[Users] Cell based workstation from Sony

[John Shalf]

Hi Gaurav.
My primary comment is that Sony would be better off pairing the PS3  
with a CUDA GPU.  The traditional GPUs, like RSX, have a programming  
model that is based around packing your problem data into a texture  
map and then rendering it into the framebuffer.  To say that is  
awkward is exquisite understatement.  Some very dedicated GPGPU  
programmers have demonstrated you can do an amazing variety of things  
with this model, but it is hardly what you would consider general- 
purpose.

The CUDA, by contrast, is more of a conventional manycore processor.   
It gets rid of the distinction between vertex processors and pixel  
processors (they are all just processors).  You are not forced to use  
a stream programming model (eg. you can gather from memory *and*  
scatter too).  You can do inter-processor communication to take  
advantage of the on-chip bandwidth (cannot do that with RSX).

So, I think that CUDA will take a lot of steam out of the current  
GPGPU programming.  Once programmers get all of the power of the GPUs  
packaged together with a more flexible programming model, older-style  
GPU programming models are going to give way to this more flexible  
architecture.  (that is to say, ATI is likely to move in this same  
architectural direction as well).

So it makes sense that Sony would put an RSX in the CPBW since it  
could be a development platform for the PS3.  However, if they are  
interested in combining GPU programming with Cell programming, they  
might be better off pairing with NVidia.

-john

On Mar 12, 2007, at 5:26 AM, Gaurav Khanna wrote:
> Hi all
>
> I was recently contacted by Sony concerning an evaluation of the  
> market
> (in the scientific HPC community) for a Cell processor based  
> workstation
> which includes Sony's RSX GPU. These are the same technologies used  
> in the
> PS3, but they are appropriately modified for general purpose  
> computing.
>
> Apparently, detailed specs don't quite exist yet (they are still in  
> flux)
> but the price point is expected to be in the 7K range.
>
> Can you folks send me your thoughts about the likelihood of success of
> such a beast in the scientific community? Thanks very much.
>
> I'm including some links below on the Cell processor and RSX, in case
> you're not very familiar with these ..
>
> Best Regards
> Gaurav
>
> http://en.wikipedia.org/wiki/Cell_processor
> http://en.wikipedia.org/wiki/RSX_%27Reality_Synthesizer%27
>
>
>
> -----------------------------------------------
>
> GAURAV KHANNA
> UMass Dartmouth, Physics
> (508) 910 6605
> http://gravity.phy.umassd.edu/
>
> "Black holes are where God divided by zero." - Steven Wright
>
>
>
> _______________________________________________
> Users mailing list
> Users at cactuscode.org
> http://www.cactuscode.org/mailman/listinfo/users