8000 Enable CUDA 12.8.0, Disable CUDA 12.4 · Issue #145570 · pytorch/pytorch · GitHub
[go: up one dir, main page]
Skip to content

Enable CUDA 12.8.0, Disable CUDA 12.4 #145570

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
24 of 26 tasks
tinglvv opened this issue Jan 24, 2025 · 13 comments
Open
24 of 26 tasks

Enable CUDA 12.8.0, Disable CUDA 12.4 #145570

tinglvv opened this issue Jan 24, 2025 · 13 comments
Assignees
@tinglvv
Labels
module: cuda Related to torch.cuda, and CUDA support in general triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module

Comments

@tinglvv
Copy link
Collaborator
tinglvv commented Jan 24, 2025 •
edited
Loading

🚀 The feature, motivation and pitch

CUDA 12.8.0 is out, adding to CI/CD.

Docker Images & Windows AMI Update

CD Update

CUDA 12.4 deprecation and CUDA 12.6 CI benchmarks

Alternatives

No response

Additional context

No response

cc @atalman @malfet @ptrblck @msaroufim @eqy @nWEIdia
This was referenced Jan 24, 2025
Closed
Closed
@cpuhrsch cpuhrsch added module: cuda Related to torch.cuda, and CUDA support in general triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module labels Jan 24, 2025
@johnnynunez
Copy link
Contributor

this will be amazing for test new blackwell.
I can say that it is compiling well: @malfet

@tinglvv tinglvv mentioned this issue Jan 24, 2025
Closed
@johnnynunez
Copy link
Contributor

@tinglvv new cudnn and tensorrt is out

@ofirkris
Copy link

Can we get 12.8 support and 2.7 in nightly?

@johnnynunez
Copy link
Contributor
johnnynunez commented Jan 25, 2025 •
edited
Loading

Can we get 12.8 support and 2.7 in nightly?

They are working on that:
pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

@johnnynunez johnnynunez mentioned this issue Jan 26, 2025
Closed
@ofirkris
Copy link

Can we get 12.8 support and 2.7 in nightly?

They are working on that: pip3 install --pre torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/cu128

ERROR: Could not find a version that satisfies the requirement torch (from versions: none) ERROR: No matching distribution found for torch

@johnnynunez
Copy link
Contributor

@ofirkris it will be available in the following days

@tinglvv tinglvv mentioned this issue Jan 27, 2025
8000 Closed
pytorchmergebot pushed a commit that referenced this issue Jan 27, 2025
@tinglvv @pytorchmergebot
Add CUDA 12.8 installation and manylinux-cuda12.8 (#145567)
93dd6bc 
Breaking #145557 into two parts.
Need to have manylinux-cuda12.8 in order to build magma.

Issue: #145570

Pull Request resolved: #145567
Approved by: https://github.com/nWEIdia, https://github.com/atalman
This was referenced Jan 27, 2025
Closed
Closed
nWEIdia pushed a commit to nWEIdia/pytorch that referenced this issue Jan 27, 2025
@tinglvv @nWEIdia
Add CUDA 12.8 installation and manylinux-cuda12.8 (pytorch#145567)
a3f598f 
Breaking pytorch#145557 into two parts.
Need to have manylinux-cuda12.8 in order to build magma.

Issue: pytorch#145570

Pull Request resolved: pytorch#145567
Approved by: https://github.com/nWEIdia, https://github.com/atalman
@tinglvv tinglvv mentioned this issue Jan 27, 2025
Closed
pytorchmergebot pushed a commit that referenced this issue Jan 29, 2025
@tinglvv @pytorchmergebot
Add CUDA 12.8 libtorch image (#145789)
f4ca989 
#145570

Builds 12.8 libtorch docker/deprecate 12.1 meanwhile

Pull Request resolved: #145789
Approved by: https://github.com/nWEIdia, https://github.com/atalman
pytorchmergebot pushed a commit that referenced this issue Jan 29, 2025
@tinglvv 8000 @pytorchmergebot
Add magma cuda build 12.8 (#145765)
354fe48 
#145570

