OpenACC May Hint to the Future of CUDA at Nvidia
An announcement from Nvidia made at the most recent SC11 conference apparently got lost in a wave of supercomputing announcements.
However, it’s a pretty significant change in Nvidia's strategy on how to approach multi-core computing in the future and worth a note.
Together with Cray, PGI and CAPS, Nvidia announced a new parallel programming standard called OpenACC. OpenACC has been published as API that includes a range of compiler directives which describe code additions for C, C++ and Fortran to take advantage of accelerators to speed up code processing in highly data parallel environments. OpenACC is designed to apply to general parallel processors and can be used on GPUs as well as CPUs.
Cray, PGI and CAPS are planning to deliver initial compiler support for OpenACC Q1 2012. There was no information when Nvidia would provide a dedicated OpenACC compiler, but the company said that its CUDA architecture is "fully compatible and interoperable" with OpenACC. Nvidia did not indicate that it would drop CUDA anytime soon, but its participation in this group of companies suggests that Nvidia may be planning for a time beyond CUDA.
While CUDA code also runs on multi-core processors, the underlying architecture is proprietary technology. The advantage of OpenACC is a much broader range of devices support. Initially, there is no indication that OpenACC will have a huge impact in consumer applications as the group of OpenACC supporters is especially hoping for interest from developers in supercomputing applications fields in industry and academia covering topics such as chemistry, biology, physics, data analytics, weather and climate, intelligence, and many other fields.
Yeah I was wondering why Nvidia wasn't throwing more weight and tools behind OpenCL. Oh wait, then it'd run on AMD hardware effortlessly, too.
LOL, you obviously have zero openCL development experience in coding or optimizing for openCL devices I see.
You know that Cuda is still the largest deployed and developed for GPGPU solution by a long shot right?
What are you talking about? First of all, what OpenCL applications are you referring to? I'm not aware of very many, and no, they haven't been widely adopted. As of now, CUDA is by far the most widely adopted parallel computing architecture.
Wow dude, you really sound like you know what you're talking about. Nvidia provides full support for OpenCL and did so before AMD. In fact, one of the few OpenCL applications I'm aware of, SmallLuxGPU, seems to perform significantly better on Fermi based GPU's, especially gf100/110.
http://www.anandtech.com/show/4239/nvidias-geforce-gtx-590-duking-it-out-for-the-single-card-king/15
Yes, CUDA is proprietary, but given Nvidia's market share relative to AMD's in GPGPU computing, introducing a proprietary standard really isn't much of a problem.
If you're just going to spam baseless AMD fanboyism in the hopes of deceiving the ignorant, may I suggest...
http://semiaccurate.com/
we don't need this !!