Google's IREE To Demonstrate Machine Learning Via Vulkan With MLIR

[phoronix]

Phoronix: Google's IREE To Demonstrate Machine Learning Via Vulkan With MLIR

One of the new open-source compiler IR advancements of 2019 has been the Google/Tensorflow MLIR as the Multi-Level Intermediate Representation designed for machine learning models/frameworks. With Google's "IREE" project, MLIR can be accelerated by Vulkan and thus allowing machine learning via this high-performance graphics/compute API...

http://www.phoronix.com/scan.php?page=news_item&px=Google-IREE-ML-Vulkan

[shmerl]

This should help displacing CUDA lock-in.

[GruenSein]

This should help displacing CUDA lock-in.

Let's hope so. My current impressionis that there are multiple efforts to replace CUDA and none of them has gained traction or progressed beyond the conceptional stage.

[zxy_thf]

Let's hope so. My current impressionis that there are multiple efforts to replace CUDA and none of them has gained traction or progressed beyond the conceptional stage.

The main problem is they're way too late, and with inferior interface, e.g., the context creation of OpenCL is a nightmare.
However, the current ML platforms, on the other hand, are making CUDA irrelevant rather than replacing CUDA.

IMO this is a right direction, just like the Android is not replacing Windows but making Windows irrelevant on mobile systems.

[cb88]

Let's hope so. My current impressionis that there are multiple efforts to replace CUDA and none of them has gained traction or progressed beyond the conceptional stage.

Not really accurate since TensorFlow and many other CUDA applications have ports to AMD cards via HIP/HCC. It exists all you have to do is use it.

[GruenSein]

Not really accurate since TensorFlow and many other CUDA applications have ports to AMD cards via HIP/HCC. It exists all you have to do is use it.

I am aware of these HIP-ports. Since I don't use ML-frameworks I can just go by the documentation I have found online. To me, these seem to suggest that compatibility is limited, the installation process is less straight forward etc. There is a reason why most HPC developments use CUDA

[Steffo]

If this works, I would seriously consider to switch to AMD!