From: Ben Goertzel (firstname.lastname@example.org)
Date: Tue Nov 09 2004 - 17:56:24 MST
As James suggests, for Novamente, shared memory is key. Thus we've opted
for Penguin's AMD Opteron based multiprocessor solution.
If we had more $$ for hardware, I'm impressed with Octiga Bay's
architecture, but I haven't studied it that closely.
-- Ben G
> -----Original Message-----
> From: email@example.com [mailto:firstname.lastname@example.org]On Behalf Of J. Andrew
> Sent: Tuesday, November 09, 2004 6:35 PM
> To: email@example.com
> Subject: Re: AI Hardware improving
> BillK wrote:
> > There are other startups doing similar projects, but Orion seem to
> > actually have something available now. So for quite reasonable money,
> > you can now give each programmer their own personal Linux cluster.
> There is wide consensus that the Orion and similar is a solution looking
> for a problem. The only way it makes sense is if your priorities are
> low power consumption above all else, and in practice hardly anyone
> interested in supercomputing has that as a priority. You can buy
> Opteron SMP systems for the same money with similar computing power and
> much broader applicability since that will also give you ultra-fast
> shared memory. And if you buy low-power Opterons (e.g. the 30W
> ultra-low power versions), you end up in the ballpark power-wise as
> well, though for a little more dough.
> Orion is mostly about buzzword compliance and the neat trick of packing
> gobs of modest processors in a tiny space. It is not a spectacularly
> powerful system for most real-world applications, nor is it a good
> solution to most supercomputing problems. Style over substance.
> j. andrew rogers
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:50 MDT