@enkore can you please review?

Interesting phenomenon. Ice Lake has indeed already the extensions, but I cannot find information on Intel.com that its successor Tiger Lake has it as well. But I think we can assume it contains them, too. Apple's processors are also worth to be named on themselves given the widespread use. Thanks for the notes.

However, I think your approach of benchmark makes more sense on the long term: the user shall make a benchmark and use what is fastest. lscpu would be an alternative to identify the extensions, although this works not on all OS.

But maybe it makes sense to enable a benchmark that does not need the creation of a repo in advance, and then replace my whole adjustment about the architectures and their extensions by one or two sentences about how and why to make the benchmark? So, facilitating an initial benchmark as first step before creating the first borg repo on a system.

I can write an adjustment in the next few days.

There's "borg benchmark cpu" specifically for doing those tests. I don't know if it does, but a nice thing there would be to just plainly tell you which mode is quickest on your system, and associated caveats if any.

There's "borg benchmark cpu" specifically for doing those tests. I don't know if it does, but a nice thing there would be to just plainly tell you which mode is quickest on your system, and associated caveats if any.

My point was that "borg benchmark *" seems to need a repo that already exists before it can run. So, a user would need to create a repo, then do the benchmark, and then potentially delete the repo again to create a new one with a different primitive. Maybe this can be simplified in future versions? It would be definitely the better approach to just let the user know how to do the benchmark and then choose the primitive that is fastest.

"borg benchmark cpu" leads to the error "borg benchmark: error: argument : invalid choice: 'cpu' (choose from 'crud')" on my system, but I expect this is because I have currently no borg repo.

borg benchmark cpu is only in master branch and does not require a repo. it does pure algorithmic tests without I/O.

borg benchmark cpu is only in master branch and does not require a repo. it does pure algorithmic tests without I/O.

I was not aware of this :) In this case, it makes sense to already prepare the docs in master branch for that. I will delete all the architecture stuff and simply refer to how to use and interpret "borg benchmark cpu". I will use your example of #6666

You could keep the architectural stuff for a 1.2-maint PR.

Yes, for master, you can just refer to borg benchmark cpu.

security.rst is updated & "borg benchmark cpu" is introduced. You may verify my syntax with commands (such as "borg benchmark cpu") because my experience with readthedocs is limited.

thanks!