00:14 mlankhorst: imirkin: yeah but nouveau gets all the formats etc from the kernel driver, so that part is fine :P
00:14 mlankhorst: resource_from_fd is dri3 though
00:15 imirkin: mmm.... maybe i got the name wrong
00:15 imirkin: resource_from_somewhere_else() :)
00:15 mlankhorst: from_handle :P
00:18 imirkin: anyways, sounds like it might be the 16 vs 32 bit attachments issue
00:18 AnotherLinuxUser: @imirkin: Well, that is interesting... now, Xorg crashes at xdm login. No login screen... but the following 3 lines comes up:
00:19 imirkin: anholt had a clever solution to that problem for vc4... basically counting the number of reads and writes or something
00:19 AnotherLinuxUser: nouveau E[ DRM] can't find encoder to force DVI-I-1 on!
00:19 AnotherLinuxUser: ... the above line is repeated 3 times... probably one for each of the crtc output that the routine is called on?
00:20 imirkin: mmmm... maybe. i wouldn't worry about it
00:20 imirkin: probably once per vbios script execution or something
00:20 imirkin: doesn't really matter
00:20 imirkin: can you post the Xorg log with the crash?
00:24 imirkin:zzz
00:26 AnotherLinuxUser: @imirkin: A quick peek at the Xorg.0.log shows the following: (EE) NOUVEAU(0): 964: Given depth (32) is not supported by this driver
00:26 AnotherLinuxUser: @imirkin: I guess that answers that question about 32bit.
00:28 AnotherLinuxUser: @imirkin: I saved it and can send the file to you another time. It seems like you need some zzz.
04:49 titan38__: hi! i tried to get dri3 working with xorf 1.1.16 and nouveau 1.0.11 but xorg still uses dri2. is there anything missing for dri3?
05:05 pq: titan38__, what did you use to check for dri3?
05:29 titan38__: pq: xorg.log says dri2 extension is loaded that and also tested breaking on dri3_swap_buffers with glxinfo under gdb (https://wiki.ixit.cz/d3d9_dri3)
05:30 titan38__: pq: the problem is that DRI3Open fails
05:30 pq: alright, I've just heard even when dri3 works, there will be nothing about it in logs, so you need something else to check
05:33 titan38__: pq: hum the gdb tricks does not work?
05:33 pq: I didn't say that, I don't know about that - I'd assume it's a good test
05:37 titan38__: ps: ah ok
06:15 infinity0: so yeah, reclocking fermi didn't work :( 07 and 0f just gave gfx corruption and keyboard input stopped working, though mouse still worked, as did alt-sysrqb-B
06:15 infinity0: back to 03 pstate :(
06:15 infinity0: and 10fps sc2
06:38 RSpliet: infinity0: that's why it's labelled unsupported...
08:56 IronTree9: hey guys. on a old laptop (toshiba portage m200-s838) and the nouveau only works if I completely disable acpi support in the kernel
08:56 IronTree9: with acpi support I get nouveau E[ VBIOS][0000:01:00.0] 0xcab9[ ]: unknown opcode 0x06
08:58 IronTree9: tried kernel versions 3.16.2 and 3.18.7
09:10 alexandernst: Thanks to everybody that helped my those days with my dual GT620 adventure. I gave you a hard time asking so many questions, but you kept answering when possible :)
09:10 alexandernst: At the end I finally got all my 4 monitors working, but sadly the performance is really bad, and so I was forced to buy a GTX750 that has 2 DVI and 2 HDMI ports and just use that one.
09:10 alexandernst: Anyways, I don't regret giving the dual NVIDIA a shot and playing with nouveau for a few days. I couldn't get what I want, but I know you're trying your best.
09:10 alexandernst: And that's why I want to donate both GT620 and a GT8600, if anybody is willing to take them and use them to improve nouveau.
09:34 alexandernst: I just saw http://nouveau.freedesktop.org/wiki/HardwareDonations/
09:34 alexandernst: from the listing there, it seems that "Samuel Pitoiset" could be interested in all the cards I mentioned
09:34 alexandernst: how can I reach him?
