KI don't think we can merge this @Dylan
Kwould break existing images
AKthey look for
-nvidiaKnot
-nvidia-legacy or -nvidia-openKwould require everyone that consumes akmods to update their pulls
Abelow escape
Awith shift
DI can change it from -nvidia-legacy to -nvidia
Kthat would be good
DI do need some help though with figuring out why the labels are making my arg not pass in
DI'm kind of stuck on that one last bit
J
J@Noel tell him this slide matches our website, nearly bullet for bullet too.
DI updated it
JOMG
Jthey're doing zstd:chunked because some of the RHEL AI images are freaking 50-100GB, LOL
Achecks out
Aand its probably like 20 files
Jlol he mentions that fedora declined the zstd:chunked proposal for this cycle.
N@j0rge Another talk from Eric Curtin and Dan Walsh up next in the main room.
NRamalama. New AI stuff!
NYeah, I got more info on that. It's because they need to support older docker versions which only support gzip. Thank you Ubuntu LTS for that 
JIt doesn't affect us, it'll be an easy thing to turn on once the builders go live, p5 handled it.
NHe is going to repropose it and the idea is they will store both gzip and zstd:chunked. Podman will be smart enough to pull the zstd Fedora image by default.
Jyeah that's what we're doing
NSo wait, we are already doing zstd chunked?
Ano
Abazzite testing has it on
NIn conjunction with rechunk?
Ayeah
NHeck yeah.
Aso if you pull a bazzite image with podman it should be extra extra fast
Anot rpm-ostree yet though I guess
NYeah, we will have to move to bootc to get the full benefit of zstd chunked.
NAFAIK
Aat least a lot of the work we're doing with rechunk fixes quirks we'd find on bootc after the podman PR merges
Jask Dan, he'll probably say "we're waiting on Colin" lol
Aseems bazzite-deck testing on zstd is pulling smooth af
Aimage is 6.4gb
Aand as i said that it fell a bit
KI also did a re-chunk for that last build
KSo we should have good numbers going forward
NI'm so happy about this.
DCan we get zstd on all images
KNot any time soon
KRequires a newer version of rpm-ostree that not all users may have
KMaybe with 41
DWould you mind approving the workflow again on https://github.com/ublue-os/akmods/pull/225? I'm curious if the quotes will fix the arg parsing
GitHub
This PR attempts to build both containers using the same file. Not yet tested, just submitting for some early feedback.
Ftbh, rechunk made wait for
zstd:chunked much better