Python and Cuda

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

It seems that Intel is provding full support for Python with API support.

Is this possible than Nvidia provides API support for C++/Python, especially for new API like Cuda 8,
maybe by re-using Pycuda.

It would efficiency by using recent API amd increase performance on both sides.

2

Have you have a chance to try PyCUDA (PyCUDA) to see whether that fits your needs? Or maybe Numba (http://numba.pydata.org/) does what you want?

3

Hi,

NUMBA is incomplete API (almost not usable), NUMABA PRO is better but paid version.

PYcuda has some functionnalities, has missing items in the API like CuDNN and lastest Cuda 8 Graph API.

For example, Intel is actively supporting Python through those API:

JIT Compiler for Intel CPU
Intel® SDK for OpenCL™ Applications

DAAL Python Library: Accelerator for Intel CPU.
Intel® Distribution for Python*

So, If NVDIA actively supports some API Python interface on top of CUDA C API (maybe reusing Pycuda and numpy), it would standardize and have new tools easily.
It also benefits back NVDIA

OPen Source can also support it, like tensor flow of google.

4

Intel Python uses MKL. You have to pay to use MKL.

5

I think it is unrealistic to expect support for CUDA 8 features in PyCUDA at this time, given that CUDA 8 hasn’t even been finalized yet (the final version is expected this month, from what I understand).