Graphics Card Computing for Cosmology
Posted: November 07 2007
Hi everyone,
Thanks to a Foundational Questions Institute minigrant, I've recently been able to experiment with doing numerical computations for cosmology on graphics cards. New cards are very fast and can now be programmed (somewhat...) straightforwardly in a [q]C[/q]-like language. At the moment they only work in single precision but double precision hardware should soon be available.
My main project on them is the numerical simulation of eternal inflation and the investigation of the possible effects of volume-weighting on the typical history. I've also done a bit on the Cholesky factorization of matrices, which is of course very useful in a wide variety of contexts, not least in CMB likelihood calculations. I've been pretty impressed with the performance so far.
I've put together a website that might be of interest if anybody is thinking of doing anything similar, see here.
Steve Gratton.
Thanks to a Foundational Questions Institute minigrant, I've recently been able to experiment with doing numerical computations for cosmology on graphics cards. New cards are very fast and can now be programmed (somewhat...) straightforwardly in a [q]C[/q]-like language. At the moment they only work in single precision but double precision hardware should soon be available.
My main project on them is the numerical simulation of eternal inflation and the investigation of the possible effects of volume-weighting on the typical history. I've also done a bit on the Cholesky factorization of matrices, which is of course very useful in a wide variety of contexts, not least in CMB likelihood calculations. I've been pretty impressed with the performance so far.
I've put together a website that might be of interest if anybody is thinking of doing anything similar, see here.
Steve Gratton.