01:06 _lyude[d]: will probably have something else for you to try later as well as I clean this up a bit and make sure this is actually reliable
03:19 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1501785626398429264/20260506_221423.jpg?ex=69fd565b&is=69fc04db&hm=0a39658d428c2b8e31b934838f33edca0cce6427cfd54d4c6cc03bd45f7bc7db&
03:19 mangodev[d]: i think i hit the turing bug
03:20 mhenning[d]: do you have anything in dmesg? what kernel?
03:22 mangodev[d]: 7.0, checking dmesg and journalclt
03:22 mangodev[d]: hit it in jetbrains
03:23 mangodev[d]: i presume it's the buffer view bug
03:23 mangodev[d]: since i didn't have the patch to disable compression
03:24 mangodev[d]: yep, it's the buffer view bug :/
03:24 mangodev[d]: context lost
03:25 mangodev[d]: https://cdn.discordapp.com/attachments/1034184951790305330/1501787018781397104/rn_image_picker_lib_temp_f0877a4f-e55e-4b93-b039-41fbc5f4b088.jpg?ex=69fd57a7&is=69fc0627&hm=a2f799ff8376be9e0f5a1649b3c50bf00015d8df768a3c8987ff139e29e50499&
03:27 mangodev[d]: testing rustrover again for the fault
03:28 mangodev[d]: fixed
03:28 mangodev[d]: mhenning[d]: ping
03:29 mhenning[d]: yeah, could be the buffer view thing
03:29 mangodev[d]: i do hope we're able to find how to disable buffer view compression
03:29 mangodev[d]: prop seems to be able to, they most definitely have color compression on turing and prior
03:30 mangodev[d]: this'll probably need a nouveau patch though
03:30 mangodev[d]: so who knows how long it'll take to get upstreamed :/
03:31 mhenning[d]: yeah, it's probably fixable, just someone needs to get around to it
03:31 mhenning[d]: not sure it'll need a kernel patch. might be a userspace bug
03:32 mangodev[d]: that'd be the nicer possibility of the two
03:33 mangodev[d]: i mean
03:33 mangodev[d]: better news
03:33 mangodev[d]: it seems AMS works on my setup now? unsure how to check for sure though
03:33 mangodev[d]: before kernel 7, AMS would fail to set up on my system and fall back to legacy modeset
03:33 mangodev[d]: now it doesn't do the black flash from failing AMS
03:36 mhenning[d]: that's good. Lyude has been working on stuff related to atomic modeset so maybe one of those patches helped
13:43 v1withstyle: hello
13:44 v1withstyle: dont aswer if its a stupid question, but why power management(clock management specifically), is hard?
14:01 phomes_[d]: fyi some numbers in the perf sheet are getting an update. I added warhammer 3 benchmark. I noticed that the first run was a bit unstable, so now I do 4 runs and use the avg of the last 3
14:02 phomes_[d]: I now also use that method for the benchmarks of dirt rally and tomb raider. I am rerunning it for baseline and the tracked MRs so those will change during the day
14:03 v1withstyle: how can you play warhammer on boot clocks of cards? arent they low?
14:05 Mary: v1withstyle: Are you asking for Maxwell/Pascal GPUs?
14:06 v1withstyle: i dont know you tell me
14:07 Mary:shrugs
14:08 v1withstyle: i should ve picked amd
14:09 Mary: v1withstyle: For Maxwell Gen 2 (GTX 9XX) and Pascal (GTX 10XX), the microcontroller on the GPU that handle power management requires signed firmware to work and we don't have those
14:09 v1withstyle: yeah i have the worst pascal to ever exist gt 1030 ddr4
14:09 Mary: for Turing and later so (RTX or GTX 16XX) we have firmware that handle power management for us provided by NVIDIA
14:10 v1withstyle: is there no way to revers engineer that firm?
14:11 Mary: v1withstyle: It's not really only a matter or reversing those firmware, you cannot run anything that isn't signed by NVIDIA on those microcontrollers
14:11 Mary: the firmware are also encrypted so that's also another problem sadly
14:11 v1withstyle: have you even tried?
14:12 Mary: yes I did.
14:13 v1withstyle: well i dont know nothing about videocarads reverse engineering but maybe just sign by nvidia key and ecnrypt?
14:13 Mary: NVIDIA keys are not accessible or known, it's fused in hardware
14:14 v1withstyle: its bios blocked from executing non signed firmware?
14:15 Mary: you can execute unsigned code but not touch any registers related to power management, you need to upload another firmware that is signed and validated by the hardware when you start executing it
14:15 Mary: it's a whole thing
14:16 v1withstyle: programmator?
14:18 v1withstyle: i also got fermi gtx 460 with one disable vram chip
14:19 v1withstyle: atleast thats doesnt have the signed firmware problem, right?
14:20 Mary: I am not too knowledgeable about Fermi situation but yes that surely does not have signed firmware
14:22 v1withstyle: and i see that most of the cards dont have auto reclocking, is theres something to do about it
14:47 karolherbst: implement reclocking correctly
15:25 v1withstyle: good luck making that
15:53 HdkR: lol, I love the statement, "Have you even tried?"
15:53 HdkR: The world sucks, trying harder won't change the problem :P
17:10 Jasper[m]: I assume this person does not speak English very well
19:10 trevorstanford_: Such devils should never be dealt with really, however i make your life short and living hell instead, I made the needed work on paper, as i am preparing to get a new computer i will attack all your services during this year with super backends until you kill yourself. I made a new song: every stalk you make, every step you take, every smile you fake, I’ll be killing you. must be idiots
19:10 trevorstanford_: dudes have stalked me for 20 years now, 7 years frozen my sockets, 5 years hammered my equipment with bootrom viruses, with severe assaults and kill off attempts starting from 2011 continuing to 2026 actually all because of AIDS medication fraud, framing me mentally ill due to inevitable software that I paradoxically enough already wrote. as I said I make all your crankgangsters lives a
19:10 trevorstanford_: living hell. and you will kill yourseselves in masses. fucking idiots.
