3

1

I'm aware of GPU computing making great strides in tomography, but are there any projects which use GPGPUs in structural biology? Or, are the algorithms from, for example, crystallography not amenable to use on GPUs?

flag

1 Answer

0

GPGPUs are excellent for parallel processing. Much of the challenge of implementing code for use with GPGPUs is adapting a serial algorithm for parallel processing. Crystallography is certainly going to be using many large matrices for pairwise calculations, maximum likelihood estimates, etc. which lend themselves nicely to parallel processing and thus GPGPUs.

I did a quick search through the crystallography literature and didn't immediately see anything.

Because it's a rising field, many applications for GPGPUs have not been yet been realized! This could be a great opportunity for you to pioneer these applications.

Here's a link to nVidia's CUDA guide for GPU programming: http://developer.download.nvidia.com/compute/cuda/3_0/toolkit/docs/NVIDIA_CUDA_ProgrammingGuide.pdf

Here's a general tutorial on GPU programming: http://www.ddj.com/architect/207200659

link|flag
The problem is that the parallelizable routines used in refinement don't consume enough of the total runtime for use of the GPU to be particularly useful. (Amdahl's Law is key here.) 3D FFTs are also very complicated, so actually writing the code is a non-trivial task. There is also the issue of memory management and bandwidth. The life science applications for which GPU computing has proven most useful tend to heavily depend on a single repeated calculation, like nonbonded interactions in molecular dynamics. (The equivalent in crystallography is much simpler, and doesn't take very long.) – Nat Jun 23 at 19:09

Your Answer

Get an OpenID
or

Not the answer you're looking for? Browse other questions tagged or ask your own question.