09:35 joi: hakzsam: ^
09:35 alexandernst: hakzsam: ping (can you read last 4 lines I wrote? Maybe you'll be interested)
10:07 imirkin: infinity0: my understanding is that with gallium-nine, sc2 goes much faster than with wined3d. you could give it a shot.
10:27 AnotherLinuxUser: @imirkin: Hi, are you awake?
10:27 imirkin: yes
10:27 imirkin: as i hope you figured out, use "24", not "32". i guess i was wrong about how X worked -- been a while since i've written that stuff in configs
10:29 AnotherLinuxUser: @imirkin: Yes, I got logs from the GPU lockup that occured using 24 bits. I will fpaste them for you.
10:35 AnotherLinuxUser: @imirkin: Xorg.0.log is at http://ur1.ca/jqwv3
10:40 AnotherLinuxUser: @imirkin: dmesg log is at http://ur1.ca/jqwvw
10:42 AnotherLinuxUser: @imirkin: My IRC chat skills are immensely rusty! Is there a way to private chat instead?
10:45 hakzsam: alexandernst, hi :)
10:45 alexandernst: hello there! :)
10:45 alexandernst: so... I have some old cards (actually pretty new, but I can't use them neither return them to the store)
10:46 hakzsam: nv50?
10:46 alexandernst: let me see...
10:46 hakzsam: ok :)
10:46 alexandernst: So, The old one is a NVIDIA 8600GT, so that is a NV84
10:46 alexandernst: which I guess is NV50 ?
10:47 hakzsam: yes
10:47 alexandernst: hakzsam: this is it: http://old.vgamuseum.info/images/zaatharen/nvidia/8600GT_fhq.jpg
10:48 alexandernst: hakzsam: and I also have 2 identical Asus NVIDIA GT620, which is...
10:48 alexandernst: NVD9 if I read correctly the CodeNames table in Nouveu's page
10:48 alexandernst: so, NVC0 family
10:49 hakzsam: GT620 is c0 or d7, not sure
10:49 hakzsam: according to the CodeNames pages
10:49 alexandernst: wiki says it's a renbarnded 520
10:50 hakzsam: ah ok
10:50 alexandernst: http://en.wikipedia.org/wiki/GeForce_600_series
10:50 hakzsam: anyways, I don't have these chipsets :)
10:50 alexandernst: do you want all 3 of them?
10:51 hakzsam: if you don't need them, it would be awesome, yeah
10:51 alexandernst: cool :)
10:51 alexandernst: where do I send them?
10:51 alexandernst: and do you preffer any service or should I just search the cheapest one?
10:52 hakzsam: go private for the shipping address ;)
10:52 alexandernst: ;)
10:57 AnotherLinuxUser: imirkin how can chat private on this irc webchat?
10:57 imirkin: hakzsam: GT620 is likely to be a GF108 (nvc1)
10:57 AnotherLinuxUser: s/can chat/can I chat/
10:57 imirkin: AnotherLinuxUser: no clue, depends on the webchat client i suppose
10:58 imirkin: AnotherLinuxUser: keep it in channel... i come and go, others can be (and often are!) helpful
10:58 imirkin: nouveau E[ DRM] GPU lockup - switching to software fbcon
10:58 imirkin: there ya go! that's what i was looking for!
10:58 AnotherLinuxUser: ok, I sent pastes to you... were you able to retrieve them.
10:58 hakzsam: imirkin, are you sure about this? because I already have a GF108 and someone else could be interested
10:58 imirkin: hakzsam: no. GT610 is a GF108 though.
10:59 imirkin: 0f01 GF108 [GeForce GT 620]
10:59 imirkin: 1049 GF119 [GeForce GT 620 OEM]
11:00 hakzsam: ok
11:01 hakzsam: well, no a big issue :)
11:05 AnotherLinuxUser: are you guys collecting cards to test the nouveau driver on?
