posts/arc-a770-testing/part-1/

[utterances-bot]

Christian Mills - Testing Intel’s Arc A770 GPU for Deep Learning Pt. 1

I tested inference performance with OpenVINO and DirectML on the A770 and attempted to train models using PyTorch-DirectML.

https://christianjmills.com/posts/arc-a770-testing/part-1/

[hshen14]

Does CPU support FP16 with HW instruction?

[cj-mills]

Hi @hshen14, You can check out how OpenVINO handles different floating-point data types for CPU inference in their documentation at the link below.

• CPU Device - Floating Point Data Types Specifics

[paradoxical-gentoo-user]

@cj-mills would you recommend me to buy intel arc a770 and does it compile all sorts of models i mainly am looking to buy arc a770 to compile different types of models

[paradoxical-gentoo-user]

and yes i am okay with having to do little patching and having to hunt down bugs i am okay if the experience is a bit rough and not plug and play

[cj-mills]

Hi @paradoxical-gentoo-user, By compiling models, are you referring to using Keras? If so, I primarily work with PyTorch and generally avoid TensorFlow.

Intel adds support for PyTorch and TensorFlow with the extensions linked below.

• intel-extension-for-pytorch
• intel-extension-for-tensorflow

I tested the PyTorch extension on Ubuntu last December and did not consider it good enough to make a tutorial yet. There is a more recent release, so maybe that's changed.

I have not tried the TensorFlow extension, so I can't speak to its model support. I know some have run Stable Diffusion with it, though.

[Danyal-sab]

@cj-mills Did you proceed to run more test on Arc 770? do you have4 any new update?

[cj-mills]

@Danyal-sab I have not revisited it recently. When I last tested it in December, I felt it needed more time to mature. There have been a few updates for Intel's PyTorch extension and the PyTorch with DirectML package, so I plan to give it another shot soon. I'll make a follow-up to this post with my new findings. If it works reasonably well, I'll make a tutorial for it.

It will be at least a week before I can get back to testing the Arc card.

[Danyal-sab]

@cj-mills Thanks mate, looking forward to read it.

[Danyal-sab]

@cj-mills did you get any interesting result? would be grateful if you share your experience.

[cj-mills]

@Danyal-sab I'm literally writing the follow-up post right now. You can get a preview of it in my comments on the fast.ai forums.

[Danyal-sab]

@cj-mills Thanks for your great work and sharing it.

[vishwanathchar]

@Christian Mills, Could you please help me to set Intel Arc A370 graphic card to execute the Machine learning algorithms? I'm trying to find it on Google, but I failed to find the exact steps for the Windows machine.

@cj-mills

[cj-mills]

Hi @vishwanathchar,

I am unsure what you are looking for, specifically. However, you can use PyTorch on Windows through Windows Subsystem for Linux (WSL).

You can follow the tutorial on the Ubuntu website to install Ubuntu on WSL2.

• Install Ubuntu on WSL2

From there, you can follow the same setup steps in my tutorial:

• Getting Started with Intel’s PyTorch Extension for Arc GPUs on Ubuntu

That said, I recommend using native Ubuntu instead of WSL2 if possible. There is a significant performance hit from using WSL2 for deep learning tasks.

• Install Ubuntu with a Bootable USB Stick