www.sicortex.com
|
From: Nick Maclaren on 2 Apr 2008 13:25 It looks interesting, but will it succeed? Any serious or even humorous comments on it appreciated. Regards, Nick Maclaren.
From: Terje Mathisen on 2 Apr 2008 16:33 Nick Maclaren wrote: > It looks interesting, but will it succeed? Any serious or even > humorous comments on it appreciated. Interesting indeed. 5.8 TFlops is just in the low end of the useful range, I didn't see any mention of how you'd gang multiple boxes together, but I assume that they'd like to sell them so they must have considered this. Terje -- - <Terje.Mathisen(a)hda.hydro.com> "almost all programming can be viewed as an exercise in caching"
From: Nick Maclaren on 2 Apr 2008 16:52 In article <-KCdnWvYdIczcG7anZ2dnUVZ8tPinZ2d(a)giganews.com>, Terje Mathisen <terje.mathisen(a)hda.hydro.com> writes: |> |> > It looks interesting, but will it succeed? Any serious or even |> > humorous comments on it appreciated. |> |> Interesting indeed. |> |> 5.8 TFlops is just in the low end of the useful range, I didn't see any |> mention of how you'd gang multiple boxes together, but I assume that |> they'd like to sell them so they must have considered this. For top-end HPC, yes. But consider alternative uses. The very small box should make an excellent development engine for 'serious' supercomputers (and is my personal interest in it). The medium one should fit into an office or laboratory without too much hassle. Also, the great unknown is whether you can get a higher proportion of its theoretical peak out of it than you can on dual-socket, quad-core Intel systems. That is the key to whether it is a good buy or a bad one - plus its price, of course. Regards, Nick Maclaren.
From: Chris Thomasson on 2 Apr 2008 20:24 "Nick Maclaren" <nmm1(a)cus.cam.ac.uk> wrote in message news:ft0fhe$7qg$1(a)gemini.csx.cam.ac.uk... > > It looks interesting, but will it succeed? Any serious or even > humorous comments on it appreciated. Very cool! Do you happen to know if they provide an ISA manual and an assembler? I assume I can use POSIX and C because it runs Linux. I am interested in the semantics of the assembly instructions that drive their DMA engines.. I think it would be neat to be able to design custom message-passing frameworks for this beast.
From: Terje Mathisen on 3 Apr 2008 03:49
Nick Maclaren wrote: > In article <-KCdnWvYdIczcG7anZ2dnUVZ8tPinZ2d(a)giganews.com>, > Terje Mathisen <terje.mathisen(a)hda.hydro.com> writes: > |> > |> > It looks interesting, but will it succeed? Any serious or even > |> > humorous comments on it appreciated. > |> > |> Interesting indeed. > |> > |> 5.8 TFlops is just in the low end of the useful range, I didn't see any > |> mention of how you'd gang multiple boxes together, but I assume that > |> they'd like to sell them so they must have considered this. > > For top-end HPC, yes. But consider alternative uses. The very > small box should make an excellent development engine for 'serious' > supercomputers (and is my personal interest in it). The medium > one should fit into an office or laboratory without too much hassle. > > Also, the great unknown is whether you can get a higher proportion > of its theoretical peak out of it than you can on dual-socket, > quad-core Intel systems. That is the key to whether it is a good > buy or a bad one - plus its price, of course. Price is the really important consideration: Currently StatoilHydro has around 20 TF (afair) in its seismic clusters, and the machine rooms have been through the needed refit to handle the cooling issue. In an upgrade, each of the cpu nodes would/will get upgraded, to roughly double the performance, something which would more or less keep up with increased needs. If SiCortex can deliver the same performance at half or lower cost, then it might be interesting, particularly if it would at the same time allow scaling to another order of magnitude without power/cooling problems. Terje -- - <Terje.Mathisen(a)hda.hydro.com> "almost all programming can be viewed as an exercise in caching" |