[geeks] Game GPU clusters for supercomputering

[Shannon Hendrix]

On May 22, 2008, at 23:29 , Joshua Boyd wrote:


>> On a related note, nVidia now sells GPU clusters.
>
> Well, they aren't very self sufficient.  I suspect that on average  
> you need a 1U computer for each 1U of GPUs, and I think 1U of GPUs  
> means something like 4 of them.

The S870 has 4 each 1U, yep.

That's actually pretty dense when you consider the power requirements  
and the huge number of cores running.

>
> The GPU stuff is cool looking, but there are a lot of things I like  
> better about the Cells.  The Cells seem a bit more straight forward  
> to program.

I thought they seemed kind of weird, but then I've only read  
documentation, not actually done either one.

I wish I had the opportunity.

> The Cells can take a lot more memory (first generation is 2 gigs per  
> chip instead of 1.5 gigs, but the new model that just started  
> shipping can take 32gigs per chip).

Yes, but the Cell doesn't actually address that memory directly.  The  
way I understand it, they only address about 256K, and there are  
instructions that treat the rest of the memory as an incoming stream  
or something like that.

I have an article on how to program them efficiently, and it seems  
anything but straight-forward to me.

> The Cells don't require a PC next to them.

I thought a front-end was a requirement for most applications though,  
since you don't want to write a lot of the I/O handling part of your  
app on the Cell.

It seems best to treat most of these things as coprocessors and do  
your dirty work on a front-end host.

> And long term most importantly, the Cells seem to be much better in  
> the documentation department.  With Nvidia, your choices are Cg,  
> GLSL, or the CUDA C compiler.  The machine details aren't publicly  
> documented.

They probably won't be doing that, since it would be giving away those  
precious GPU secrets.

> With the Cell you have a lot of documentation and you have GCC,  
> which means you have C, C++, Objective C, Fortran, and ADA.  I'm not  
> sure that C++, Objective C or Ada are practical on the SPUs, but  
> Fortran has got to be a big deal for a lot of people.  Oh, and we  
> have assembly.  And having assembly means that we can get a lot more  
> languages for it.

Heh... I'm not sure lacking GNU C support is a bad thing.

GNU compilers are only good in so far as they are widely available.  I  
personally would love to see better compilers without the GPL baggage.


-- 
"Where some they sell their dreams for small desires."