3 ms·
NB: Only decompression in implemented on the GPU (via HLSL compute shader). Compression is still done on the CPU.
by supernes 4y ago
NB: Only decompression in implemented on the GPU (via HLSL compute shader). Compression is still done on the CPU.
- singhrac 4y agoThis is likely related to Microsoft’s DirectStorage API: https://devblogs.microsoft.com/directx/directstorage-1-1-now-available/ https://devblogs.microsoft.com/directx/directstorage-1-1-now.... While this is mostly game-focused, I wonder if ML training libraries will take advantage of this, since transferring training examples is often a huge bottleneck.