From the mailing list …
We’d like to announce the availability of GPU-HMMER, a port of HMMER’s hmmsearch utility to NVIDIA CUDA-enabled GPUs. Anyone who is interested is invited to visit www.mpihmmer.org to get the source. A brief user’s guide and installation guide are provided at mpihmmer.org as well. This version has been tested on 8800 GTX Ultra and Tesla C1060 GPUs, though we expect that the code should work on a wider range of GPUs than has been tested.
The current source code supports a single GPU with speedups that range from 20x-38x, depending on the size of the HMM. In the coming weeks we plan to release support for multiple GPUs within a single node. Current tests demonstrate speedups that exceed 100x using 3 Tesla GPUs, also depending on the size of the HMM. Thereafter we plan to integrate GPU support directly into MPI-HMMER to support multiple GPUs via MPI.
As always we welcome any feedback or bug reports. We are especially interested in compatibility reports and will post compatibility updates to www.mpihmmer.org as necessary.
Viewed 7431 times by 1491 viewers