That is impressive. I wonder how much follding would have been done if there was a decent PS3 game?

what surprises me is folding not using double-precision floating point arithmetic, which is the bare minimum in computational science - cell's double precision performance is over 10x slower than its single precision performance

Taking TFLOPs/number of CPUs as a kind of measure of how quick these things are, PS3s wipe the floor with general purpose CPUs. It's interesting to note though, that by this metric, GPUs are indeed better, but there's less of em. Do GPUs only take certain work unit types too? (does anyone know?)

Following on, on the point that PS3's can only do one type of work unit, does anyone know how CPUs compare to the PS3 on those work units? is it still such a whitewash?

jamiecockrill
…Do GPUs only take certain work unit types too?…

My understanding is that they take (pretty much) most work units - but that their unique parallelism only allows them to excel with certain units

Could be wrong - but I believe this is the case :)

This doesn't impress me whatsoever.

Average TFLOPS per processor
GPU average: 0.0595
PS3 average: 0.0247

If there were more GPUs folding, they would crush the PS3 by producing over twice as much processing power.

And, more pointedly, GPUs would crush the PS3 on the basis that it can munch through more types of work unit. I guess the next question is, why are there so few graphics cards that support folding? The PS3 should be applauded for doing its bit as well, even if it's only one type of work unit, it must be a pretty amazing contribution! I'd love my 360 to be able to do the same.