The last few weeks I have been looking at the map-reduce area; specifically Hadoop and IBM’s BigInsight. This led to questioning how the data-parallel activities could be sped up. Found a reference to the usage of GPU for such data parallel tasks. A quick yahoo search kicked off the following site navigation:
- Mars: A MapReduce Framework on Graphics Processors: http://www.cse.ust.hk/gpuqp/Mars.html
- NVIDIA’s CUDA support: http://developer.nvidia.com/what-cuda
- NVIDIA’s support for OpenCL: http://developer.nvidia.com/opencl
- WebCL: http://www.khronos.org/webcl/
- OpenCL podcasts at MacResearch: http://macresearch.org/opencl. This rather interesting video from MacResearch made me spend some time on their podcasts. I really liked them
- An IBM research focusing on OpenCL: https://researcher.ibm.com/researcher/view_page.php?id=1835. I read a couple of their publications, but wasn’t able to find a public link. Rather interesting read.
- Finally, looking for a Java based API for OpenCL, I came across JOCL
The promise of a GPU powered speed-up in processing is a rather seductive one. There are ofcourse a bunch of disadvantages going with the GPU (refer to one of the Khronos, or MacResearch links for details).
This is an area I will be looking at in some detail going forward.