CUDA-Python Reaches "GA" With NVIDIA CUDA 11.5 Release, __int128 Preview

[phoronix]

Phoronix: CUDA-Python Reaches "GA" With NVIDIA CUDA 11.5 Release, __int128 Preview

NVIDIA has made available CUDA 11.5 today as the latest version of their popular but proprietary compute stack/platform. Notable with CUDA 11.5 is that CUDA-Python has reached general availability status...

https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-CUDA-11.5

[hoohoo]

Python binding for CUDA. This is the first really exciting thing I have heard about in months.

[schmidtbag]

Python binding for CUDA. This is the first really exciting thing I have heard about in months.

Huh? It's been around for years.

[hoohoo]

Huh? It's been around for years.

I use CUDA with C++. Thought about it and I suppose I have never gone looking for a Python binding, So I thought this was something new.

[schmidtbag]

I use CUDA with C++. Thought about it and I suppose I have never gone looking for a Python binding, So I thought this was something new.

Haha nope. Not only has it been around for a while, but it's also quite polished. If you ever want to get into CUDA quickly, Python is great.
In fact, Nvidia's efforts with Python are to me a prime example of why CUDA reigns supreme in GPU compute - it's all so well-documented and implemented that anyone with intermedia programming experience can figure it out. They provide lots of clean and helpful examples too.

The OpenCL implementation, on the other hand, is basically just a wrapper to the C++ plugin and you're left to "figure it out", because most of the documentation is how you do things in C++ but not Python. At least, that was the case a few years ago when I first attempted it. Not sure if it got anywhere since then. With the direction Intel and AMD, maybe it got a little better. I'd much rather OpenCL because it is far more portable, but the documentation is so lacking that I would argue it's easier to just write your whole program in C++.

[hoohoo]

Haha nope. Not only has it been around for a while, but it's also quite polished. If you ever want to get into CUDA quickly, Python is great.
In fact, Nvidia's efforts with Python are to me a prime example of why CUDA reigns supreme in GPU compute - it's all so well-documented and implemented that anyone with intermedia programming experience can figure it out. They provide lots of clean and helpful examples too.

The OpenCL implementation, on the other hand, is basically just a wrapper to the C++ plugin and you're left to "figure it out", because most of the documentation is how you do things in C++ but not Python. At least, that was the case a few years ago when I first attempted it. Not sure if it got anywhere since then. With the direction Intel and AMD, maybe it got a little better. I'd much rather OpenCL because it is far more portable, but the documentation is so lacking that I would argue it's easier to just write your whole program in C++.

I read through a couple of the Python example programs and it looks quite friendly.