Feat/add gpu vram

[garloff]

Preliminary:

• Spec adjustment done
• Implementation notes done (including GPU table)
• Code adjustments (parser, flavor generator, pretty name, ...) NOT yet done

Please check whether that incremental change is the small step that we agreed upon in the last team meeting.
It keeps the current scheme intact -- it's certification preserving, thus no new version required.

We may reconsider the flavor naming more fundamentally at some point in the future and maybe change direction from the systematic naming, but this is a major deviation and needs more work and preparation.

Related to #366

[cah-patrickthiem]

Looks good overall.
But I have got some things to mention anyway:

1. Minor thing, Nvidia is mostly written either with all capital letters NVIDIA or only the "N" is capital: Nvidia (e.g. https://de.wikipedia.org/wiki/Nvidia). I basically never see nVidia (funny thing is, Github marked this writing as false, Nvidia and NVIDIA not). So my suggestion would be to change the writing to "Nvidia" for all mentions.

2. keep in mind that there are still many Nvidia generations officially supported by Nvidia: see https://endoflife.date/nvidia-gpu . I would suggest to include this link in the document and to also include all GPUs from Nvidia which are at least listed with "Active Support", meaning we go down to Maxwell generation. We at C&H do have at least Turing (Nvidia T4) GPUs which are used by customers. Turing is even listed as "In Production" For AMD and Intel this does not apply since they have not that many generations out there for so long as Nvidia.

[mbuechse]

Interesting. I still know the old spelling nVidia from the 90s as well. It can be seen on old chips, for instance: https://de.m.wikipedia.org/wiki/Nvidia#/media/Datei%3A6600GT_GPU.jpg

[garloff]

Spelling fixed.
Feel free to add further nVidia^WNvidia generations with tables.
The scheme allows for them, no mysteries ...

[cah-patrickthiem]

Spelling fixed. Feel free to add further nVidia^WNvidia generations with tables. The scheme allows for them, no mysteries ...

Lets merge this PR here first and I will create another PR to fill up the table with more GPUs.

[cah-patrickthiem]

Looks good for me, LGTM.

[cah-patrickthiem]

For detailed information regarding why this PR was made: #546

[garloff]

Well, I still need to adjust the parser, generator, pretty-printer in the code to accept/generate the new syntax before merging.
I'm glad to have support before investing the work.

[cah-patrickthiem]

After Matthias commits, I again went through all documents and found some more things which should be changed.

[mbuechse]

@cah-patrickthiem Please perform the changes yourself. Kurt is on vacation, and he won't mind.

[garloff]

After Matthias commits, I again went through all documents and found some more things which should be changed.

You're speaking in mysteries, @cah-patrickthiem here.

• If you have minor improvements, just mention and commit them, like @mbuechse suggests.
• If you have major concerns, please bring them up and let's discuss.

I did change the wording in the meaning of h to say high frequency (rather than perf) for the CUs and high-bandwidth (rather than perf) for VRAM, so things should be clearer in the console output and on https://flavors.scs.community/
I hope that is appreciated.

From my perspective this is ready to be merged, but happy to see additional refinement if you have any ...

[cah-patrickthiem]

After Matthias commits, I again went through all documents and found some more things which should be changed.

You're speaking in mysteries, @cah-patrickthiem here.

• If you have minor improvements, just mention and commit them, like @mbuechse suggests.
• If you have major concerns, please bring them up and let's discuss.

I did change the wording in the meaning of h to say high frequency (rather than perf) for the CUs and high-bandwidth (rather than perf) for VRAM, so things should be clearer in the console output and on https://flavors.scs.community/ I hope that is appreciated.

From my perspective this is ready to be merged, but happy to see additional refinement if you have any ...

Sorry, usually I/we handle it like so, that the reviewer is not supposed to push anything to the PR to be reviewed, therefore I did not do this but planned to do it after Matthias told me its ok. You were just quicker.

Nevertheless, I also think we can merge it, yes.

[garloff]

Thanks, @cah-patrickthiem and @mbuechse to get this over the finish line!