12:01:27 #startmeeting CIP IRC weekly meeting 12:01:27 Meeting started Thu Jun 11 12:01:27 2026 UTC and is due to finish in 60 minutes. The chair is jki. Information about MeetBot at http://wiki.debian.org/MeetBot. 12:01:27 Useful Commands: #action #agreed #help #info #idea #link #topic #startvote. 12:01:27 The meeting name has been set to 'cip_irc_weekly_meeting' 12:01:34 #topic AI review 12:01:48 I had nothing noted 12:02:00 5 12:02:02 4 12:02:03 3 12:02:05 2 12:02:07 1 12:02:09 #topic Kernel maintenance updates 12:02:47 i pushed 4.4 12:02:56 I reviewed 6.12.90 12:04:10 "This week reported 63 new CVEs and 7 updated CVEs." - masami 12:04:22 anything else to add? 12:04:50 5 12:04:52 4 12:04:54 3 12:04:56 2 12:04:57 1 12:05:00 #topic Kernel release status 12:05:13 all on track, 6.1 (both) will be next 12:05:32 still no 6.1-rt-stable, I suppose... 12:06:08 anything else to add? 12:06:22 5 12:06:24 4 12:06:26 3 12:06:27 2 12:06:29 1 12:06:31 #topic Kernel testing 12:07:16 https://lava.ciplatform.org/scheduler/worker/lab-cip-siemens-muc is green again 12:07:29 Thanks 12:07:37 Not testing related, but I'm working with gitlab to get our license renewed 12:08:37 No other updates 12:08:53 if there is nothing more on testing, then move on... 12:09:01 5 12:09:03 4 12:09:04 3 12:09:06 2 12:09:07 1 12:09:09 #topic AOB 12:09:37 there were 1 or 2 repos without licenses related to the kernel... 12:10:06 cip-kernel/lts-commit-list was one 12:10:26 still unsure about just pushing GPL or first asking TSC 12:11:32 patersonc: who should delete cip-kernel/cip-kernel-tests? 12:11:54 I think the original license for this data is GPLv2. Therefore, the license for this repository will also be GPLv2. 12:12:28 if that is true, we cannot deviate anyway 12:12:50 iwamatsu: would you push a license file then so that gitlab is identifying it? 12:13:18 OK 12:13:35 cip-kernel/cip-kernel-tests should be okay now. I pushed a license to it. I'm happy to delete it though if it's not needed 12:13:37 GPLv2 or GPLv2+? 12:13:59 patersonc: it's confusing as the real repo is under cip-testing 12:14:02 same name 12:14:12 Shall I just delete cip-kernel/cip-kernel-tests? 12:14:22 I would say, yes 12:15:02 Done 12:15:08 thanks 12:15:42 iwamatsu: do you also know if we should use GPLv2+ for the other repo? 12:15:51 So just lts-commit-list remaining now. Plus a couple of non-code task-only projects that break the GitLab tooling (working with gitlab on this) 12:16:07 some scripts are v2+ IIRC 12:16:26 I don't know now. 12:17:14 I can do it if you want? 12:17:50 do what? 12:18:16 Add the license to lts-commit-list 12:18:43 Gitlab has a template for GPL v2, not v2+. But easy to copy in 2+ if that's what you want 12:19:18 the point is first of all identifying the required license 12:19:27 ok 12:19:36 although v2+ allows us to pin to v2-only as well 12:20:13 we would simply push the right text as COPYING file commit, and gitlab should recognize that 12:23:17 I will proceed with that work. 12:23:40 thanks! you were the initial committer anyway, just found out ;) 12:23:45 good 12:24:09 then maybe give pavel a change to catch up - we are already in AOB 12:24:19 I am late, sorry. 12:24:26 anything you want to add from other topics? 12:24:29 I did reviews. 6.12.92 and .93. 12:24:53 And it looks like this cycle is really a bit different, as we are pretty late 12:25:04 and there are still big series flying around. 12:25:12 Other than that, not much to report. 12:25:26 did you ping regarding 6.1-rt? 12:25:35 Not yet, sorry. 12:25:53 will be due now for us very soon 12:25:56 ok 12:26:29 then maybe switch to the AI review topic 12:26:40 uli: thanks again for the write-up 12:26:47 np 12:26:54 anyone else has comments / ideas regarding it? 12:27:47 patersonc: I suppose our AWS account would be able to cover some instances with GPUs as well, right? 