Although the voting has concluded, I was able to repeat my testing on NVIDIA's
internal CI, so my late vote for the record is '+1 (binding)':
My testing involved taking the 1.8.0.rc2 codebase and adding the files needed
to build and run against my internal CI system. The value here is not in the
number of build configurations per se, but in the number of different GPU
architectures tested. All the gpu unittests under
./tests/python/{gpu,train,tensort} passed on machines with: A100, V100, TU102,
TU104, P100 and P40 GPUs. Also, all the cpu unittests under
./tests/python/unittest passed (on a machine with a V100).
The build was for ubuntu 18.04 with cuda 11.1 and cuDNN 8.0.4 using the cmake
command:
cmake -GNinja -S. -DCMAKE_BUILD_TYPE=Release -Bbuild \
-DUSE_MKLDNN=OFF -DMXNET_CUDA_ARCH="6.0 6.1 7.0 7.5 8.0" \
-DUSE_CUDA=ON -DUSE_CUDNN=1 \
-DUSE_NCCL=ON \
-DUSE_DIST_KVSTORE=ON \
-DONNX_NAMESPACE=onnx \
-DUSE_NVTX=1 \
-DUSE_INTGEMM=OFF \
-DUSE_TENSORRT=1
On 2020/10/30 03:08:36, "Skalicky, Sam" <sska...@amazon.com.INVALID> wrote:
> Dear MXNet community,
>
> This is the vote to release Apache MXNet (incubating) version 1.8.0. Voting
> will start October 29, 20:10:00 PT and close on November 1, 20:10:00 PT.
>
> Link to release notes:
> https://cwiki.apache.org/confluence/display/MXNET/1.8.0+Release+Notes
>
> Link to release candidate:
> https://github.com/apache/incubator-mxnet/releases/tag/1.8.0.rc2
>
> Link to source and signatures on apache dist server:
> https://dist.apache.org/repos/dist/dev/incubator/mxnet/1.8.0.rc2
>
> Please remember to TEST first before voting accordingly:
> +1 = approve
> +0 = no opinion
> -1 = disapprove (provide reason)
>
> Here are the changes comparing to 1.8.0.rc1:
>
> * Updated LICENSE for all submodules
> * Updated oneDNN to v1.6.5 for RNN bugfix
> * Fixed type issue for broadcast_like
> * Fix MKLDNN BatchNorm with even number of channels
> * CI enhancements
> * Fix SoftReLU fused operator numerical stability
> * More Extensions fixes
> * Changed intgemm to a submodule instead of fetch
> * Updated DISCLAIMER
> * Update cuDNN version
>
> Best regards,
> Sam Skalicky
>
>
>
