Richard Jones' Log: PS3 distributed compute farm?

Fri, 23 Mar 2007

So apparently PS3s are kicking butt in distributed protein-folding. According to the folding@home FAQ, "With about 40,000 such machines, we would be able to achieve performance on the petaflop scale."

Makes me wonder how feasible would it be for PS3 owners to lease their hardware to a distributed processing company...

Just so we're clear on this, the fastest supercomputer on the planet* as of November 2006 was the IBM BlueGene/L system with 280.6 teraflops. Now, the benchmark is different of course, but the PS3 contribution to folding@home is (as of today) 497 tflops (that's with 20,287 participants). Not bad for a video game console.

I can't find rankings of distributed computing systems... does anyone know whether such a thing exists?

*: OK, except the one the NSA uses to decode the brainwave patterns scanned from the heads of every single person on the planet using their secret laser satellites.

Comment by Xander on Sat, 24 Mar 2007

The big difference in benchmarks is that supercomputers use 64-bit floats, whereas the PS3 is almost certainly being measured with 32-bit floats.

Comment by Xander again on Sat, 24 Mar 2007

Starting price for a BlueGene is a measly $1.5M[1], whereas 20,000 PS3s will cost over $20M (Sony, when will you learn?).

So BlueGene is a better gaming platform than PS3, but not quite as price-competitive as the Wii.


Comment by Xander again again on Sat, 24 Mar 2007

Dividing the total TFLOPS contribution of each platform by the number of active nodes shows that the GPU is twice as floppy as the PS3 (which is itself 25x floppier than the PCs and Macs).

The GPU client is for ATIs only at the moment, so it can't be using any features beyond a 7900... why aren't the PS3's using their GPUs instead of the Cell then?

Comment by Tristan Seligmann on Mon, 26 Mar 2007

If only the PS3 were nearly as good at running games as it is at running Folding@Home...