High-level library for GPGPU

Question

I want to use GPU for scientific computation. I know CUDA and OpenCL, but is there more high-level library (if possible, in Python) for GPGPU?

score 1 · Accepted Answer · answered May 07 '18 at 19:53

1

There are lots of ways doing this depending on what kinds of computations you want to do and how deeply you're willing to dive into the guts of the GPU.

If you're using an NVIDIA GPU with CUDA, you can use the NVBLAS library supplied with the CUDA toolkit as a drop-in replacement that accelerates Level 3 BLAS functions.

The Anaconda distribution of Python has a package called Accelerate that offers a bunch of features for GPU acceleration using NVIDIA GPU's.

The pycuda package is another option.

answered May 07 '18 at 19:53

Brian Borchers

128
4

This answer seems to be too NVIDIA-leaning – HEKTO May 16 '18 at 19:41
I'd be interested in seeing an answer that pointed to resources based on OpenCL on AMD GPU's. I'm aware of an OpenCL version of pycuda, but not much else. – Brian Borchers May 17 '18 at 02:46

High-level library for GPGPU

1 Answers1