NVIDIA Talks Up Numba For GPGPU Computing With Python

[phoronix]

Phoronix: NVIDIA Talks Up Numba For GPGPU Computing With Python

Numba is designed to allow for high performance Python JIT-compiled code designing for C/C++ levels of performance while using LLVM for optimizations and allowing GPU offloading too. NVIDIA is promoting Numba in the context of CUDA...

http://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Python-Numba

[Guest]

From the link in the article:

1. Numba is 100% Open Source

Doesn't really count, because nvidia blocks development of opensource drivers by not providing enough documentation for their GPUs, and makes it impossible to reverse engineer proprietary drivers, because of their "security" features. Compared to AMD, which puts efforts itself into development of opensource drivers.

[extraymond]

From the link in the article:



Doesn't really count, because nvidia blocks development of opensource drivers by not providing enough documentation for their GPUs, and makes it impossible to reverse engineer proprietary drivers, because of their "security" features. Compared to AMD, which puts efforts itself into development of opensource drivers.

Numba is developed by continuum I believe, they're just integrating it with numba.
Numba had hsa support when apu was introduced, but no major news from hsa/rocm/opencl recently. Maybe bridgman can share us some plans?

[edoantonioco]

When is nvidia going to improve their opengl performance? It has improved lately on the windows driver, but now the linux performance is a bit (noticeable) behind

[thebear]

edoantonioco you do realize that this has absolutely nothing to do with graphics right?

[hoohoo]

I would like to see Intercal used in this manner. The 'comefrom' instruction could make GPU offload much more intuitive.

[caligula]

From the link in the article:



Doesn't really count, because nvidia blocks development of opensource drivers by not providing enough documentation for their GPUs, and makes it impossible to reverse engineer proprietary drivers, because of their "security" features. Compared to AMD, which puts efforts itself into development of opensource drivers.

And the users don't give a feck. CUDA still dominates the field.