I have been wondering if the Afterburner FPGA will be opened up for other usage.
Well AMD is working hard on ROCm. Or may be it should be Google working more on Tensorflow ROCm rather than focusing on CUDA?
I think ROCm is close, very close. But like everything when the work is 80% done and you thought the 20% is going to be easy, that 20% normally takes the same amount of time as the first 80%. People are aware of the CUDA lock in, and once AMD reaches the point of good enough, more people will switch and more people will help with ROCm
And needless to say Apple by themselves are also heavy user of Machine Learning.
Not being an expert on the intricacies of ML, my take is that Apple does not want NVIDIA to subvert or be able to exert any influence or control over the on device ML that any A11 or A12 equipped hardware can perform. Apple wants its devs using CoreML and writing native for its iOS devices, not letting CUDA get its foot in the door on the Mac side and then developers focus on the largest marketshare or think they can forgo CoreML. It would give NVIDIA an even bigger presence in the market and Apple does not want another company dictating this technology to them as they use it to their competitive advantage when adding features to the iPhone, especially with photo processing. Just my 2¢.
Last edited: