The rCUDA team has been working for the last 6 months on porting the rCUDA middleware to the ARM processor architecture. The port has been made on the CARMA and KAYLA nVIDIA development platforms. We are now completing the first release of rCUDA for this low-power processor, which will be available very soon. Our tests include running an application on the ARM processor that demands GPGPU services on an x86 remote rCUDA server. Conversely, it is also possible to locate the GPU on an ARM-based server and demand GPGPU services from an x86 computer.
The rCUDA remote GPU virtualization framework was recently updated in order to provide support for the last CUDA 5.0 release. Now, with the launch of the new CUDA 5.5 release candidate, the rCUDA team has tested the rCUDA middleware with this new CUDA release and is working in updating rCUDA to the new CUDA version.
The new features of rCUDA, as well as its last developments, were presented on June 16 at the HPC Advisory Council European Conference 2013, held in Leipzig (Germany) during the International Supercomputing Conference (ISC’13). The slides of the presentation are available here. You can also access a video with the presentation in this link.
We are happy to announce next rCUDA updates that will complete the rCUDA 4.0 version. The new features will complete the new rCUDA version, which includes binary compatibility, InfiniBand native support and enhanced data transfers.
In addition to the full binary compatibility feature recently released, the new features are going to be:
- rCUDA 4.0 release with CUDA 5.0 API support (excluding graphics interoperability).
- SLURM scheduler integration with rCUDA.
The rCUDA team thanks Bright Computing, Inc its donation of an NVIDIA Tesla K20 GPU during the SuperComputing 2012 exhibition held in Salt Lake City (Utah) last November. This last generation GPU will be used to further improve the features and performance of the rCUDA middleware.