[geeks] Game GPU clusters for supercomputering

[Patrick Giagnocavo]

Joshua Boyd wrote:
> And of course, you can get Cell blades from IBM and Mercury.

I have a client that is perfectly satisfied with nVidia GPU for 
computing and has decided not to port to Cell.  Part of the reason is 
cost, part of the reason is that nVidia is "good enough".  Also Matlab 
has a plugin for the nVidia GPU, so he can quickly prototype stuff.

>> 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.  Now each of those 4 has an insane number of 
> cores, and each core has an insane number of hardware contexts for doing 
> something like hyper threading.  Basically, it looks like a 512 vector 
> processors.

I don't think they are very cheap.

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

Just remember that being better technically doesn't necessarily mean it 
will be a hit in the marketplace.

--Patrick