Has anyone here worked with OpenCL ? I was interested in learning it for a while but couldnt find any reliable course/tutorial or even sane documentation to follow. And is it even worth it now!?

Played with it for a bit, but dropped it ask couldn't find a sensible project to use it for. Despite being open, it seems to have kind of lost out to CUDA for the most part, so that may be the more sensible thing to invest time into learning.

@AlmondSauce The thing i find frustating with CUDA is, it is not compatible with AMD GPUs. I was told to skip CUDA because of this and OpenCL being a higher level language.

@purist Yeah, I wish OpenCL won for that reason, but it hasn't. Depends what your use case is - but look at libraries like TensorFlow for instance, CUDA baked in and OpenCL is just a half-complete experiment. Same with MXNet, same with a whole other bunch of stuff.

Hopefully the situation changes, but at present it looks pretty dire for OpenCL.

@AlmondSauce It is infuriating that the entire industry just accepted Nvidia to be the sole winner of this battle and is leaving AMD to bleed out and die. Scientific computing is the future, and i cant help but wonder how tf did anyone not see this as a big problem well ahead.!? I have learnt some CUDA and it is nice. But i dont want to learn an enterprise language, but i also dont want to learn something which will be useless very fast.

I am learning pytorch. But i feel like, it looks at the problem of parallelisation very differently than CUDA. No offense, pytorch/TFs graph based parallelisation maybe the future for ML applications, but that seems like a very specific application of parallelisation to me.