PyTorch Foundation Formed By Meta, AMD, NVIDIA, & Others To Advance AI

[phoronix]

Phoronix: PyTorch Foundation Formed By Meta, AMD, NVIDIA, & Others To Advance AI

The PyTorch Foundation has been formed under the Linux Foundation umbrella by Meta (Facebook), AMD, AWS, Google Cloud, Meta, Microsoft, and NVIDIA to focus on providing the leading-edge AI framework...

https://www.phoronix.com/news/PyTorch-Foundation

[ddriver]

I think meta deserves another mention

[Klassic Six]

"Cool demo to turn your photo into anime style"

Wow! Great achievement guys, congrats!

[Setif]

Meta (Facebook), AMD, AWS, Google Cloud, Meta, Microsoft, and NVIDIA

You missed Alphabet (Google), AWS (Amazon) and Azure (Microsoft)

[peterdk]

I am familiar with Tensorflow, and did hear about Pytorch. Are these related, or 'competitors'? Is Pytorch now becoming the standard? Or is still just one of the options?

[JellyBrain]

I am familiar with Tensorflow, and did hear about Pytorch. Are these related, or 'competitors'? Is Pytorch now becoming the standard? Or is still just one of the options?

Tensorflow is still very much alive (got new commit on GitHub 3 mins ago).
Yep they are still kinda of a competing libraries.

And it seems this foundation will have more focus on deep learning on the cloud, but still I am sure their upstream improvements will help all of us.

[brucethemoose]

Users have been asking for AMD PyTorch support for ages... especially in the Vega days, where the hardware was relatively well suited to it (unlike the RX 6000+ series).

I tried to the ROCM PyTorch branch working locally, and eventually just gave up.



Hence I don't believe AMD is focused on bringing PyTorch upport to consumer hardware. Which is a mistake, as Nvidia quickly learned that students playing with CUDA on their laptops/desktops go on to work with CUDA as adults.

[coder]

Tensorflow is still very much alive (got new commit on GitHub 3 mins ago).
Yep they are still kinda of a competing libraries.

Since Google's TensorFlow couldn't kill off PyTorch, Google's interest is probably to try preventing their platforms and devices from being significantly disadvantaged by any developments in PyTorch.

[coder]

Users have been asking for AMD PyTorch support for ages... especially in the Vega days, where the hardware was relatively well suited to it (unlike the RX 6000+ series).

Why do you say that? Vega did introduce packed arithmetic, but RDNA also has it. I guess the biggest problem with RDNA is that lacks the matrix extensions found in CDNA. You could also cite memory bandwidth, but Infinity Cache can be useful for that.

[ET3D]

The upcoming RDNA 3 is said to have matrix extensions, so it seems that AMD does intend its consumer products to run AI. Hopefully this and being part of PyTorch means that AMD is trying to be serious about AI.

Still, based on its track record, I don't expect AMD to provide anything easy to use with wide support in the near future. I hope that AMD proves me wrong.