11:12 AnotherLinuxUser: ... play the latest GPU intensive games?
11:13 AnotherLinuxUser: ... mine bitcoins?
11:14 imirkin: unfortunately what we need the most is time, not hardware
11:15 AnotherLinuxUser: ... act as another CPU node in a multi-processor setup?
11:15 AnotherLinuxUser: Aaahh, I see.... you need developers.
11:15 imirkin: gtg
11:18 Yoshimo: throwing money for cards or cards themselfes would be a lot easier for us
11:28 AnotherLinuxUser: Easier in the short run, maybe... but in the long run, hmmmm...
11:33 Yoshimo: but if you don't have coding experience and want to help, that is the best thing you can do, isn't it?
11:36 AnotherLinuxUser: As I said, in the short run, maybe, but in the long run, that is a more philosophical question.
11:48 hakzsam: AnotherLinuxUser, I'm working on performance counters, not bitcoins ;)
11:56 AnotherLinuxUser: :) @ hakzsam
11:56 AnotherLinuxUser: Is the GeForce 7300 LE a NV46 card?
11:57 AnotherLinuxUser: ... so it is a part of the NV40 family?
12:01 AnotherLinuxUser: ... and what is the connection/compatibility, or lack thereof, between NV40 cards and NV04 cards?
12:17 KingEdgar: how is maxwell support right now >.>?
12:27 KingEdgar: nvm, I found enough information eventually
12:28 RSpliet: KingEdgar: sorry, I was distracted with tinkercad
12:28 KingEdgar:nods.
12:28 RSpliet: it's not well supported, mostly because firmware is missing
12:28 KingEdgar: Oh?
12:28 RSpliet: NVIDIA is looking into providing firmware binaries
12:28 RSpliet: but it's taking a bit too long :p
12:28 KingEdgar: oh :/
12:29 RSpliet: (trust me, we've created that firmware for older cards, it's not evil)
12:30 KingEdgar:nods.
13:14 imirkin: KingEdgar: which maxwell? GM20x or GM10x?
13:15 KingEdgar: GM20x
13:15 imirkin: yeah, what RSpliet said.
13:15 KingEdgar: lol
13:15 imirkin: GM107 is better supported... still have the firmware issue, but it's resolvable by an individual
13:15 imirkin: whereas the GM20x issue isn't :)
13:49 giuseppe_82: hello
13:52 giuseppe_82: i wanted to know if options showed in "man nouveau" command are the only ones I can use on xorg configuration file. Can I possibly define other options like "UseFastTLS" or Backingstore or UseEvents I used with nvidia driver? Thanks
13:53 imirkin: it won't hurt to have those defines in place
13:53 imirkin: but they will have no effect
13:53 giuseppe_82: so all options except the ones on config are useless. Is that right?
13:54 imirkin: mmm... it's conceivable one or two may have escaped the man page
13:54 giuseppe_82: i mean,, on the man
13:54 giuseppe_82: 1
13:54 imirkin: but there's no correlation between blob driver ones and nouveau ones
13:55 giuseppe_82: i'm actually running nouveau with my ooold geforce4 mx 440 agp
13:55 giuseppe_82: (it's the only driver working with xorg>1.12)
13:55 imirkin: k. should work largely fine.
13:55 imirkin: that's a nv17, right?
13:56 imirkin: the 3d driver is missing support for a few features it could potentially support
13:56 giuseppe_82: do you know any options, or kernelmodules I could activate to get it faster?
13:56 giuseppe_82: it's NV18
13:57 giuseppe_82: actually i'm using Option "PageFlip" "on"
13:57 giuseppe_82: Option "HWcursor" "off"
13:57 giuseppe_82: Option "AccelMethod" "exa"
13:57 giuseppe_82: Option "GLXVBlank" "on"
13:57 giuseppe_82: Option "NoAccel" "off"
13:57 giuseppe_82: Option "SwapLimit" "2"
13:57 imirkin: not really... it's not exactly a speedy card
13:57 imirkin: ohhhh
13:57 imirkin: turning off hwcursor is a *big* mistake
13:57 imirkin: if that's an actual option
13:57 imirkin: yeah, it is... and that's gonna make it perform at like... WAY slower rate :)
13:58 giuseppe_82: why? i'm running SWcursor as I didn't see any improvements
13:58 imirkin: you want to use the hw cursor.
