Giter VIP home page Giter VIP logo

Comments (18)

jakirkham avatar jakirkham commented on June 12, 2024 2

Just a note here, we are currently waiting for CentOS 7 CUDA 11 images to be published. More details on the current status in this comment.

from docker-images.

isuruf avatar isuruf commented on June 12, 2024 1

CF packages work just fine on CentOS 7 even though built in CentOS 6.

from docker-images.

isuruf avatar isuruf commented on June 12, 2024 1

We set up two variants of docker images for CUDA 10.1 & 10.2 that is based on CentOS 7, and put them in production in conjunction with existing ones.

Sure, but you'll also need to add __glibc >=2.17 to all the recipes using CentOS 7 to make sure that users using older GLIBC will not get these new packages that don't work on their systems.

from docker-images.

leofang avatar leofang commented on June 12, 2024

cc: @jakirkham

from docker-images.

leofang avatar leofang commented on June 12, 2024

FYI, according to the Release Notes of CUDA 10.2, RHEL 6.x will be dropped in the upcoming CUDA 11:

Note that support for RHEL 6.x is deprecated and support will be dropped in the next release of CUDA.

from docker-images.

isuruf avatar isuruf commented on June 12, 2024

Where is cuTENSOR coming from?

We've had this discussion before and updating the base image just to accommodate a proprietary binary blob is no. There are still many people using CentOS 6.

from docker-images.

leofang avatar leofang commented on June 12, 2024

@isuruf

Where is cuTENSOR coming from?

Like cuDNN, cuTENSOR is available from NVIDIA Developer Program. @jjhelmus built the former in Anaconda's channel locally and then upload the package, and I verified that the same approach applies to cuTENSOR.

We've had this discussion before and updating the base image just to accommodate a binary blob is no. There are still many people using CentOS 6.

Unfortunately we don't get to choose when CUDA 11 is out (see my above #137 (comment)). The base won't be able to support it.

from docker-images.

leofang avatar leofang commented on June 12, 2024

One more thing: as of cuTENSOR 1.0.1 only CUDA 10.1 and 10.2 are supported, so a compromise could be that we only update the base for 10.1+

from docker-images.

leofang avatar leofang commented on June 12, 2024

One final note 😅: I am not saying this change is urgent/critical (to CuPy or any other CUDA-based packages). I'm simply bringing this issue to CF core devs' attention. This doesn't have to be addressed right now, but an update plan should be considered/revisited in the near future.

from docker-images.

isuruf avatar isuruf commented on June 12, 2024

Why do you want to update the whole conda-forge system to CentOS 7 just to accommodate a proprietary software? Why not just change the few feedstocks you need it for CentOS 7?

from docker-images.

leofang avatar leofang commented on June 12, 2024

Why not just change the few feedstocks you need it for CentOS 7?

@isuruf Could you elaborate on how to do this please?

from docker-images.

isuruf avatar isuruf commented on June 12, 2024

You can override docker_image variable in recipe/conda_build_config.yaml.

from docker-images.

leofang avatar leofang commented on June 12, 2024

But how do I know the package would be compatible with the rest of CF packages that also depend on cudatoolkit and other CUDA packages?

from docker-images.

leofang avatar leofang commented on June 12, 2024

Thanks for confirmation, @isuruf. It's good to know.

from docker-images.

leofang avatar leofang commented on June 12, 2024

Sorry @isuruf, one follow-up question:

CF packages work just fine on CentOS 7 even though built in CentOS 6.

Were you implying the converse wouldn't work? (Building on CentOS 7, running on CentOS 6)

from docker-images.

isuruf avatar isuruf commented on June 12, 2024

Mostly it would work because the conda compilers build against CentOS 6 even on CentOS 7. Problems arise when you use hacks like https://github.com/conda-forge/cupy-feedstock/blob/master/recipe/meta.yaml#L28-L30 and it would no longer be usable in CentOS 6. (Of course if your dependencies don't work in CentOS 6, the package wouldn't either)

from docker-images.

leofang avatar leofang commented on June 12, 2024

Sorry, @isuruf, I still feel it's necessary to revisit your comment:

Why do you want to update the whole conda-forge system to CentOS 7 just to accommodate a proprietary software?

It is not just "one" proprietary software. As I mentioned earlier, it is the entire CUDA Toolkit (and everything built on top of it). The current docker images simply will not work with the upcoming CUDA 11.

Perhaps I am being naive, but I wonder if the following intermediate solution is possible while we continue the discussion: We set up two variants of docker images for CUDA 10.1 & 10.2 that is based on CentOS 7, and put them in production in conjunction with existing ones. Once CUDA 11 is out, we can simply add the support (so that from CUDA 11+ we use CentOS 7 only).

from docker-images.

jakirkham avatar jakirkham commented on June 12, 2024

Newer CUDA images use CentOS 7 (or in the most recent CentOS 8) as needed. Also older CUDA images had CentOS 7 variants added in PR ( #167 )

from docker-images.

Related Issues (20)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.