Why is the neural network community so fond of CUDA and not OpenCL? What happened to Linux users and their opensource and Muh-ltiplatform?
>>54382409
The botnet happened
help why is single precision bad
Because CUDA is much easier to learn. There are even OpenCL guides that suggest you learn CUDA first.
You can either get a job and give in to the botnet or be a smelly hobo who use free as in no cost and free as in freedom software.
>>54382742
Yeah i can understand that, but neural networks are mainly if not only usable on Linux, the people who develop the Neural Networks are the freefags yet they keep using CUDA for every singe GPU implementation. It's not like OpenCL is an AMD only thing as Nvidia and Intel also have OpenCL enabled hardware.
>>54382409
CUDA was first, CUDA has a lot of libraries, CUDA is easier too
OpenCL is also broken on some systems, mainly drivers issues, on both AMD and Nvidia
>>54382409
Opencl is framework for heterogeneous computing with fuckeverything from fpgas to graphics cards. Anything that can do computations can be made to support opencl. Meanwhile cuda is not only for gpus but only for nvidia gpus. Without actually using either of these, one can already guess that cuda is going to be the smaller, simpler, more well defined one.
>>54382830
>the people who develop the Neural Networks are the freefags
Topkek
