This announcement was a bit buried in the news, but soon, you will be able to use Tensor Cores in graphics shaders.
https://devblogs.microsoft.com/directx/enabling-neural-rendering-in-directx-cooperative-vector-support-coming-soon/
https://developer.nvidia.com/blog/nvidia-rtx-neural-rendering-introduces-next-era-of-ai-powered-graphics-innovation/
The API is super elegant, neural and non-neural stuff, a very impressive work of my colleagues, stay tuned.
https://devblogs.microsoft.com/directx/enabling-neural-rendering-in-directx-cooperative-vector-support-coming-soon/
https://developer.nvidia.com/blog/nvidia-rtx-neural-rendering-introduces-next-era-of-ai-powered-graphics-innovation/
The API is super elegant, neural and non-neural stuff, a very impressive work of my colleagues, stay tuned.
Comments
For a preview of what Tensor Cores do, check out WMMA API and instructions in the CUDA manual - much wider vectors and larger matrices, including smaller float/int types. But require a coherent/scalar matrix.
Not a dumb question.
float4x4 stuff: no
big matrix stuff like for ML: yes, but has to be coded that way.