Meeting started at 2026-05-26 16:00:08 UTC

The Meeting name is 'riscv-sig'

Hey folks, anyone around for a riscv meeting today?

!hi

davidlt: David Abdurachmanov (davidlt)

!hi

abologna: None (abologna)

Cool. My laptop just rebooted. 🙃

definitely monday.

Anyhow, how about a quick fedora 44 update, then hardware, then open floor?

Sure

!hi

Conan Kudo 😌: Neal Gompa (ngompa) - he / him / his

Not a lot of building. I started landed a huge 500+ qt 6.11 (and related things) update, which got blocked on qt webengine. Marcin yesterday finished rebasing relevant patches for us.

So I should be able to submit a new qt webegine and continue, but haven't done that yet.

Otherwise we also gained COMPS and initial f44 repo (but not f44-staging and omini, or similar).

at what point should we look at making some images?

that probibly needs to wait until after that 6.11 batch is done?

I believe, abologna has been looking at kiwi configuration and might have built cloud image already.

Some things are broken now for sure, but could, and generic server images probably will build. Desktop images definititely not until that large update fully goes in.

!hi

kashyapc: Kashyap Chamarthy (kashyapc)

anything else on f44? pain points or ?

I think, there is again interest to move to a new forge.

We still don't have change on Koji hub to allow usage of it.

that should be easy enough... just a line or two in policy...

We also added one more P550 board to the build farm.

Yes, one line if someone finds that magic 30 seconds to get it done 😄

I'd be happy to do it if someone gives me context/urls/etc.

A question - was this qt 6.11 update "mandatory" to land, before we could finish F44 building? Asking out of ignorance :)

nirik: it's allowing repos under https://forge.fedoraproject.org/riscv in SCM list.

Ah, the Kiwi config, I guess.

I think, some stg stuff already might have it, so copy & paste and add "/riscv" probably.

ah, ok.

so for packages patches/overlay? or ?

If you want containers, and server it should be ok. I will block desktop stuff (and maybe some CLI that attempt to pull in qt, etc.)

Yes, it's for moving from fedora.riscv.rocks:3000 where we keep other dist-git "overlay".

Ah, right. Maybe for F45, we could push out the minimal images (server and container) and then get to the rest?

right, ok. Is that going to be one overlay repo with all the packages under it? or multiple repos/one per package?

multiple repos

alright.

Again, any repos under https://forge.fedoraproject.org/riscv

https://paste.centos.org/view/c0cdd6a5 should perhaps do it. (well, might need firewall tweaking too)

Ah, okay, thanks!

wait, that's going to be allowed?

that seems kind of crazy since there's no enforcement of dist-git rules on forge.fpo

it's just moving the existing already used one to forge... this is for the secondary efforts only. It will not exist when things move to main koji.

... okay I suppose

do we intend to rebootstrap the architecture when we move it to main koji?

it sounds like we should

possibly. I don't think thats something we need to decide right now... given that I don't think it's particularly soon. ;)

Depends what you mean by rebootstrap :smi

Depends what you mean by rebootstrap :smil

a complete build of the whole tree from scratch using the original secondary koji repo as a seed environment

Yeah, it should get a full rebuild on import.

(or during mass rebuild, or both)

well I would probably argue it should be before that, since it "de-taints" the buildroot

I wouldn't call it rebootstrap (as bootstrap is more painful process, I would say)

in this sense, it's a bootstrap in main koji

rather than a plain import

It works for me, once we are ready, which isn't today. The goal is to have no changes (well, no dist-git overlay, etc.) before we ready to go "in".

So we trim thing, and reduce diff (Marcin did some epic job here), but we are not at 0 diff wise.

There is also no rush as there are not hardware that would be suitable for main Koji.

right

So really a future discussion.

But yeah, it will be rebuilt 1-2 (or more) before it gets shipped as anything official.

How or how many times it's a future topic.

Should we move on to any hardware news?

does anyone have any? ;)

There is a conversation going on with Brian Redbeard about a shortcut to getting things in place faster, but I have no idea what's going on internally at Red Hat about it.

K3 started shipping, some folks in #riscv:fedoraproject.org already received theirs and started working on hacking a kernel.

I know kashyapc was doing some benchmarking.

Yes!

Conan Kudo 😌: these talks are unknown to me, but I don't think there is anything public today suitable for the main Koji instance.

Not yet, but I spoke with him at Red Hat Summit about it, and there's something expected to come online next month and there's a commercial discussion going on with Red Hat about it.

I have a remote access that'll end in a few days. But I could re-provision for another 20 days or so, I was told.

Anyways, if something happens next month or even 2-3 months from now we wouldn't be ready.

Yeah.

True, but we would be able to be ready _sooner_.

FWIW, I'm also working with relevant management at Red Hat to secure funding for hardware for Fedora data-centers. It's an evolving situation, as there's the "Waiting for hardware vendor to do their thing - ship HW" variable

I've learned to not be in a hurry on it... it will be ready when it is

Absolutely. I agree with it.

Also conserves energy.

Once someone tells me there is a hardware plan we will get things ready + plan how to get it happen.

