search

microsoft / DirectML

DirectML is a high-performance, hardware-accelerated DirectX 12 library for machine learning. DirectML provides GPU acceleration for common machine learning tasks across a broad range of supported hardware and drivers, including all DirectX 12-capable GPUs from vendors such as AMD, Intel, NVIDIA, and Qualcomm.
MIT License
2.11k stars 284 forks source link

why DirectML isn't being further promoted #575

Open muyu66 opened 3 months ago

muyu66 commented 3 months ago

I'm very puzzled why DirectML isn't being further promoted. I find the idea of using CPUs to accelerate machine learning to be extremely groundbreaking. Although I don't work at the deep end of machine learning, I wonder if GPU + CPU Crossfire could further speed up the training process? There are vast clusters of CPUs worldwide, leaving them idle rather than using them for training models seems like a loss. With graphics card prices generally soaring, if CPUs could provide acceleration comparable to GPUs, it might just be a form of technological egalitarianism.

uniartisan commented 2 months ago

I have the same issue, it's a good project and should be promoted further, like updating to a newer version of pytorch

jk4e commented 1 month ago

I'll go along with that. I wonder if there will be more or if the project is already dead/suspended. Too bad for ML developers who want or need to work with Windows as their operating system. A faster, more active and continuous development of such tools and libraries would be very desirable, especially since the possibilities under Windows are lagging behind the other operating systems and ML and AI are so much in the spotlight.

hez2010 commented 1 month ago

It's waiting for the Microsoft Build 2024 announcement. Now a new version of torch-directml has been released: https://pypi.org/project/torch-directml/ Including both GPU and NPU support. For details, see https://blogs.windows.com/windowsdeveloper/2024/05/21/unlock-a-new-era-of-innovation-with-windows-copilot-runtime-and-copilot-pcs/

PyTorch is now natively supported on Windows with DirectML
We know that a lot of developers do their PyTorch development on Windows. So, we’re thrilled to announce that Windows now natively supports PyTorch through DirectML. Native PyTorch support means that thousands of Hugging Face models will just work on Windows. Not just that – we’re collaborating with Nvidia to scale these development workflows to over 100M RTX AI GPUs.
PyTorch support on GPUs is available starting today, with NPU support coming soon.