Intel announced at Siggraph 2010 OpenCL BOF, they will release an Alpha SDK by the end of the year: done!
This SDK is tagged as an 'Alpha' SDK and it only provides a CPU implementation of OpenCL. Is Intel going to provide a GPU implementation? Are Intel GPUs actually capable to efficiently accelerate OpenCL?
It's good to see some interest by Intel for OpenCL but let's remember that OpenCL 1.0 as been release on 8th December 2008 and Intel only provides an CPU implementation almost 2 years later... On AMD side this step has been reach on 5th August 2009 (CPU implementation) and nVidia has released a GPU implementation on 28th September 2009 for public... a long time ahead of Intel!
At that time nVidia doesn't have a CPU implementation of OpenCL but for a GPU only company it makes sense. However, this could change as nVidia has announced at the GPU Technology Conference CUDA x86 which is a x86 implementation of CUDA. As OpenCL is built on top of CUDA at nVidia, chances are that nVidia will sooner or later provide an implementation of OpenCL for CPUs.
Currently, AMD CPU implementation of OpenCL runs as well on Intel CPUs which makes me wonders how Intel and AMD implementations compete against each other.
nVidia is far ahead on the topic of GPU computing on the regard of performance and the long experience with CUDA might paying of even if is seams that nVidia OpenCL implementation is significantly slower than CUDA. I thought that AMD could reach at least the same level of performance thanks to impressive theorical performance but as time goes, it becomes less likely to see it happening with Radeon 4000, 5000 and 6000 series or maybe AMD implementation is still not mature enough.