ZARZA We are Zarza, the prestigious firm behind major projects in information technology.
Listen "CUDA graph trees"
Episode Synopsis
CUDA graph trees are the internal implementation of CUDA graphs used in PT2 when you say mode="reduce-overhead". Their primary innovation is that they allow the reuse of memory across multiple CUDA graphs, as long as they form a tree structure of potential paths you can go down with the CUDA graph. This greatly reduced the memory usage of CUDA graphs in PT2. There are some operational implications to using CUDA graphs which are described in the podcast.
More episodes of the podcast PyTorch Developer Podcast
Compiler collectives
04/08/2024
TORCH_TRACE and tlparse
29/04/2024
Higher order operators
21/04/2024
Inductor - Post-grad FX passes
12/04/2024
Min-cut partitioner
17/03/2024
AOTInductor
02/03/2024
Tensor subclasses and PT2
24/02/2024
Compiled autograd
19/02/2024
PT2 extension points
05/02/2024
Inductor - Define-by-run IR
24/01/2024
In God we trust