12:28:34 Probably 12:28:42 If we want to pay lots :P 12:28:46 cost-wise, we would obviously be prepared to handle any overrun as CIP 12:29:19 uli: do you have an idea about what would match your needs from the AWS offerings? 12:29:35 i haven't looked into aws yet 12:29:51 we will need some rough math on costs when running this, even when just for some weeks 12:29:56 ...first 12:30:33 advantage might be that his should be able to be started and stopped as needed 12:30:36 Maybe LF has a central service of some kind 12:30:57 Maybe we need to get Google or someone similar to join the project and get free tokens ;) 12:31:01 regular LF IT will likely be too costly 12:31:25 getting tokens means also buying into specific commercial models 12:32:04 we could try that as well, later on, and specifically if there something comming via Glasswing 12:32:25 but the first idea would be replicating uli's setup for shared use 12:32:46 what would be an acceptable budget anyway? like per month? 12:33:32 I would have not problem to ask for, say, 10K first, and we see how far we would get with it 12:34:05 days, weeks, or some month - and then we can check again the value and make a real plan 12:34:23 the only option i have looked at so far (except for pay-as-you-go stuff like vast.ai) is a server from hetzner for 800 something per month (96 GB VRAM) 12:34:23 likely days would be too short, though ;) 12:34:46 +2000 or 3000 (don't remember) setup or therabouts 12:35:16 additional supplier may complicate a quick setup, I guess 12:35:27 yeah, i expected that... 12:35:29 that's why I thought of AWS - we have our credit card registered 12:36:51 further thoughts? 12:37:12 Thank you for the good analysis uli 12:38:10 happy to help 12:38:16 https://aws.amazon.com/ec2/instance-types/g5/ - this sounds... expensive :) 12:40:23 i have difficulties parsing that web site 12:40:53 You can click and talk to their "certainly not halucinating" AI assistent :) 12:42:21 there are also special inference instances, but probably with special accelerators 12:42:36 known to work is something with NVIDIA or AMD, I suppose 12:42:57 sooo may different instance types, easily losing overview 12:43:45 you want nvidia. everything else theoretically works but in practice is a pain 12:43:52 especially intel 12:44:08 https://aws.amazon.com/ec2/instance-types/g7e/ ? 12:44:37 "NVIDIA RTX PRO 6000 Blackwell Server Edition" 12:44:40 https://instances.vantage.sh/aws/ec2/g5.12xlarge?currency=USD&duration=annually 12:45:42 5k/h - hmm, that would be short fun with my 10K 12:47:05 jki: That's likely $5/h, right? 12:47:20 jki: It is US, so dot is decimal dot. 12:47:32 yeah, dot-confusions 12:47:36 sound much better 12:47:40 https://www.hetzner.com/dedicated-rootserver/gex131/ 1k/month incl vat 12:48:06 uli: that can still be an option once we know we want to run that 24/7 12:48:16 ...or have to 12:48:35 I would go for some on-demand first, if possible, also to play with instance sizes 12:48:41 in that case vast.ai might be an option. it's not super-reliable, but then i always choose the cheapest, most unreliable option there... :) 12:48:43 or with new models 12:50:01 looks a bit cheaper, if I chose the right comparison, but not by factors, right? 12:50:35 it's actually even a bit more expensive than the hetzner option, but then you can get everything by the hour there 12:51:16 and no setup fees 12:51:49 my point is what would be easiest for us to start playing, also from account handling perspective 12:52:10 I don't want anyone to swipe their personal credit cards here 12:53:56 what speaks against trying out an EC2 instance first on our existing account? later, we could check what is needed to use other hosters as well 12:55:12 for me that's ok 12:57:09 then I would still ask for a 10K reserve from TSC->Board so that we have some buffer in case we overshoot the AWS grants 12:57:51 anything else for today? 12:58:22 5 12:58:23 4 12:58:25 3 12:58:27 2 12:58:29 1 12:58:30 #endmeeting