Google's Tensorflow + Intel = oneDNN API

[tech.monk]

Intel and Google are working together on Tensorflow and positive news in the market.


Finally another replacement for costly Nvidia GPUs in deep learning segment is available.

https://venturebeat.com/2022/05/31/tensorflow-now-defaults-to-intel-onednn-ai-optimizations/

These days PyTorch is more popular than Tensorflow, still this will be good move (assumed Intel GPUs will be available for cheaper).

 

[vishalrao]

Pytorch also seems to be in active development with added support for platforms like apple silicon etc, at least in the nightly builds...

 

[tech.monk]

Pytorch also seems to be in active development with added support for platforms like apple silicon etc, at least in the nightly builds...

This is true and inclined to get into it but at times few tensorflow features - one has to go through the pain of integrating them into PyTorch. Right now this is the deal breaker point on integration for me.

Rest Pytorch is going in right direction on bringing support of AMD and Nvidia (won't comment on compatibility)

 

[vishalrao]

Kool... Yup I don't do this stuff I was just commenting from what I saw in a YouTube video about running pytorch nightly build which recently added apple silicon support.

I would like to dabble a little bit into this area someday soon and I currently have an rx580 might upgrade to a newer AMD GPU (currently despise Nvidia) so hoping team red gets some better platform support just enough so I can play around and explore.

 

[tech.monk]

PyTorch will run on RX 580 with AMD ROCm. Not sure of apple silicon.

Let me know your findings.

Thanks