Blender 2.79: OpenCL on par with CUDA

Re:Blender 2.79: OpenCL on par with CUDA by lettergram on Wednesday April 12, 2017 @11:13AM (#14098442)

I tried picking up OpenCL then CUDA as well and had a similar experience.

CUDA simply feels less hacked together to me. It's like working with a well thought out and documented code base (CUDA) vs working with some library that has little documentation and contradicting syntax and formats (OpenCL).

IMO that's why AMD is losing the "deep learning" battle, it's just not easy to develop using OpenCL. At least, not as easy as it should be.

[ Parent | on HN ]

Re:Blender 2.79: OpenCL on par with CUDA by happycube on Wednesday April 12, 2017 @11:22AM (#14098543)

Losing? When it comes to TensorFlow et al they haven't really shown up for the fight.

It's a vicious cycle though, since everything uses CUDA/CuDNN everyone buys nVidia cards, which means that developer interest in getting AMD cards fast isn't there.

If AMD GPU cores were in phones etc, they would probably be getting TF support for client work at least. Heck, considering they sold off that dept to Qualcomm, which is getting supported...

[ Parent | on HN ]

Re:Blender 2.79: OpenCL on par with CUDA by lettergram on Wednesday April 12, 2017 @11:33AM (#14098657)

There's really no reason AMD themselves couldn't hire 5 developers to integrate with TensorFlow. I agree it's a vicious cycle, but I feel if they release support for tensorflow it wouldn't take long to shift people from Nvidia to AMD. AMD has a significantly cheaper offering, and no company wants to be tied by only one supplier.

Considering how small of a dent that would make in their bottom line (financially), it's really amazing they haven't yet... The potentially upside is easily in the tens (if not hundreds) of millions of dollars.

So definitely, I agree they intentionally have decided to not support deep learning. For the life of me, I don't know why.

[ Parent | on HN ]

Re:Blender 2.79: OpenCL on par with CUDA by happycube on Wednesday April 12, 2017 @02:00PM (#14100094)

Looking at things some more, it seems they are working on it... just a year or two after it needed to be done. :P

https://github.com/GPUOpen-ProfessionalCompute-Tools/HIP/iss...

[ Parent | on HN ]

Re:Blender 2.79: OpenCL on par with CUDA by ansible on Wednesday April 12, 2017 @12:01PM (#14098962)

There's really no reason AMD themselves couldn't hire 5 developers to integrate with TensorFlow.

The problem is that it isn't just TensorFlow. It is also Caffe, Theano, Torch, etc., etc., etc.. For deep learning, Nvidia GPU acceleration is almost a given.

[ Parent | on HN ]

Re:Blender 2.79: OpenCL on par with CUDA by lettergram on Wednesday April 12, 2017 @12:31PM (#14099259)

If AMD can write an abstraction layer similar to CuDNN, and write the initial integration to TensorFlow and Caffee I think they'd gain enough community support to finish out the rest.

80% of deep learning users use the TensorFlow, Caffe, or Torch frameworks.

[ Parent | on HN ]