On Aug 3, 8:02 am, "stefanba...@yahoo.com" <stefanba...@yahoo.com>
wrote:
> >Since these chips are not so common
>
> Larabee and G300 (GT300) are Intel & NVIDIA projects of next GPU
> generation. Intel is open about Larabee architecture and it is clear
> that Larabee is very different from today's GPU architeture. It is
> closer to massive multi-core & hyper threading PC; each core/thread
> appears as really undependable scalar processor (well, superscalar
> more accurately). Information about GT300 is more speculative and not
> very reliable so some NVIDIA's "leaks" states that it is going to be
> a truly MIMD architecture sounds for me more like marketing BS similar
> to CUDA/MIMD hype - definitely I expect it to be mach closer to MIMD
> yet I have my doubts that it is a match for Larabee in term of MIMD
> performance but I would not be surprised to see that Larabee may be
> slower for existing game then G300.
>From here:
http://electronicdesign.com/Articles/ArticleID/19641/19641.html
Larabee uses the "CPU array" architecture, unlike those of GPU, which
is much easier for compiler optimization.
And as its have the x86 as its core, I think it will take off much
faster than other CPU architecture.
>
> >I don't think there exists any opensource OS
> > running on these processor yet.
>
> Well, they are not available yet (expected next year??? or may be end
> of this) but Larabee seems may be capable to run let say Linux (x86)
> with minor modifications and I'm sure it is not the case for G300
> (yet, too little reliable information about G300 to consider my
> personal opinion as a reflection of indisputable facts).
>
> >So the personal computers from
> > Nvidia must be running some proprietary OS.....am I right?
>
> "must be running" sounds more wishful joke ;o)))
>
I just read this:
http://electronicdesign.com/Articles/ArticleID/21156/21156.html
or to quote:
The Tesla S1070’s four 1.44-GHz GPUs house a total of 960 processing
cores that have a peak double-precision floating-point performance of
345 GFLOPS. For single-precision work, the system delivers 4.14
TFLOPS. The four-node cluster delivers four times this performance.
For many applications, the GPU approach offers significant benefits
including lower power requirements, lower cost, and less space
sometimes by more than a factor of 30.
Linux cluster software is available now, and Windows Server 2003/2008
will be available later in the year. Larger configurations are
available.
Hmmmmmm, Nvidia does use Linux after all.....and considering that the
processor is NOT x86 compatible.....wow...that is a lot of work...esp
on the compiler side....:-).
