Release 3.3.0 corresponding to NGC container 20.08
PyProf - PyTorch Profiler for GPUs
PyProf is a tool that profiles and analyzes the GPU performance of PyTorch models. PyProf aggregates kernel performance from Nsight Systems or NvProf. The tool can identify the layer that launched the kernel, identify the tensor dimensions and precision, perform forward-backward correlation, determine Tensor Core usage, and correlate the line in the user's code that launched a particular kernel.
What's New In 3.3.0
-
Capture PyTorch API information and data loading configuration
-
Added CUTLASS to the list of GEMM kernels
-
Added optional function stack tracing to NVTX markers. Enable with
pyprof.init(enable_function_stack=True)
Known Issues
-
This software only supports PyTorch 1.6.
-
Forward-Backward kernel correlation heuristics do not work correctly with PyTorch 1.6. Recommended work arounds include:
- Use with PyTorch 1.5
- Use DLProf in the 20.08 NGC PyTorch container
- Use the 20.03-py3 PyTorch NGC container
$ docker pull nvcr.io/nvidia/pytorch:20.03-py3