-
Notifications
You must be signed in to change notification settings - Fork 6.8k
[BACKPORT]Enable CUDA 11.0 on nightly + CUDA 11.2 on pip (#19295)(#19764) #19930
Conversation
Hey @access2rohit , Thanks for submitting the PR
CI supported jobs: [windows-gpu, windows-cpu, clang, website, unix-gpu, unix-cpu, edge, miscellaneous, centos-gpu, centos-cpu, sanity] Note: |
@waytrue17 @leezu Can you review ? |
e7d5935
to
bbe179e
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also, since we need to support the last 2 major versions (11.x and 10.x) we should not remove support for 10.0 in this PR. Rather, can we add 11.1 and 11.2 and leave 10.0 intact? Thanks.
Remove CUDA 9.2 and CUDA 10.0
8b1440a
to
8f86404
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM, thanks!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looking good! we can try this pr on a duplicate cd pipeline to verify this works before merging.
blocked on unix-GPU failing due to error
@Zha0q1 @josephevans are looking into this issue. |
the base image |
Checked https://hub.docker.com/r/nvidia/cuda |
apache#19764) (apache#19930) * Enable CUDA 11.0 on nightly development builds (apache#19295) Remove CUDA 9.2 and CUDA 10.0 * [PIP] add build variant for cuda 11.2 (apache#19764) * adding ci docker files for cu111 and cu112 * removing previous CUDA make versions and adding support for cuda11.2 Co-authored-by: waytrue17 <52505574+waytrue17@users.noreply.github.com> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu>
apache#19764) (apache#19930) * Enable CUDA 11.0 on nightly development builds (apache#19295) Remove CUDA 9.2 and CUDA 10.0 * [PIP] add build variant for cuda 11.2 (apache#19764) * adding ci docker files for cu111 and cu112 * removing previous CUDA make versions and adding support for cuda11.2 Co-authored-by: waytrue17 <52505574+waytrue17@users.noreply.github.com> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu>
apache#19764) (apache#19930) * Enable CUDA 11.0 on nightly development builds (apache#19295) Remove CUDA 9.2 and CUDA 10.0 * [PIP] add build variant for cuda 11.2 (apache#19764) * adding ci docker files for cu111 and cu112 * removing previous CUDA make versions and adding support for cuda11.2 Co-authored-by: waytrue17 <52505574+waytrue17@users.noreply.github.com> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu>
…20015) * [BACKPORT]Enable CUDA 11.0 on nightly + CUDA 11.2 on pip (#19295)(#19764) (#19930) * Enable CUDA 11.0 on nightly development builds (#19295) Remove CUDA 9.2 and CUDA 10.0 * [PIP] add build variant for cuda 11.2 (#19764) * adding ci docker files for cu111 and cu112 * removing previous CUDA make versions and adding support for cuda11.2 Co-authored-by: waytrue17 <52505574+waytrue17@users.noreply.github.com> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> * [FEATURE]Migrating all CD pipelines to Ninja build + fix cu112 CD pipeline (#19974) * migrating cd builds to ninja + removing static links to nvidia libs and leagacy cuda versions * installing NCCL manually for cuda11.2 container * set MSHADOW_USE_CUDNN=1 in CMakelists of mshadow to build properly for CUDNN support * adding coverage to cd requirements file to fix cu100, cu101 and cu102 tests * updating cd_test containers to ubuntu 18 * adding cmake config for linux native and adding USE_KV_STORE in linux_cpu * updating zmq builds to statically link to libmxnet.so * updating toolchains for r, clang and llvm for ubuntu18. OpenBlas Static link for 'distribution' build type only. Fix caffe build to use openCV 3. Remove leagacy Clang 3.9 from CI * fix versions for pip install in ubuntu_core_sh add new search path for cuDNN * finxing cudnn link problem for CUDA<=11.0 * adding library paths for libjpegturbo and lapack to fix failing CI on ubuntu 18 images * removing ASAN integration test from miscellaneous CI as its not required * fix lapack path for gpu builds * correctly installing libjpegturbo for ubuntu 18 * updating docker images of r,jekyll,julia etc test containers+ fix java version to 8 * installing libomp.so * removing debug test as its not required. Code clean-up * adding alternate URL source for MNIST dataset as original website is down * skipping flaky tests issue tracked #20011 Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> * update cudnn from 7 to 8 for cu102 (#19506) * update cudnn from 7 to 8 for cu102 (#19522) * downloading MNIST dataset from alternate URL (#20014) Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> * fixing CI issue with v1.8.x * addressing review comments Co-authored-by: waytrue17 <52505574+waytrue17@users.noreply.github.com> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> Co-authored-by: Manu Seth <22492939+mseth10@users.noreply.github.com>
…pache#20015) * [BACKPORT]Enable CUDA 11.0 on nightly + CUDA 11.2 on pip (apache#19295)(apache#19764) (apache#19930) * Enable CUDA 11.0 on nightly development builds (apache#19295) Remove CUDA 9.2 and CUDA 10.0 * [PIP] add build variant for cuda 11.2 (apache#19764) * adding ci docker files for cu111 and cu112 * removing previous CUDA make versions and adding support for cuda11.2 Co-authored-by: waytrue17 <52505574+waytrue17@users.noreply.github.com> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> * [FEATURE]Migrating all CD pipelines to Ninja build + fix cu112 CD pipeline (apache#19974) * migrating cd builds to ninja + removing static links to nvidia libs and leagacy cuda versions * installing NCCL manually for cuda11.2 container * set MSHADOW_USE_CUDNN=1 in CMakelists of mshadow to build properly for CUDNN support * adding coverage to cd requirements file to fix cu100, cu101 and cu102 tests * updating cd_test containers to ubuntu 18 * adding cmake config for linux native and adding USE_KV_STORE in linux_cpu * updating zmq builds to statically link to libmxnet.so * updating toolchains for r, clang and llvm for ubuntu18. OpenBlas Static link for 'distribution' build type only. Fix caffe build to use openCV 3. Remove leagacy Clang 3.9 from CI * fix versions for pip install in ubuntu_core_sh add new search path for cuDNN * finxing cudnn link problem for CUDA<=11.0 * adding library paths for libjpegturbo and lapack to fix failing CI on ubuntu 18 images * removing ASAN integration test from miscellaneous CI as its not required * fix lapack path for gpu builds * correctly installing libjpegturbo for ubuntu 18 * updating docker images of r,jekyll,julia etc test containers+ fix java version to 8 * installing libomp.so * removing debug test as its not required. Code clean-up * adding alternate URL source for MNIST dataset as original website is down * skipping flaky tests issue tracked apache#20011 Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> * update cudnn from 7 to 8 for cu102 (apache#19506) * update cudnn from 7 to 8 for cu102 (apache#19522) * downloading MNIST dataset from alternate URL (apache#20014) Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> * fixing CI issue with v1.8.x * addressing review comments Co-authored-by: waytrue17 <52505574+waytrue17@users.noreply.github.com> Co-authored-by: Sheng Zha <szha@users.noreply.github.com> Co-authored-by: Rohit Kumar Srivastava <srivastava.141@buckeyemail.osu.edu> Co-authored-by: Manu Seth <22492939+mseth10@users.noreply.github.com>
Remove CUDA 9.x add CUDA 11.2 support
Backport #19295, #19764 as a part of effort #19911