4 ms·
The insane complexity of these ML models is limited to the data being handled, billions of numbers in hundreds of tensors. The GPU-running compute kernels are s
by Const-me 3y ago
The insane complexity of these ML models is limited to the data being handled, billions of numbers in hundreds of tensors. The GPU-running compute kernels are surprisingly simple, they just called many times with different input tensors.
This makes these compute kernels relatively portable across GPU APIs and languages.
I think on Windows, Direct3D is often a better tech choice for GPGPU. (1) It runs on GPUs of all vendors (2) Direct3D is an essential Windows component; the OS even composes windows on desktop with D3D11. Unlike CUDA, D3D runtime libraries are already available in the OS by default, which simplifies software installation and support.