| *** ykarel_ is now known as ykarel | 04:17 | |
| *** vhari_ is now known as vhari | 12:53 | |
| *** haleyb is now known as haleyb|out | 15:26 | |
| opendevreview | sean mooney proposed openstack/project-config master: add ops to cyborg and watcher channels https://review.opendev.org/c/openstack/project-config/+/977022 | 18:28 |
|---|---|---|
| sean-k-mooney | fungi: ^ finally got around to that but no rush | 18:29 |
| fungi | i had already approved it | 18:30 |
| sean-k-mooney | :) | 18:30 |
| sean-k-mooney | i know i still need to look at the acls for cybrog and gerrit at some point but i dont think that required before FF so its on my todolist for later | 18:31 |
| sean-k-mooney | im pretty sure everthing else is workign well enough for now | 18:31 |
| sean-k-mooney | do you know if the meetings ical also configred by the projec-config repo or is that seperate? | 18:32 |
| sean-k-mooney | i think gouthamr mentioned there is a irc-meetings repo but not sure which one that is | 18:34 |
| sean-k-mooney | oh its proably https://opendev.org/opendev/irc-meetings | 18:34 |
| fungi | sean-k-mooney: yes, there, you found it | 18:36 |
| fungi | i can approve changes to that as well if you have any | 18:37 |
| sean-k-mooney | fungi: on cool i just submitted https://review.opendev.org/c/opendev/irc-meetings/+/977024 | 18:38 |
| fungi | i saw, and approved | 18:39 |
| sean-k-mooney | we are going to try restarting it tomorrow. not expecting many to attend but if noting else we will start looking at open bugs | 18:39 |
| fungi | best of luck, i hope it picks up | 18:40 |
| sean-k-mooney | i was speaking to at least 1 person at ovh who was interested in better gpu managment brifly so there may be some apitate form others to add features | 18:42 |
| sean-k-mooney | and hopefully also maintain it but we will see. | 18:42 |
| fungi | yeah, the ai craze, for however long it lasts, seems to have found ways to reuse all the cryptomining gpu overspend, so i expect providers to be looking for easier ways to resell that capacity in the near term now that the bottom has fallen out of the currency markets | 18:44 |
| opendevreview | Merged openstack/project-config master: add ops to cyborg and watcher channels https://review.opendev.org/c/openstack/project-config/+/977022 | 18:47 |
| sean-k-mooney | huh really i was not expectign that but i assume you jsut mean the gpu over spend rather then teh crypto asics | 18:48 |
| sean-k-mooney | those are jsut such a waste fo resouces | 18:48 |
| fungi | yeah, mostly cloud providers stocked up on gpu hardware due to cryptocurrency proof-of-work mining demand | 18:50 |
| fungi | at least from what i've seen, granted i don't have direct customers and am mostly exposed to the user side instead | 18:51 |
| * sean-k-mooney could maybe upgrade my gaming gpu if they ever sell some of those... | 18:51 | |
| fungi | seriously | 18:51 |
| sean-k-mooney | with dram prices wehre they are right now buying any new hardware consomer or server grade is kind of insane | 18:52 |
| fungi | i'm just glad i spent my equipment refresh stipend before the hardware market went stupid | 18:54 |
| sean-k-mooney | to be fair i dotn play game that much any mroe and when i do its often factoio. but i was considerign doign some local ai stuff but im not goign to spend 4-10k just on that | 18:57 |
| sean-k-mooney | i havent made time to do this yet but im plannign to swap to GLM5 of a week or month and see how it as an openweight model comapres to antropic models. but ultimatly it would be nice to evenually be able to run some of those locally too | 18:58 |
| fungi | the vast majority of my gaming these days relies on valve's proton fork of wine, which is tuned for best performance on amd gpus, so the emerging glut of used nvidia hardware doesn't help me much | 18:59 |
| sean-k-mooney | ya i actully bought an intel gpu for my desktop arch 750 and other wise just use my steam deck | 19:00 |
| sean-k-mooney | im very tempted to reformat my windows vm and replace it with linux its the only windows anythign i have in my home anymore | 19:01 |
| fungi | right, the deck is the reason valve sank so much money into optimizing for amd hardware | 19:01 |
| fungi | granted, i don't have a deck, but steamclient on debian works great | 19:01 |
| sean-k-mooney | i was plannign to pick up one of there steam machine when it lachuch but it rumared to be alot more expensive then planned due to ram pricing so i might wait a bit | 19:01 |
| sean-k-mooney | that or wait for the next framework desktop release bacuse i really like what they are doing as a company in general | 19:02 |
| fungi | yeah, the reboot of the steam machine is definitely interesting to me, but i put together a decked-out minisforum hx99g which is probably physically smaller while being plenty powerful for that purpose (also doubles as a workstation) | 19:05 |
| fungi | for portable devices the beefiest/newest ones i've been running steam games on are gpd pocket 4 netbooks | 19:06 |
| fungi | and then i use bluetooth handheld peripherals and displayport-over-usb video glasses for relaxed gaming | 19:07 |
| sean-k-mooney | on an unerelated related tangent :) a few years ago oen fo the upstream ci providers i belive made some kind fo gpu hardware aviable at one point | 19:08 |
| sean-k-mooney | do we knwo fi that is still a thing | 19:08 |
| fungi | vexxhost, and yes afaik | 19:08 |
| sean-k-mooney | we are looking into if we can set up third party ci for cybrog testing but also wonderign what is already possible 1st party | 19:08 |
| sean-k-mooney | cool ill look into that more | 19:09 |
| sean-k-mooney | we are explorign useing sample kernl module sto fake device like nvme or mtty to fake an mdev serial port | 19:09 |
| sean-k-mooney | so im pretty sure we can do some level of first praty testign with fake devices | 19:10 |
| sean-k-mooney | but i want to see if we can do spot checsk on real hardware or perodics ectra too | 19:10 |
| fungi | mnaser can probably explain what's available to us there in greater detail, if you catch him when he has a free moment | 19:11 |
| fungi | problem is we probably want to do nested testing where we can pretend the virtual machine for devstack has dedicated gpu hardware, so that may not be easy to orchestrate | 19:12 |
| sean-k-mooney | ya so in recent nvidia driver they now actully supprot that | 19:13 |
| sean-k-mooney | but only on new cards | 19:13 |
| sean-k-mooney | and tahts $$$ | 19:13 |
| sean-k-mooney | i.e. with blackwell gpus and i think hopper you can pci passthough the gpu to a vme and then use all the virutalisation feature in that vm with nested virt for second level mvs | 19:14 |
| sean-k-mooney | i have not been following the nodepool in zuul work too closely but is https://opendev.org/opendev/zuul-providers/src/branch/master/zuul.d/providers.yaml the new place we configrue the ndoe pool lables? | 19:15 |
| fungi | do they still require a separate licensing server and proprietary client software running in the guest? | 19:15 |
| sean-k-mooney | and the mappign to the providers | 19:15 |
| sean-k-mooney | fungi: i think techniclly yes but with our a lices you get like vga level resolution | 19:15 |
| fungi | sean-k-mooney: yeah, nodepool is dead, long live zuul-providers | 19:15 |
| sean-k-mooney | i started on a kernel moduel to fake out gpus and was chattign to the kubvirt folks and they are tryign to do the same | 19:16 |
| sean-k-mooney | so im somewhat confient we cak fake enouch to have confidance the code is right then do spotchecks with real hard ware preodicly | 19:16 |
| fungi | we've been using it in opendev as an early adopter for nearly a year, but as of last week nodepool is officially deprecated: https://lists.zuul-ci.org/archives/list/zuul-announce@lists.zuul-ci.org/thread/KJWMMIZNUN3AGIY47IS25HKPYPWXMK4L/ | 19:17 |
| sean-k-mooney | https://github.com/kubevirt/kubevirt/pull/16717/changes | 19:17 |
| sean-k-mooney | ah zuul 14 was released | 19:18 |
| sean-k-mooney | im currently using 13 something but ill eventully get around to that | 19:18 |
| fungi | well, the launcher service has been implemented for many releases, but 14.0.0 is the point where we finally feel like it's stable enough to replace nodepool officially | 19:19 |
| sean-k-mooney | yep but im usign the zuul operator and im not sure if it supprot it yet | 19:20 |
| sean-k-mooney | i just went with its default config | 19:20 |
| fungi | also documented, which was the biggest missing piece until recently | 19:20 |
| fungi | i think there was some discussion in the zuul matrix room over the weekend about updating the operator | 19:20 |
| sean-k-mooney | that remind me i should proably usetream the 2 patches im carryihng at some point | 19:21 |
Generated by irclog2html.py 4.0.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!