Pull Request resolved: #145765
Approved by: https://github.com/malfet
@tinglvv tinglvv mentioned this issue Jan 30, 2025
Closed
pytorchmergebot pushed a commit that referenced this issue Jan 31, 2025
@tinglvv @pytorchmergebot
Use Magma-cuda 12.8 for libtorch (#146019)
08d8812 
#145570

Build failure for libtorch wheel
`CUDAContext.cpp:(.text+0x157): additional relocation overflows omitted from the output
/usr/bin/ld: failed to convert GOTPCREL relocation; relink with --no-relax
collect2: error: ld returned 1 exit status`

Unsure if this is related, fixing as a start
Pull Request resolved: #146019
Approved by: https://github.com/eqy
pytorchmergebot pushed a commit that referenced this issue Jan 31, 2025
@tinglvv @pytorchmergebot
Add CUDA 12.8 manywheel x86 Builds to Binaries Matrix (#145792)
9232355 
#145570

Adding cuda 12.8.0 x86 builds first
Pull Request resolved: #145792
Approved by: https://github.com/nWEIdia, https://github.com/malfet, https://github.com/atalman
@tinglvv tinglvv mentioned this issue Jan 31, 2025
Closed
@gdubroeucq
Copy link

waiting for the windows build as well :)

@tinglvv tinglvv mentioned this issue Feb 2, 2025
Merged
@tinglvv
Copy link
Collaborator Author
tinglvv commented Feb 3, 2025 •
edited
Loading

Nightly manywheel torch builds are available https://download.pytorch.org/whl/nightly/cu128/torch
Windows build PRs are in the process, should be the week after as the AMI takes a week to build.

This was referenced Feb 3, 2025
Merged
Closed
@tinglvv tinglvv self-assigned this Feb 3, 2025
Raymo111 pushed a commit that referenced this issue Feb 20, 2025
@tinglvv @Raymo111
windows Magma build for cu128 (#146653)
d97c50c 
#145570

removing `.ci/pytorch/windows/internal/cuda_install.bat` as it is a duplicate with` .github/scripts/windows/cuda_install.bat`. The later one is the one in use - https://github.com/pytorch/pytorch/pull/146653/files#diff-613791f266f2f7b81148ca8f447b0cd6c6544f824f5f46a78a2794006c78957bR8

Pull Request resolved: #146653
Approved by: https://github.com/atalman
Raymo111 pushed a commit that referenced this issue Feb 20, 2025
@tinglvv @atalman @Raymo111
windows Magma build for cu128 (#146653)
abbbbaf 
#145570

removing `.ci/pytorch/windows/internal/cuda_install.bat` as it is a duplicate with` .github/scripts/windows/cuda_install.bat`. The later one is the one in use - https://github.com/pytorch/pytorch/pull/146653/files#diff-613791f266f2f7b81148ca8f447b0cd6c6544f824f5f46a78a2794006c78957bR8

Pull Request resolved: #146653
Approved by: https://github.com/atalman

Co-authored-by: atalman <atalman@fb.com>
Raymo111 pushed a commit that referenced this issue Feb 20, 2025
@tinglvv @atalman @Raymo111
windows Magma build for cu128 (#146653)
71db602 
#145570

removing `.ci/pytorch/windows/internal/cuda_install.bat` as it is a duplicate with` .github/scripts/windows/cuda_install.bat`. The later one is the one in use - https://github.com/pytorch/pytorch/pull/146653/files#diff-613791f266f2f7b81148ca8f447b0cd6c6544f824f5f46a78a2794006c78957bR8

Pull Request resolved: #146653
Approved by: https://github.com/atalman

Co-authored-by: atalman <atalman@fb.com>
Raymo111 pushed a commit that referenced this issue Feb 20, 2025
@tinglvv @Raymo111
Add CUDA 12.8 windows nightly build (#147037)
232595f 
#145570

windows AMI is deployed to prod today, prepping the windows cuda 12.8 build

Pull Request resolved: #147037
Approved by: https://github.com/atalman
pytorchmergebot pushed a commit that referenced this issue Feb 21, 2025
@tinglvv @pytorchmergebot
Add libtorch nightly build for CUDA 12.8 (#146265)
fe100c3 
Try removing sm50 and sm60 to shrink binary size, and resolve the ld --relink error

"Architecture support for Maxwell, Pascal, and Volta is considered feature-complete and will be frozen in an upcoming release." from 12.8 release note.

