03:06 peterwergaahen: obviously it does not work so that in a war Russia has military equipment and west just takes their powering electronics over, I.e responds with only hybrid war. War is not war cowards who butcher conspired ways as cowards their own people and say oops we did it again and that all after their implants, you go neck to neck or get killed, since the defense for electronics and powerplants
03:06 peterwergaahen: is very simple update. you get bigger problem in front of you than you ever imagined with your ill thinking Estonian and Finnish scammers, you are meant to die scum and you will. none gives mercy to those cheater tyran gangteams.
12:08 phomes_[d]: I finally figured out the issue with mangohud when we do multiple queues
12:10 phomes_[d]: debugging these things seems to always be 95% time spent looking in the wrong place 🙂
12:59 karolherbst[d]: yep
15:34 x512[m]: Any ideas why display start working only after connecting it to different port on RISC-V machine with Nvidia GPU? Observed on both Linux Nouveau and Haiku OpenRM.
17:28 _lyude[d]: If you could get a log with `drm.debug=0x16` in the kernel options, push buffer debugging enabled in the kernel config, along with `nouveau.debug=disp=trace` that would be appreciated
17:28 _lyude[d]: what kind of machine do you have that you're able to run a nvidia GPU on that's also riscv btw?
17:30 _lyude[d]: btw airlied[d] - gonna tidy up my office in a bit to make some space for another machine so I can take a look at the issues you were talking about
17:31 x512[m]: It is HiFive Unmatched RISC-V board.
17:36 _lyude[d]: sounds unmatched
17:36 x512[m]: https://www.sifive.com/boards/hifive-unmatched
17:36 _lyude[d]: oh dear it is discontinued
17:36 _lyude[d]: it truly is unmatched
17:37 x512[m]: Milk-V Titan should start shipping soon.
17:37 _lyude[d]: i'll have to see if I can convince work to expense one perhaps
18:27 tdaven[d]: FYI, the rev b version of the Unmatched is available. Only the older rev is discontinued.
21:21 _lyude[d]: Huh. airlied[d] jfyi, it seems like ada actually works fine with eDP. so I guess it must just be ampere
21:23 airlied[d]: blackwell fails for me as well, so it might be laptop dependant, but also make sure you are seeing the eDP panel two a disable/enable operation in the first atomc modeset
21:27 _lyude[d]: yeah imma search through my other laptops, I k now I had at least one with this issue
21:29 airlied[d]: seeing it on a HP at the moment, but I think I have a lenovo with it, make sure to find the discrete graphics toggle 🙂
21:40 _lyude[d]: Ok - I found a laptop that takes like 5 minutes to turn on, and I'm now remembering i'm pretty sure this was the machine that did the thing.
22:11 _lyude[d]: airlied[d]: got it. Though, the behavior on mine is turning the display on with a bunch of garbage!
22:11 _lyude[d]: but it's probably the same issue
22:11 _lyude[d]: will confirm by removing the inherit once I get a custom kernel built
22:14 _lyude[d]: oh no - my screen doesn't seem to come back after the initial modeset though, it just comes up in the same state as before
22:26 Blaine: hello anyone here?
22:27 assi-dude8787[m]: Yes
22:27 Blaine: hey where is the developer documentation>
22:27 Blaine: *?
22:28 Blaine: like nouveau.freedesktop.org is hella old and I have no idea where to even start
22:30 mhenning[d]: There's some documentation on the mesa website https://docs.mesa3d.org/drivers/nvk.html
22:31 Blaine: will check it out
22:32 Blaine: I am trying to trying to add my nvidia tesla k80 to the driver
22:33 Blaine: I picked it up for $20 a few months ago
22:35 mhenning[d]: Hmm. I'd expect a k80 to work already. If it's not working then that's worth a look
22:35 karolherbst: the datacenter GPUs are sometimes very funky in very unexpected ways
22:35 karolherbst: ohh wait
22:36 karolherbst: it's a different k80
22:36 karolherbst: oh no, it's a DC k80
22:36 karolherbst: is it... gk210?
22:36 Blaine: sorry just looked back
22:36 Blaine: yeah gk210
22:36 karolherbst[d]: indeed.. gk210
22:36 Blaine: one sec
22:36 karolherbst[d]: I don't think nouveau loads on that one yet 🙃
22:37 Blaine: yeah ik
22:37 Blaine: any how https://www.techpowerup.com/gpu-specs/nvidia-gk210.g773
22:37 karolherbst[d]: yeah....
22:37 Blaine: similar to the gk110 https://www.techpowerup.com/gpu-specs/nvidia-gk110b.g760
22:37 Blaine: someone did get it working
22:37 karolherbst[d]: the fun part is going to figure out how much of maxwell and how much of gk110 and how much of gk208 that one is
22:38 karolherbst[d]: nah the gk110 is a gk110, the gk210 is different
22:38 Blaine: here https://bugzilla.kernel.org/show_bug.cgi?id=209345
22:38 karolherbst[d]: might even have subtle ISA changes
22:38 Blaine: yeah I was seeing that but from my knowlege it is just a minor update to gk110b
22:39 karolherbst[d]: well... if it would, it wouldn't crash
22:39 karolherbst[d]: *Would be
22:40 Blaine: yeah I was going to see but I would like to make a "better" fix b.c. it does have a lower base clock
22:40 karolherbst[d]: mhh gk210 has twice as many registers.. fun
22:41 Blaine: where are you seeing that?