19:14 trevorstanford_: if that software does not get awarded to my name, it’s as simple as that that someone other writes it anyways, it’s very inevitable you see.
19:15 HdkR: 💃
20:01 calim: hey ... does tracing the nvidia binary driver to get command buffers and shader code still work or did they start hiding things better or are the tools out of date?
20:02 calim: I tried to get nsight graphics to show me shader assembly but it won't comply
20:07 karolherbst: calim: soooo... you can do a core dump and search for instructions...
20:07 karolherbst: they usually keep multiple copies of shaders in memory for a bit
20:08 karolherbst: but yeah.. we don't really have a reliable way of dumping shaders atm
20:17 Mary: calim, karolherbst: If you have a pipeline cache you can actually do something with https://gitlab.freedesktop.org/nouveau/nv-shader-tools
20:18 calim: I can make a pipeline cache yeah it's my own code
20:18 Mary: nvparse *might* work for you then
20:18 Mary: for shader code at the very least
20:19 calim: karolherbst so the whole tracing machinery (what used to be renouveau and dedma etc.) is not working anymore? whyyy
20:19 Mary: we have https://gitlab.freedesktop.org/nouveau/envyhooks now btw
20:19 mhenning[d]: If you have spirv, that's even easier with nvdump in that repo
20:20 mhenning[d]: ("that repo" being nv-shader-tools)
20:20 calim: mhenning[d]: well I have spirv but how does that help if it's not assembled?
20:21 mhenning[d]: nvdump creates a vulkan context, compiles the spirv with some default options, and then disassembles the cache
20:22 calim: ah that's useful
20:23 calim: how do they upload shaders nowadays then? if you can dump command buffers the shaders are usually either inline or in some other mapped memory
20:24 calim: at least they're not uploading spirv to a shader compiler in the firmware :D
20:31 calim: what prompted replacing the stuff in envytools with envyhooks, I need a histroy lesson lol :3
20:32 calim: what did I miss in the last 10 years pleasethanks :)
20:38 mhenning[d]: envytools fell into a state of disrepair. envyhooks takes advantage nvidia's open source kernel module, which is fairly new
20:47 calim: ah, that makes sense, thanks!
21:21 calim: so if envytools isn't working anymore, how do you decode the command buffers from envyhooks?
21:26 airlied: nv_push_dump in mesa
21:26 calim: ah, you hid it in there, thanks
22:07 mwk: holy shit I haven't seen you in ages
22:09 mwk: yeah um. I kinda haven't touched anything nvidia-related in 8 years or so myself.
22:12 calim: haha yes I've been kind of ... away
22:14 karolherbst: calim: well the old command buffer tracing infra was valgrind based, sooo...
22:15 calim: I vaguely remember valgrind-mmt ... I had a look at the command buffers for fun but I can't see any shader uploads, they must all be in dma transfers that aren't captured by envyhooks
22:16 karolherbst: yeah..
22:16 calim: and everything uses this annoying mme
22:16 karolherbst: but we can just compile spirvs to gpu binaries, so it's not that big of an issue
22:16 karolherbst: pr compile ptx to gpu binaries
22:16 calim: I figured out one that likely sets the shader address but then I can't see any uploads near there so yeah I'll just go try the spirv compilation
22:17 karolherbst: looking at their binaries is more helpful in terms of what kind of optimizations they are doing and the likes
22:18 karolherbst: are you interested in a specific instruction or something?
22:20 karolherbst: I already forgot what I wanted to look at when I get home.. ohh right.. ULDC and stuff..
22:21 karolherbst: ohh why integer dot product is so damn slow
22:51 gfxstrand[d]: What the hell?!? nouveau.ko is segfaulting in getparam()
22:51 gfxstrand[d]: Blowing up on nvkm_gr_units
22:52 gfxstrand[d]: Fedora 44, latest kernel
22:52 gfxstrand[d]: `Linux animal 6.19.14-300.fc44.x86_64 #1 SMP PREEMPT_DYNAMIC Thu Apr 23 15:17:50 UTC 2026 x86_64 GNU/Linux`
22:53 mhenning[d]: Haven't seen that before. I'm currently on 7.0.3
22:53 gfxstrand[d]: Everything is fine and then RIP
22:54 gfxstrand[d]: My Blackwell box is fine. Same kernel.
22:54 gfxstrand[d]: I guess I'll CTS there for now
22:55 mhenning[d]: What card is failing?
22:55 gfxstrand[d]: 3060
22:55 gfxstrand[d]: But that's the box I have 2 cards in so it could be that somehow
22:56 mhenning[d]: Okay. Logs say I previously had 6.19.14 on this machine without seeing the issue, although this is also on blackwell
23:02 mohamexiety[d]: nothing here on fedora 44 blackwell too
23:05 airlied: that suggests fw loading failed
23:05 airlied: or you fell back to the non-accelerated path somehow
23:05 airlied: probably want to dig in dmesg a bit earlier
23:06 airlied: or boot with nouveau.debug=trace
23:11 calim: karolherbst: I was mainly interested in how well my shaders get optimized
23:12 calim: I very much dislike the idea of profiling something when I don't even know what it actually is