PyTorch 1.8 Released With AMD ROCm Binaries

[phoronix]

Phoronix: PyTorch 1.8 Released With AMD ROCm Binaries

PyTorch 1.8 was released on Thursday as the newest version of this widely-used machine learning library. Exciting many will be easier AMD Radeon ROCm support with Python wheels now provided for that Radeon Open eCosystem support...

http://www.phoronix.com/scan.php?page=news_item&px=PyTorch-1.8-Released

[boffo]

Finally official Radeon support

[Steffo]

The AMD support is really huge! I hope, this will be a game changer in the AI market!

[haydn-j]

Do you have the means to get some benchmarks against NVIDIA for the new ROCm binaries? This is exciting.

[zexelon]

Can we see some benchmarks on this? If its even remotely competitive with CUDA solutions (i.e. within say 80 - 90%) this will be a huge game changer as Radeon tends to be quite a bit cheaper for their performance.

[coder]

Can we see some benchmarks on this? If its even remotely competitive with CUDA solutions (i.e. within say 80 - 90%) this will be a huge game changer as Radeon tends to be quite a bit cheaper for their performance.

AFAIK, he lacks a MI100 and that's the only AMD GPU with Matrix Cores. Otherwise, your performance is going to be potentially quite a bit slower than any Nvidia GPU with Tensor cores.

I mean, you could always just use fp32, but then it's no longer a very realistic benchmark (other than showing things like relative driver overhead of each stack).

[coder]

So, let me get this straight, boffo and Steffo both registered their accounts in 2013 and both have nearly the same # of posts? Interesting...

[plinkyplonky]

I wasn't sure whether to go with Tensorflow or PyTorch. I hope that PyTorch supports AGPUs as well, like the laptop Zen+ CPUs with built-in GPU.

[coder]

I wasn't sure whether to go with Tensorflow or PyTorch.

Hmmm... Google vs. Facebook.

People used to say that Tensorflow had better training tools, but I don't know if that's still the case.

[boffo]

Hmmm... Google vs. Facebook.

People used to say that Tensorflow had better training tools, but I don't know if that's still the case.

You can use tensorboard with pytorch too.