hahn wrote:
> > Sorry Greg, top500 list came out and you lost!
> 
> sigh.  Apple bought a benchmark.  does this make Apple products better?

Dell bought a benchmark at NCSA (See Tungsten
at #4).  Does this mean that Dell knows clusters?

I would be more than happy to have a vendor
decide to let me have their stuff at significant
discount so they can have a good benchmark.

> 
> > BigMac G5 (aka "X")         (2200 CPUs) - 10.3 TFlops
> 
> OK, so we already know that their pricing disclosures are ah,
> "optimistic".  (nodes probably cost >6k apiece, not half that)
> and their yield is around 59%.  not to mention the little bit 
> of missing ECC...

I am really curious to know why their are only
getting 59% efficiency when Quadrics system with
similar node counts is above 70%.  We know it
isn't an issue with the node speed, because 
the X cluster ran at 80% efficiency with 128 nodes.

Whats going on with the Infiniband?????

> 
> actually, I'd love to know what their yield is if they had only
> 4GB per node - almost certainly lower.
> 
> > Lightning (LANL Opteron)    (2816 CPUs) -  8.1 TFlops
> 
> yield is 72%.

How much was this system?  You sure Linux Networx
didn't 'buy' the business?  It seems to be happening
alot these days with large systems.

Craig


> 
> > Apple BigMac is number 3, while the Opteron cluster is
> > number 6.
> 
> for the billionth time: rmax is just a matter of how much money
> you have.  rmax/rpeak is the only part of top500 that matters.
> 

Yes but besides to politicians and the people funding
the systems, how relevant is the rmax/rpeak ratio?

Craig

> I'd like very much to know the actual prices and discounts for Big Mac.
> it's a shame this isn't required for Top500...
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

_______________________________________________
Beowulf mailing list, Beowulf at beowulf.org
To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list