Also updating the runner for cuda 12.8 test to g4dn (T4, sm75) due to the drop in sm50/60 support.

#145570

Pull Request resolved: #146265
Approved by: https://github.com/atalman
@tinglvv tinglvv mentioned this issue Feb 21, 2025
Closed
@jeffrey-cochran jeffrey-cochran mentioned this issue Feb 22, 2025
Open
@Skylion007
Copy link
Collaborator

I really think we should try to keep the lower SM arches in the 12.8 build while they are supported. We just need to fix the ld --relink script operation for libtorch, and we are unblocked from that, right? That likely involves passing the link script provided by nvcc to it?

pytorch-bot bot pushed a commit that referenced this issue Feb 24, 2025
@tinglvv @Ryo-not-rio
Add CUDA 12.8 windows nightly build (#147037)
4c9980b 
#145570

windows AMI is deployed to prod today, prepping the windows cuda 12.8 build

Pull Request resolved: #147037
Approved by: https://github.com/atalman
pytorch-bot bot pushed a commit that referenced this issue Feb 24, 2025
@tinglvv @Ryo-not-rio
Add libtorch nightly build for CUDA 12.8 (#146265)
40cb3bd 
Try removing sm50 and sm60 to shrink binary size, and resolve the ld --relink error

"Architecture support for Maxwell, Pascal, and Volta is considered feature-complete and will be frozen in an upcoming release." from 12.8 release note.

Also updating the runner for cuda 12.8 test to g4dn (T4, sm75) due to the drop in sm50/60 support.

#145570

Pull Request resolved: #146265
Approved by: https://github.com/atalman
@tinglvv tinglvv mentioned this issue Feb 25, 2025
Merged
@tinglvv tinglvv changed the title Enable CUDA 12.8.0 Enable CUDA 12.8.0, Disable CUDA 12.4 Feb 25, 2025
@PandaWei
Copy link

when release the version for cu128？

atalman pushed a commit to pytorch/test-infra that referenced this issue Feb 26, 2025
@tinglvv
Disable CUDA 12.4 nightly build (#6333)
2a18211 
pytorch/pytorch#145570

deprecate cuda 12.4 nightly build

cc @atalman
@FredGoo
Copy link
FredGoo commented Mar 2, 2025

when release the version for cu128？

Please be patient and kind

majing921201 pushed a commit to majing921201/pytorch that referenced this issue Mar 4, 2025
@tinglvv @atalman @majing921201
windows Magma build for cu128 (pytorch#146653)
17b4cb4 
pytorch#145570

removing `.ci/pytorch/windows/internal/cuda_install.bat` as it is a duplicate with` .github/scripts/windows/cuda_install.bat`. The later one is the one in use - https://github.com/pytorch/pytorch/pull/146653/files#diff-613791f266f2f7b81148ca8f447b0cd6c6544f824f5f46a78a2794006c78957bR8

Pull Request resolved: pytorch#146653
Approved by: https://github.com/atalman

Co-authored-by: atalman <atalman@fb.com>
majing921201 pushed a commit to majing921201/pytorch that referenced this issue Mar 4, 2025
@tinglvv @atalman @majing921201
windows Magma build for cu128 (pytorch#146653)
dbf1c61 
pytorch#145570

removing `.ci/pytorch/windows/internal/cuda_install.bat` as it is a duplicate with` .github/scripts/windows/cuda_install.bat`. The later one is the one in use - https://github.com/pytorch/pytorch/pull/146653/files#diff-613791f266f2f7b81148ca8f447b0cd6c6544f824f5f46a78a2794006c78957bR8