22:41 karolherbst[d]: there is a whitepaper
22:41 Blaine: ah
22:41 karolherbst[d]: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/tesla-product-literature/NVIDIA-Kepler-GK110-GK210-Architecture-Whitepaper.pdf
22:41 Blaine: I was looking at that
22:41 Blaine: let me look again
22:41 karolherbst[d]: now I'm curious on why it's crashing...
22:42 karolherbst[d]: I _Think_ copying the gk110 vtable and going from that is probably the way to go, but it might also run into unexpected issues apparently
22:42 karolherbst[d]: ohh mhh
22:42 karolherbst[d]: the k80 comes without cooling, right?
22:43 Blaine: yeah doesn't stop me tho cobbled by own cooling
22:43 karolherbst[d]: well as long as hte GPU doesn't reach ~110ºC you are good
22:44 karolherbst[d]: probably
22:44 karolherbst[d]: the GPU might shut down a bit earlier
22:44 Blaine: lol I would also like to save my power bill will need to fix the clock too
22:45 karolherbst[d]: if you select the lowest pstate after boot it should put the GPU into a bit of a lower power state
22:45 mhenning[d]: karolherbst[d]: probably starting with the gk208 vtables would be closer?
22:45 karolherbst[d]: there is also clock gating that ain't enabled by default that should reduce power draw at idle
22:45 karolherbst[d]: mhenning[d]: mhhh it seems like the gk210 was released earlier
22:46 karolherbst[d]: or.. uhm...
22:46 karolherbst[d]: later?
22:46 mhenning[d]: I'd still guess another kepler2 part would be closer than kepler1
22:46 karolherbst[d]: gk110 is kepler2
22:47 karolherbst[d]: like it comes with the new ISA and stuff
22:47 Blaine: one sec so 2x 32-bit Registers / Multiprocessor , larger Max Shared Memory / Thread Block 112K vs 48k, and lower base clock gk110b vs gk210
22:47 karolherbst[d]: clocks don't really matter
22:48 karolherbst[d]: and the other stuff is only relevant for userspace
22:48 karolherbst[d]: mhh but it does look like the gk210 was release late...
22:48 karolherbst[d]: so maybe it is closer to the gk208 indeed
22:48 Blaine: how so I saw you guys use some sort of prec to pull it from firmwarebins?
22:50 mhenning[d]: karolherbst[d]: ugh, right the naming is weird
22:51 Blaine: is there any other documentation other that mesa3d I can try to work back from the nouveau/nvkm/engine/device/base.c but I feel like I don't quite grasp how these gpu profiles are made
22:52 karolherbst[d]: now that I've landed all the shared memory occupancy stuff we can even make better use of those DC card for compute workloads 🙃
22:52 karolherbst[d]: Blaine: just copy the table from "gk208", and change the name to "gk210" and see if that works better than the patch in the bz
22:53 karolherbst[d]: `nv108_chipset` or `nv106_chipset`
22:53 karolherbst[d]: but I think they are identical
22:53 Blaine: ok let me try I am also new at building the linux kernel so I might put my server as unrecoverable and I will need to go home
22:54 karolherbst[d]: huh.. but hte chipset is `0xf2` _funky_
22:54 Blaine: 208 or 208b
22:54 karolherbst[d]: shouldn't matter
22:54 karolherbst[d]: I think the patch on the bz copied from `nvf1_chipset`
22:55 karolherbst[d]: but that apparently didn't work out to well
22:55 karolherbst[d]: but without a serial console attached it's gonna be pain to debug this
22:56 Blaine: lol I have a second k80 card if you guys want I can just ship it to you it just has 1 of the 2 gpu's dead
22:57 Blaine: but this card is really nice for cupy (numpy cuda python) when doing science stuffs
23:00 Blaine: was wanting to try the galium cuda and see if I can run my card on some somewhat new drivers
23:04 mhenning[d]: there's no open source cuda implementation, if that's what you're implying
23:11 Blaine: ah this is what I saw https://docs.mesa3d.org/teflon.html
23:13 karolherbst[d]: well the k80 not gonna help with that at all
23:14 Blaine: yeah but I got it confused with that (k80 does not have fp16 no tensorflow) cluda and found cluda when loooking up teflon https://www.phoronix.com/news/Mesa-CLUDA-MR-CUDA-Gallium
23:15 mhenning[d]: yeah cluda is the reverse direction (gallium on cuda, not cuda on gallium)
23:15 Blaine: which is indeed not open source but I was thinking I could use the opensource galium driver point it to lubcuda.so and have something work
23:16 karolherbst[d]: nope, that's not how it works
23:17 Blaine: darn guess I can settle for binding one gpu to nvidia close source one nouveau
23:44 Blaine: darn server not responding got to go home