[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: introduction



Hi Brian,

On 2024-07-19 23:57, Brian DeRocher wrote:
> I've been taking notes here https://hackmd.io/@openbrian/BJccdIuOR ; If
> anyone wants to make contributions, just let me know.  Or, I can move
> this to the wiki on Salsa.

Having updates to the wiki would be great. I'm happy to open it up but I
need to figure out how to set the appropriate permissions in
Salsa/GitLab first. I won't get to that today, but I'll try to get that
done tomorrow.

By the way, two more things came to mind:

First, if you have an older computer with a consumer mainboard around,
you could also sidestep obstacle (1) by trying the 7900 there.

I don't remember if my failed experiments from April were already done
on the box with SuperMicro H13 + EPYC Genoa where my gfx1100 currently
lives. Compared to consumer devices, these offer a ton more
virtualization-related options, and the "bad address" failure could
theoretically also be a BIOS misconfiguration.

Second, the qemu-rocm-{build,run} scripts were superseded by the
rocm-qemu-{build,run} scripts here [1], which you can install as package
rocm-qemu-support from our team's APT repository [2].

These automate some of the steps listed on the wiki. However, this is
done pretty defensively: they abort on IOMMU devices other than VGA,
Display, Audio, so you'd have to comment that part out.

Best,
Christian

PS: I wonder though whether the wiki is the right place because it has
really poor discoverability (robots.txt bans all crawlers).


[1]: https://salsa.debian.org/rocm-team/rocm-dev-tools/-/tree/master/bin?ref_type=heads
[2]: https://apt.rocm.debian.net/


Reply to: