[00:14] what's needed on Ubuntu 22.04 to get an Intel ARC GPU working? [00:14] I'm on 6.8 (-hwe) kernel already [00:14] intel_gpu_top seems to report there's a GPU, but it doesn't show the name, it only shows the PCI ID [00:26] all I did to make it work [00:26] but to make it *stable*, that has been impossible [00:30] I feel you left out a key point of what you did! [00:30] hmm [00:30] nope [00:30] I installed the intel gpu stuff [00:31] then when 6.5? or was it 6.8, I uninstalled all that to use the kernel built in driver [00:32] hmm, I still have the intel libdrm-intel1 and xserver-xorg-video-intel stuff install though [00:32] but that is for desktop, not server [01:51] patdk-lap, sorry, was away -- which "intel gpu stuff"? :) [01:53] hmm? [01:53] I listed them above [01:53] directly from intels gpu webpage [01:53] https://www.intel.com/content/www/us/en/download/747008/intel-arc-graphics-driver-ubuntu.html [02:09] That just tells you to go to https://dgpu-docs.intel.com/driver/client/overview.html [02:09] yes [02:10] not sure what you want [02:10] if your going install hardware, you should consult the manual for that hardware [02:13] tl;dr: stuff seems to work, but `intel_gpu_top` doesn't even report the GPU name [02:13] all I'm getting is [02:13] intel-gpu-top: 8086:56a5 @ /dev/dri/card1 - 0/ 0 MHz; 100% RC6; 0 irqs/s [02:14] I'm assuming there should be some newer tools/packages/driver/something that can actually show me the correct GPU name, amongst others [02:15] hmm, the name it will only report is the pci id [02:15] maybe ask intel to add a name mapping table to intel-gpu-top? [02:16] well, if you read the page I posted it tell you how [02:16] clinfo | grep "Device Name" [02:16] Device Name Intel(R) Arc(TM) A580 Graphics [02:16] I know the clinfo tool reports it [02:17] well, intel-gpu-top doesn't support that [02:17] it *did* on a 12th Gen iGPU when I last used it [02:18] for example: intel-gpu-top: Intel Alderlake_s (Gen12) @ /dev/dri/card0 - 0/ 0 MHz; 100% RC6; 0.00/20.69 W; 0 irqs/s [02:18] that is really old [02:19] I'm kinda bummed out that neither nvtop and neither intel_gpu_top report the actual vram used by my ffmpeg processes [02:22] no, you have to use the nv smi tool to see memory usage [02:22] always been a limit for me, constantly running out of gpu memory [02:26] xpu-smi stats works [02:33] weird, my system froze while transcoding [02:33] no kernel dump, no nothing [02:33] ya, that kept happening to me, after 2 transcodes [02:33] so I stopped using that [02:33] as I said, intel arc wasn't very stable for me at all [02:33] I've seen that *before* on a different system (the one with 12th Gen), and I blamed it on the crappy iGPU [02:34] never used the built in cpu gpu, only the arc cards [02:34] didn't think ARC would suffer from the same issue [02:34] built in gpu with 4 4k monitors is never very happy [02:35] thank f* I have Intel AMT on that device, and I was able to hit a reset, because my BliKVM is not connected to the ATX pins [02:35] this workstation has ipmi :) [02:35] this one doesn't :( [02:35] it's an old ThinkStation 510 (P510?) [02:36] yeah, P510 [02:37] glad to know I'm not the only one seeing random freezes [02:37] lol, crashed again [02:38] ya, using ffmpeg transcodes, it always happened [02:38] without that, just normal desktop usage, screensaver would cause it to happen [02:38] I would get artifacts on screen, things would act odd, after a day or two [02:38] then a few more days later, the gpu would lockup [02:39] can still ssh in and do stuff, but video was no good [02:39] less screensaver activations, the longer it would last, it seems [02:41] well, this is wild [02:41] I was hoping to replace the small 1050 Ti I have in there [02:41] with the ARC, which has more VRAM and a more modern encoder, I'd assume [02:42] vram isn't really an issue for ffmpeg [02:42] it will never exceed 2gigs [02:42] I do [02:43] odd, a310 works well in windows using ffmpeg [02:43] I run about ~15 concurrent transcodes, so it eats it up [02:43] been really happy with the cards on windows machines [02:44] hmm, the arc only supports 2 streams [02:44] uhm, why would it support only 2 streams? as far as I know it shouldn't be limited to anything...? [02:45] there are only two video encode/decode chips [02:45] same deal on nvidia cards [02:46] that's not the issue, I'm not running the encoder/decoder at full speed, you can initiate much more many sessions [02:46] https://123.456.ro/share/2024/11/kitty-0.76.1.13_Hj5wr3dCTh.png [02:46] Tesla P4 [02:48] GTX 1070: https://123.456.ro/share/2024/11/kitty-0.76.1.13_Whd4E8XrXD.png [02:48] etc. [02:49] dunno, I am just never liked using gpu for video, transcodes sure [02:49] but I rarely transcode [02:51] it's pretty the #1 reason I use GPUs under linux servers :D [02:52] ya, I only use it for my desktop, and not liking it, but nvidia has become such a pain [02:52] do use nvidia in servers, cause the ai software requires it [02:54] I don't really use linux much on desktops, but at least on server-side, I had the _least_ troubles with nvidia [02:54] intel always gives me headaches, amd too [02:55] amd drivers are kindof hard, but that got fixed lately when they went into the kernel [02:55] but I opted to try intel instead of a more powerful amd card this last time [02:55] I dont need a powerful gpu, I just need the vram [02:56] it's mostly why I also wanted to give ARC a try [02:57] everything after the GTX 10 series is pretty damn expensive to use just for video transcoding [02:57] yep [02:57] the gtx 1080 is also kinda weird [02:57] the 1050/1070 have a single nvenc unit [02:57] the 1080 has 2 [02:57] I could use a tesla series, but that wouldnt last much for future use for me :( [02:58] pascal is too old to even use :( [02:58] same ffmpeg command would use about ~20% more VRAM on the 1080 due to the nvenc/nvdec count [02:58] which is weird as hell [03:00] I would have loved AV1 encode support that the ARC supports/provides :( [03:00] We use those Tesla P4's, but they're a bit tricky [03:00] Due to lower clocks, the encoder gets 100% busy with much less streams than on a 1070 [03:01] helps a bit to raise the frequency with nvidia-smi (not like yuo can much, it just has 2 settings lol) [03:01] my stuff doesn't matter how long it takes, but it has 24gigs of ai data to load :( [03:01] and to start doing two ai jobs on the same file now, so that is going be 50gigs of ram needed :( [03:01] or have to deal with using two cards [03:02] now, WTF do I do about this ARC... [03:03] my daughter has loved it in windows :) [03:03] I was like 100% sure it would work just "fine" [03:03] the amd I had in her system, failed, and it's worked well [03:03] ...I could dist-upgrade to ubuntu 24.04, maybe... [03:03] it might, but ya, not currently [03:03] wonder when intel will release updates for linux [03:04] maybe, I havent wanted to deal with that on my desktop yet [03:04] so I cannot speak to what it would do with 24.04 [03:05] have upgraded many servers, but, well, desktop upgrades are so much more finicky [03:05] #yolo [03:05] I have BliKVM, I have Intel AMD, can't break it! [03:05] ya, breaking isn't my problem [03:06] how long till it works enough nicely, so I can do work, is :) [03:06] probably clone my laptop, upgrade it, and see how it goes, then try this system [03:48] oh, the XPU-stuff is for data-center cards only [03:48] hmm? works on my card just fine [03:49] xpu-smi stats -d 0 [03:49] intel-gpu-top: Intel Dg2 (Gen12) @ /dev/dri/card1 - 1613/2455 MHz; 0% RC6; 3945 irqs/s [03:49] maybe only works on pcie based cards (arc) [03:49] works on my a580 and a310 [03:50] neither are datacenter, dont have an a40 yet :( [03:52] this is from xpumanager, correct? [03:52] yes [03:53] atleast from the git project [03:53] https://dgpu-docs.intel.com/installation-guides/index.html [03:53] >_< [03:53] I just did apt-get install xpu-smi [03:53] let me see, the 24.04 upgrade is done [03:53] maybe it's in their ppa again [03:54] nope, it's not in "https://repositories.intel.com/gpu/ubuntu noble client" [03:55] Are you on 24.10? [03:55] and using kobuk-team/intel-graphics-testing ? [03:55] 22.04 [03:55] no [03:56] https://repositories.intel.com/gpu/ubuntu jammy client [03:56] you sure? what does apt show xpu-smi show ? [03:58] hmm, that command doesn't say [03:58] should say repo? [03:58] nope [03:58] ie: APT-Sources [03:59] ...how [03:59] APT-Sources: https://repositories.intel.com/gpu/ubuntu jammy/client amd64 Packages [03:59] like I said [03:59] wtf [03:59] ...shit crashed again [04:00] worked for about ~15 minutes with 14 transcodes [05:00] and crashed again https://123.456.ro/share/2024/11/kitty-0.76.1.13_TN72mR22Wx.png === ubuntu_4321 is now known as ubuntu4321