Pull Request resolved: pytorch#146653
Approved by: https://github.com/atalman

Co-authored-by: atalman <atalman@fb.com>
majing921201 pushed a commit to majing921201/pytorch that referenced this issue Mar 4, 2025
@tinglvv @majing921201
Add CUDA 12.8 windows nightly build (pytorch#147037)
8044bf9 
pytorch#145570

windows AMI is deployed to prod today, prepping the windows cuda 12.8 build

Pull Request resolved: pytorch#147037
Approved by: https://github.com/atalman
majing921201 pushed a commit to majing921201/pytorch that referenced this issue Mar 4, 2025
@tinglvv @majing921201
Add libtorch nightly build for CUDA 12.8 (pytorch#146265)
e84a1fe 
Try removing sm50 and sm60 to shrink binary size, and resolve the ld --relink error

"Architecture support for Maxwell, Pascal, and Volta is considered feature-complete and will be frozen in an upcoming release." from 12.8 release note.

Also updating the runner for cuda 12.8 test to g4dn (T4, sm75) due to the drop in sm50/60 support.

pytorch#145570

Pull Request resolved: pytorch#146265
Approved by: https://github.com/atalman
pytorchmergebot pushed a commit that referenced this issue Mar 5, 2025
@tinglvv @pytorchmergebot
Deprecate sm70 for cuda 12.8 binary (#147607)
9dd46a9 
follow up for https://github.com/pytorch/pytorch/pull/146265/files, dropping sm_70 as well, since "Architecture support for Maxwell, Pascal, and Volta is considered feature-complete and will be frozen in an upcoming release."

#145570

Pull Request resolved: #147607
Approved by: https://github.com/atalman
This was referenced Mar 5, 2025
Closed
Closed
Closed
pytorchmergebot pushed a commit that referenced this issue Mar 7, 2025
@tinglvv @atalman @pytorchmergebot
[CI][CUDA] Move away from cuda12.4, Add cuda12.6 eager CI tests (#148602
a0bc6d8 
)

#145570

breaking #140793 into eager and inductor benchmarks to unblock

Pull Request resolved: #148602
Approved by: https://github.com/atalman, https://github.com/malfet

Co-authored-by: atalman <atalman@fb.com>
pytorchmergebot pushed a commit that referenced this issue Mar 7, 2025
@tinglvv @atalman @pytorchmergebot
[CI] [inductor] Add cu126 inductor jobs and move away cu124 (#148612)
9769618 
#145570

breaking #140793 into eager and inductor benchmarks to unblock

Seems many inductor yml are added after initial change was prepared.

Pull Request resolved: #148612
Approved by: https://github.com/nWEIdia, https://github.com/atalman

Co-authored-by: atalman <atalman@fb.com>
pytorchmergebot pushed a commit that referenced this issue Mar 7, 2025
@tinglvv @pytorchmergebot
Remove Cuda 12.4 from nightly Binaries (#148625)
1239176 
#145570

removes cuda 12.4 nightly builds

Pull Request resolved: #148625
Approved by: https://github.com/atalman
@Skylion007
Copy link
Collaborator

Cudnn 9.8.0 just dropped: https://pypi.org/project/nvidia-cudnn-cu12/ Might be good to include it in this release given all the Blackwell performance optimizations that were made to it, especially regarding SDPA and given FlashAttention4 is still a ways off.

@tinglvv tinglvv mentioned this issue Mar 10, 2025
Closed
pytorchmergebot pushed a commit that referenced this issue Mar 10, 2025
@tinglvv @pytorchmergebot
Remove 12.4 x86 builds and 12.6 sbsa builds from nightly (#148895)
2a1eeae 
#145570

redo #148625

Pull Request resolved: #148895
Approved by: https://github.com/atalman
@tinglvv tinglvv mentioned this issue May 28, 2025
Open
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@tinglvv tinglvv

Labels
module: cuda Related to torch.cuda, and CUDA support in general triaged This issue has been looked at a team member, and triaged and prioritized into an appropriate module
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
8 participants
@cpuhrsch @Skylion007 @PandaWei @FredGoo @ofirkris @gdubroeucq @johnnynunez @tinglvv
0