13:58 imirkin: trust me.
13:58 giuseppe_82: i'll try it
13:58 giuseppe_82: what's exaclty the difference from SW?
13:58 imirkin: every time you move the mouse
13:59 imirkin: it's the difference between writing the X/Y coordinates into 2 registers on the card
13:59 imirkin: and repainting the 32x32 mouse cursor square, and then compositing the mouse cursor onto the fb again
14:00 giuseppe_82: ok
14:00 imirkin: also i think sw cursor turns off a bunch of stuff in X
14:01 giuseppe_82: i also try to deactivate dri2 but i think it's builtin coz it runs anyway. i checked the xorg log file
14:01 imirkin: not 100% sure what though... it just never comes up, nobody cares about sw cursors since all hw since... a long time... has hw cursor support
14:01 imirkin: why would you want to do that?
14:01 imirkin: you probably want to turn off any sort of GL compositor you're using though
14:01 giuseppe_82: i was trying to configure X for best performance
14:01 imirkin: performance of what?
14:02 giuseppe_82: graphically i mean
14:02 imirkin: :)
14:02 imirkin: what application
14:02 giuseppe_82: i tested it with gtkperf
14:02 imirkin: ah
14:02 giuseppe_82: :)
14:02 imirkin: i doubt 3d accel would have an effect on gtkperf either way
14:03 imirkin: if the cpu in the box is much newer than the card, it may be faster to do 3d on the cpu though
14:03 giuseppe_82: i'm actually running X Server 1.16.0 on lubuntu 14.10
14:03 giuseppe_82: not so much newer it's an Athlon XP 3000
14:03 imirkin: (or if you need shaders or something)
14:03 giuseppe_82: :)
14:04 giuseppe_82: how should I turn off GL compositor?
14:04 imirkin: sorry, not sure -- lots of fancy new window managers/desktop environments/whatever-you-want-to-call-thems...
14:04 imirkin: i just stick with WindowMaker, and that's never steered me wrong
14:05 giuseppe_82: i have lightdm as display manager
14:05 imirkin: that's the xdm clone right?
14:05 imirkin: not sure that would have anything to do with it
14:06 giuseppe_82: it's lxde
14:08 imirkin: ok, that's among the multitude of new-age stuff i know little to nothing about
14:16 imirkin: skeggsb: any thoughts on userptr support in nouveau? seems to be what all the other drivers are doing now..
14:17 mlankhorst: imirkin: seems like an easy project to do :P
14:18 imirkin: mlankhorst: thanks for volunteering!
14:18 imirkin: it does seem like it should be modestly simple on nv50+
14:18 mlankhorst: just copy paste from radeon or something :p
14:19 imirkin: hehehe
14:19 imirkin: and stick waits on random fences
14:19 mlankhorst: I should really get something that runs nouveau at my gf's place
14:19 imirkin: i guess that memory is just pinned and you don't have to ever really worry about it
14:20 mlankhorst: I've been meaning to add android fences too
14:20 imirkin: there's no eviction (unless there's vm pressure, in which case we're screwed anyways)
14:20 mlankhorst: oh speaking of which..
14:23 mlankhorst: maybe tegra k1 if it ever ships :P
14:23 imirkin: you can get one now... available in stores near you!
