First Die-Shot Photograph of Intel Larrabee
NV55
discussion
http://forum.beyond3d.com/showthread.php?t=54144
apparently there are 32 cores in this chip, plus a number of other
structures / functional units
Terje Mathisen
> http://img2.pict.com/51/47/27/8651f48224d11fe0dcf2ebc330/53NzS/intellarrabeedieshot6.jpg
>
> discussion
> http://forum.beyond3d.com/showthread.php?t=54144
Quite nice thread actually!
>
> apparently there are 32 cores in this chip, plus a number of other
> structures / functional units
Assuming Intel can get this to work nearly as well as Mike Abrash' and
Tom Forsyth's talks at GDC indicated, it looks a lot like a single-chip
solution for a cheap-to-manufacture box with very decent 3D performance.
Terje
--
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"
NV55
wrote:
> NV55 wrote:
> >http://img2.pict.com/51/47/27/8651f48224d11fe0dcf2ebc330/53NzS/intell...
>
> > discussion
> >http://forum.beyond3d.com/showthread.php?t=54144
>
> Quite nice thread actually!
>
>
>
> > apparently there are 32 cores in this chip, plus a number of other
> > structures / functional units
>
> Assuming Intel can get this to work nearly as well as Mike Abrash' and
> Tom Forsyth's talks at GDC indicated, it looks a lot like a single-chip
> solution for a cheap-to-manufacture box with very decent 3D performance.
>
> Terje
> --
> - <Terje.Mathisen at tmsw.no>
> "almost all programming can be viewed as an exercise in caching"
Yes. I could imagine low-end PCs using a single 32 core Larrabee chip
(later 48 or 64 cores) having very nice 3D performance (2 TFLOPs)
Then I could see higher-end PCs, with 2 Larrabee chips, perhaps 4,
as high-end gaming rigs and other machines as personal supercomputers
--much like what Nvidia does with their GeForce-based Telsa chips.
I could also see Larrabee, or perhaps Larrabee2, being the heart of a
next generation game console from Sony or Microsoft, or less likely,
Apple or SEGA.
And like with IBM's CELL in Roadrunner, I could see Intel getting a
contract to use Larrabee in a next-gen record breaking Supercomputer.
Morten Reistad
There are more possibilities.
With 2 Tflops, or for some applications, 2 Bips, you can actually run
stuff like a video conference with 25 participants, and actually merge
the video streams correctly and make consolitdated streams for
all participants. This means you can cease all the corner-cutting
video telephony applications do today, and just do it the simple and
reliable way.
You can do all the transcoding you like for voip. You can do it in
high quality too, and make interpolated waveforms instead to the
crude stuff we do today when we have to handle GSM sound in a
real phone network. You can do so for GSM too, of course.
You can do realtime signal processing to you heart's content, so
you can push yet more bits through the rotten copper wires the telcos
have put in just after ww2.
We can do radio processing as well. We can dig bits out even deeper
from the sea of noise that surround us. A teraflop should be good for
another 10 dB depth into the noise.
All of these have straightforward, well parallell-suited algorithms
available that just have been waiting for larabee. We just need
cycles by the billions to do this.
We may even see advances in sensing, good enough for cars to have
some self-steering ability, but they probably won't rival horses
yet.
-- mrr
Ken Maltby
"NV55" <nvidi...@mail.com> wrote in message
news:c70aa95c-9c5a-4580...@z7g2000vbh.googlegroups.com...
What might have more effect, will be the acceptance of a
common API for the efficient utilization of parallel processing
environments/architectures. If something like OpenCL takes
off and programers can have some expectation that their code
will work regardless of the exact hardware, then we might see
some real progress.
Luck;
Ken
P.S. As one who transcodes video, I would love to see how
the Larrabee might work along side the SpursEngine.
Martin Thompson
> We may even see advances in sensing, good enough for cars to have
> some self-steering ability, but they probably won't rival horses
> yet.
>
*Some* self steering ability has been around for a while now:
http://youtube.com/?v=xlVx4Dhglkg
(see about 4:40 )
That was 1994... similar systems are in production now
(eg. http://ir.trw.com/releasedetail.cfm?ReleaseID=345841). And no, they
don't rival horses just yet :)
(I worked peripherally on the Lucas/Jaguar project, and was very
involved its follow-ups and on the TRW production one)
Cheers,
Martin
--
martin.j...@trw.com
TRW Conekt - Consultancy in Engineering, Knowledge and Technology
http://www.conekt.net/electronics.html
Mike Ray
Yeah and Skynet and all those machines....