I tried to keep things with minimal diff before, but it costs a lot of energy. Hard to keep it up without a plan.

davidlt: I'm following it up closely. I don't know if K3 suits the needs of Fedora DC.

yeah, upstreaming things and fixing up packages only helps when the day comes.

K3 is too slow, I think. Needs another 2x perf push at least.

Also too little memory I would say too.

X200 (K5), maybe Atlantis could do it.

Hmm.

Anything else might exist, but it's not a public knowledge.

I'm just wondering if K3 can be a temporary boost in Fedora DC until Atlantis starts shipping for real.

yep. and need to exist in the sense that people can actually buy them and have them delivered. ;)

Spacemit should also do a server one at some point, but I doubt it will be better than K3 ST perf.

X200 (IPC + Fmax) changes makes it 100% or more bump. Again, they don't have SoC with it yet, which means it could be anything in real world.

Also X200 is based on a new core. X60 and X100 was based on OpenC910 IIRC.

nirik: Later, maybe you can look at K3 specs and see if it's worth the hassle to fit them in Fedora DC? I ask because, to not fall into the slippery slope of "if only we wait a bit longer, even better HW will come around" (which is kinda true, but it risks a "perpetual wait" situation)

X200 is based on open source XiangShan.

!hi

Jason Montleon: Jason Montleon (jmontleon) - he / him / his

As of now K3 is already bringing noticeable improvements: 2-ish days for GCC instead of 5-ish days, and so on.

Putting something to DC was never a problem. Anything hacky is a problem, and performance for Fedora community is the biggest problem.

Sure... faster is always better, but the real trick is fast enough to not cause a bunch of waiting/pain vs other arches...

Right. What's the benchmark for it, then? Close enough in build time to Fedora Arm builds?

I was told not the best stories about aarch64 and it's slow performance and impact to the community. I was suggested not to do that with riscv64.

I don't know how K3 compares to some modern Ampere (or other HW that won't go defunct soon)

well, it was armv7 that was initially really slow...

I don't know that arm is the slowest anymore...

davidlt: Here's the history from 2012 :-) https://lwn.net/Articles/487622/ -- Fedora mulls ARM as a primary architecture

The new aarch64 machines in Koji are quite fast. I think, even faster than x86_64 the last time I checked.

I think nowadays that honor belongs to either of the two IBM architectures

davidlt: Here's the history from 2012 :-) https://lwn.net/Articles/487622/ -- "Fedora mulls ARM as a primary architecture"

I suspect it's probibly s390x... but not in all cases...

Yeah, I just checked; GCC build completes on s390x in 17 hours. vs on some fast-enough x86 hardware in 27 hours.

Fedora failed to be relevant on ARM, I would hope we don't wind up in the same state with RISC-V

right, but kernel aarch64 is slowest.

Fedora failed to be relevant on ARM for the longest time, I would hope we don't wind up in the same state with RISC-V

Oh yeah, ppc64le was 3-4 times slower compared to aarch64 and x86_64 to compile GCC.

so, it's nuanced. not one single simple hot take. ;)

Conan Kudo 😌: Please say something cheerful for me. My day is deadly hot here already :D

Conan Kudo 😌: Please say something cheerful for us. My day is deadly hot here already :D

We're probably going to have a RISC-V based cloud this summer in north america?

is that cheerful for you?

Heh, yes! If that is really true.

it's part of what I discussed at Summit

I still hope that QC might announce something interesting in a month.

lol no

What is QC, again? My brain is melting here, and is oozing out of my ears

and with Tenstorrent being potentially acquired in the next few months, I've counted them out too

Qualcomm.

Who is acquiring Tenstorrent?

QC == QCOM == Qualcomm

either QCOM or Intel

both are rumored to actively be courting them for a buyout

QC already got Ventana

yeah and now Ventana is a nonentity

Folks, we're going in free-form discussion here :-) Which is fine for me, unless we have more pressing stuff :)

It looks like all companies working on riscv64 + high performance + AI are being acquired.

yes, and I expect that all of them will stop working on riscv64 if their acquirers are into competing arches

It kinda always has been for me. They were announce v3, v4, while no v1, v2 in sight. Last year in Paris (at the EU RISC-V Summit) they had a giant empty booth with nothing to show for. Just a suit-person was sitting in the booth 😆

It kinda always has been for me. They were announcing v3, v4, while no v1, v2 hardware in sight. Last year in Paris (at the EU RISC-V Summit) they had a giant empty booth with nothing to show for. Just a suit-person was sitting in the booth 😆

well, they took the Red Hat toolchain people working on it, so...

we have about 10min left... open floor?

Well, it depends on when QC started acquiring Ventana. Empty booth might have been part of that process.

Yes, we get back to topic here.

Open floor is the topic 😄

I think, heat is getting me today.

We have a Flock talk upcoming. I still need to update some slides, been saddled with life-stuff

I almost fell off the bicycle a bit ago due to dizziness from the heat, after a short ride to do an errand

I am hiding in a corner where ☀️ cannot reach me.

good for solar production tho! </brightside>

Same here 😆 Looking at the sun from a safe distance in my place. The sun is baking the couch through the curtains

anyhow, thanks for coming everyone. Lets go back at it in #riscv:fedoraproject.org

Yes