15:02 hakzsam: I just tried to trace every MMIO bar access using QEMU and VT-d and I must say that's really awesome! The trace is huge, but this works really fine. :)
15:04 hakzsam: looks like (from the host) -> 28286@1424213721.877195:vfio_region_write (0000:05:00.0:region0+0xa404, 0xdeadbeef, 4)
15:06 hakzsam: I'll try to post a howto on my blog in few days
15:37 imirkin: skeggsb: any ideas on https://bugs.freedesktop.org/show_bug.cgi?id=28095 ? seems like among other things, edid isn't working on that nv04
15:47 skeggsb: imirkin: none yet, but i'm looking
15:48 imirkin: he also has the more serious issue of the display being totally fubar'd
15:48 imirkin: which got broken at some point between the 'nv' driver and now :)
15:48 skeggsb: imirkin: i also have nfi how pre-nv50 display works, at all
15:49 imirkin: "occasionally"? :)
15:49 skeggsb: haha, i more meant the hardware than our code
15:58 mlankhorst: don't you have the evo stuff now for < nv50? :P
15:58 skeggsb: eh?
15:59 skeggsb: there is no evo pre-nv50
15:59 mlankhorst: ah
15:59 mlankhorst: ok that makes it worse :P
15:59 mlankhorst: will atomic be just for >= nv50 then?
15:59 skeggsb: it's some messed-up fusion of legacy vga, extended vga, and priv registers
16:00 skeggsb: yes, unless someone else does it
16:00 mlankhorst: doubtful :p
16:00 skeggsb: nvidia actually *do* stick a nicer interface on top of the horrible hw for <=nv40
16:01 skeggsb: they have this NvVideoLutCursorDac software object class, which is used much like evo's channels are
16:01 mlankhorst: ah
16:02 mlankhorst: nvkitchensink? :P
16:02 skeggsb: NvIReallyLikeLongDescriptiveClassNames
16:02 xexaxo: everything&thekitchensink
16:02 skeggsb: they seem to have gotten over that though :P
16:03 aaronp: Calling that thing "much like evo" is being pretty generous....
16:03 skeggsb: aaronp: haha well, i meant "there's a command stream", but that's where it ends
16:03 aaronp: That's fair. :)
16:06 imirkin: skeggsb: i'm guessing that "nouveau D[ DRM] native mode from largest: 0x0@0" is probably the sign that edid didn't go so well
16:06 imirkin: i haven't plugged my nv05 in for a while, i guess i could do that and see if it seems like it works.
16:07 mjg59: VideoLutCursorDac is a pretty solidly descriptive name
16:10 mlankhorst: yeah but they're different parts of hw :P
16:14 mjg59: VideoLutCursorDacTriangleTextureShader
16:28 AnotherLinuxUser: clear
16:29 AnotherLinuxUser: imirkin: Do you need anything else from me that could help with the NV46 card?
16:30 imirkin: AnotherLinuxUser: i doubt i'll be able to help you, sorry
16:32 AnotherLinuxUser: ok... I did notice that it seems that nv46 is somehow linked to the nv04 driver code... some init_tile code(?) ...
16:34 AnotherLinuxUser: ... somehow it didn't seem right to me, since nv46 would be a part of the nv40 family of cards and nv04 would be too old.
16:36 imirkin: AnotherLinuxUser: yeah, a bunch of code is shared between generations
17:56 infinity0: imirkin: but my card is running on 50MHz when the specs say it should run at 880MHz (similar numbers for memory), it has an intrinsic disadvantage
17:57 infinity0: i mean i could try gallium but i doubt it would improve things *that* much
17:57 imirkin: infinity0: it won't execute anything faster, but people have noticed that at a particular "execution speed", gallium-nine can be considerably faster than wined3d
17:57 imirkin: due to fewer things being executed? who knows.
17:58 infinity0: RSpliet: what sort of tools would one use to try to improve things? like, how did you know to make those specific changes to the other cards?
17:59 infinity0: yeah i suppose i could try it at some point
17:59 imirkin: infinity0: basically you collect mmiotraces, and try to make sense of them
17:59 infinity0: imirkin: do you have specific pointers outside of what i could find by searching online?
17:59 imirkin: infinity0: pointers for what?
