The Khronos Group has released the OpenCL 1.0 specification today, after an incredible effort to ratify a complex specification in a very short amount of time.
With all the major hardware vendors onboard, OpenCL has the potential to become the de-facto interface to access heterogeneous compute devices (be they GPUs or CPUs) on a wide variety of operating systems.
http://www.khronos.org/news/press/releases/the_khronos_group_releases_opencl_1.0_specification/
The Guardian has an article on the desire to bring GPGPU to mobile platforms as a means of gaining efficiency and thus longer battery life. It also discusses the emergence of OpenCL has a candidate for an open-standard compute library under Khronos’s new Compute Working Group.
It is interesting to read in light of the recent information on DirectX 11 which will offer compute shaders. It seems that the current state of affairs in real-time computer graphics may repeat itself with GPGPU: one API on Windows and another for everyone else. Of course, there are a number of other options at this time, namely from NVIDIA, AMD and RAPIDMIND, which could also end up dominating the market.
No matter what happens with respect to the APIs, it’s certainly good news that GPGPU is becoming more ubiquitous and approachable every day.
CUDA is NVIDIA’s architecture and API for GPGPU – general purpose GPU programming. The fact is, those graphic cards are hugely powerful parallel computing units, and everyone stands to benefit by exploiting them to do far more than just outputting images to a screen.
Well today, NVIDIA made CUDA available for Mac OS X. This is a sign that times are changing for Apple when they start to make such a kind of inroad. And of course, everyone in the (probably small) Mac HPC business and in the scientific community is probably very happy about this.
So if you own NVIDIA hardware, go grab it and give it a try!