Rebuild for CUDA 11.8 w/arch support

[regro-cf-autotick-bot]

This PR has been triggered in an effort to update cuda118.

Notes and instructions for merging this PR:

1. Please merge the PR only after the tests have passed.
2. Feel free to push to the bot's branch to update this PR if needed.

Please note that if you close this PR we presume that the feedstock has been rebuilt, so if you are going to perform the rebuild yourself don't close this PR until the your rebuild has been merged.

---

If this PR was opened in error or needs to be updated please add the bot-rerun label to this PR. The bot will close this PR and schedule another one. If you do not have permissions to add this label, you can use the phrase @conda-forge-admin, please rerun bot in a PR comment to have the conda-forge-admin add it for you.

_{This PR was created by the regro-cf-autotick-bot. The regro-cf-autotick-bot is a service to automatically track the dependency graph, migrate packages, and propose package version updates for conda-forge. Feel free to drop us a line if there are any issues! This PR was generated by https://github.com/regro/cf-scripts/actions/runs/6503670537, please use this URL for debugging.}

[conda-forge-webservices]

Hi! This is the friendly automated conda-forge-linting service.

I just wanted to let you know that I linted all conda-recipes in your PR (recipe) and found it was in an excellent condition.

[jakirkham]

FWIW CI appears to pass

[leshikus]

I believe a good thing to do is replacing cuda 11 with cuda 12 in a list of supported configurations

[jakirkham]

Well CUDA 12 drops Kepler. So that may be a consideration depending on the user base of hdk

Perhaps one CUDA 11 build is sufficient (either 11.2 or 11.8)

[alexbaden]

Dropping Cuda 11 or keeping just one version (11.8) is fine with me.

[jakirkham]

@h-vetinari , do you recall why this is marked as draft?

[h-vetinari]

It was originally part of #37, but since this blew up the CI matrix to such an extreme extent, the utility of 11.8 (over the compatible 11.2 builds) was not there compared to another 40% more CI jobs, especially as all those are quite long running.

But perhaps that calculus can be redone in case someone really needs the additional arches of 11.8

[jakirkham]

In that case, perhaps it is worth dropping CUDA 11.2 and adding CUDA 11.8. The latter still covers the same hardware that the former does

Should add the CUDA 11.2 images are planned for removal in May 2024. So we will want to start migrating off them before then

Edit: It's worth noting we went through a similar discussion in PyTorch and decided to add CUDA 11.8 & drop CUDA 11.2 ( conda-forge/pytorch-cpu-feedstock#195 ). Similarly TensorFlow added CUDA 11.8 ( conda-forge/tensorflow-feedstock#344 ) and plans to drop CUDA 11.2 ( conda-forge/tensorflow-feedstock#347 (comment) ). Think this trend will continue

[jakirkham]

^ @h-vetinari what do you think of the idea above?

[h-vetinari]

In that case, perhaps it is worth dropping CUDA 11.2 and adding CUDA 11.8. The latter still covers the same hardware that the former does

It supports the same hardware, but not the same drivers. To be honest I don't have good insight into the kind of consumers using conda-forge GPU packages and what hardware/drivers they're on, but we've so far been very conservative in dropping the lower end of the CUDA drivers, e.g. conda-forge/conda-forge-pinning-feedstock#1708 only got merged 4 months ago (after being open for 2 years...), and there was substantial discussion in arrow (e.g. here). @kkraus14 seemed to care quite a bit about supporting old drivers.

While I don't care very much about the exact lower bound, I think such a discussion should be had conda-forge wide (conda-forge/conda-forge.github.io#1981 IMO got closed wrongly by the migrator, its title is unresolved), and not on individual feedstocks (c.f. also conda-forge/conda-forge-pinning-feedstock#3491, where this was more widely announced because it was unavoidable for pytorch at the time).

Should add the CUDA 11.2 images are planned for removal in May 2024. So we will want to start migrating off them before then

That's one argument in favour of dropping <11.8, but it's not like we haven't used EOL images before. Another issue is that 11.8 is quite a ways away from parity with 11.2 in terms of covered feedstocks.

[jakirkham]

JFYI we are planning to drop CUDA 11.2 support in conda-forge soon

Please see this announcement and this issue ( conda-forge/conda-forge-pinning-feedstock#5339 ) for more details

The smoothest upgrade path would be to add CUDA 11.8. IOW completing this migration PR

Please let us know if you have any questions on next steps

[jakirkham]

JFYI CUDA 11.2 is now officially dropped