cloudhan

Results 191 comments of cloudhan
trafficstars

@byronyi that patch is only in nvidia's fork, any chance will they merge it into upstream tf2?

@hawkinsp Have you ever benchmarked it with more workers, e.g., 4 worker with 4 V100 with only 10Gbps Ethernet. I think then you can see the difference. And yes, we...

@sorawee I suggest use angle bracket. Potentially unify the generic syntax in other language...

As a non-Scheme/Racket programmer, or more specifically, a programmer comfort with C++ or python, I take this type of character usage as abusing. You can simply naming something without these...

You might consider #574 > So, in my opinion, sync should not fetch, should not have network activities involved. It should just automatically move all my working commit to a...

On some new architectures say sm_80, the very basic form `copy(...)`, that is, without specifying the copy atom, might generate cp.async instruction. So you need further safety guarantee. #1231

I don't think there will be "general consensus" which one is better. They might be just be trail and error results. What they don't mention in the paper (explicitly) is...

Simply enable CUPTI will cause `torch_cpu.dll` reference cudart symbols ``` > ninja.exe .\bin\torch_cpu.dll [1/1] Linking CXX shared library bin\torch_cpu.dll FAILED: bin/torch_cpu.dll lib/torch_cpu.lib cmd.exe /C "cd . && "C:\Program Files (x86)\Microsoft...

For my local build on linux: ``` $ ldd build/lib/libtorch_cpu.so | grep cuda libcudart.so.11.0 => /usr/local/cuda-11.1/lib64/libcudart.so.11.0 (0x00007fa9456cb000) ``` For official release 1.9.0: ``` $ ldd /home/guangyunhan/miniconda3/envs/py37/lib/python3.7/site-packages/torch/lib/libtorch_cpu.so | grep cuda libcudart-6d56b25a.so.11.0...

``` $ grep cudart -r build/third_party/tensorpipe build/third_party/tensorpipe/CMakeFiles/Export/share/cmake/Tensorpipe/TensorpipeTargets.cmake: INTERFACE_LINK_LIBRARIES "tensorpipe;/usr/local/cuda-11.1/lib64/libcudart.so" ``` I am pretty sure `libcudart` is coming from `tensorpipe`. I suspect that if we disable tensorpipe with kineto cupti enabled,...