17:59 infinity0: and do you need special equipment for that?
17:59 infinity0: to compile gallium-nine with wined3d
18:00 imirkin: https://wiki.ixit.cz/d3d9
18:00 imirkin: for the mmiotrace: https://wiki.ubuntu.com/X/MMIOTracing
18:01 infinity0: ah, thanks
18:01 skeggsb: wine developers' stance on nine is unfortunate, it really is rather nice
18:01 imirkin: meh, i get where they're coming from. it doubles their test surface
18:01 imirkin: and they feel like they can get all the benefits without doing that
18:02 imirkin: fwiw i think they're wrong, but i certainly won't be convincing them of that
18:02 skeggsb: yeah, not worth wasting the energy debating that one :P
18:02 skeggsb: many others have tried
18:03 infinity0: so, one reason i don't think gallium will work that well, is even the native-linux steam client runs dota 2 at about 7fps
18:03 skeggsb: they get the speed with that "csmt" or whatever it is, but do so by abusing way too many cpu cores
18:03 skeggsb: (than what's needed otherwise)
18:04 skeggsb: my vote would be for them to ditch wined3d altogether, and turn their d3d->opengl layer into a gallium->opengl layer and use nine
18:04 skeggsb: buuut, that's a pipe dream
18:04 infinity0: that wouldn't affect the native steam client though right?
18:05 imirkin: nope
18:05 imirkin: native steam client probably uses a similar conversion layer though
18:05 imirkin: [to what wined3d does]
18:07 infinity0: tbh even nvidia drivers are buggy, one major reason i switched is because every [1-20] days, seemingly pretty randomly, i'd come back home to find my X showing gfx corruption, sometimes disabling local input (ssh still worked)
18:07 infinity0: doesn't happen with nouveau :p
18:08 imirkin: that's ok, nouveau has other problems :)
18:08 infinity0: this happened for like 2 years, across many multiple version between 200-340-whatever they are at now
18:08 infinity0: heh
18:08 imirkin: skeggsb: their main business is wine on os x, so that probably won't happen
18:09 imirkin: skeggsb: also even sticking on linux, won't work with intel, at least until someone decides to fix up ilo
18:09 skeggsb: imirkin: that's what the gallium->ogl driver i mentioned is for ;)
18:10 airlied: some thing wine does could be cut down with GL4.5 extensions as well
18:10 imirkin: skeggsb: ah yeah, like virgl
18:10 airlied: ARB_clip_control would let them stop editing vertex shaders in some ways
18:10 imirkin: skeggsb: iirc i made that suggestion at one point... mostly in jest
18:11 airlied: my main thing is their belief that wine can be as fast as nine with same CPU usage seems mostly deluded
18:11 airlied: and diversionary
18:11 imirkin: airlied: you planning on convincing them of that?
18:11 skeggsb: i think they skim past the "with the same cpu usage" part
18:12 imirkin: tbh, the nine guys could have also made things easier on themselves
18:12 airlied: imirkin: I don't even play native games, wine seems like even more trouble :)
18:12 imirkin: like having a more plug & play solution than having big wine patches
18:51 buhman: 02:01:13 skeggsb wine developers' stance on nine is unfortunate
18:51 buhman: stance?
18:52 Leftmost: buhman, the wine developers feel that Nine is not suitable for use as a replacement for wine's d3d9 implementation.
18:52 imirkin: "the attitude of a person or organization toward something; a standpoint."
18:53 buhman: I mean, I want to read where the wine developers say this :D
18:54 Leftmost: I mostly know of it from spending time in the wine channels and listening to their responses to people discussing it.
18:54 imirkin: buhman: hmmm... i think it's been convered on mesa-dev
18:54 imirkin: covered*
18:56 buhman: you appear to have participated ;p
19:24 imirkin: buhman: it's possible :)
19:24 imirkin: skeggsb: when's the last time you booted nouveau on a nv04 or nv05?
19:25 skeggsb: when i did the whole kernel driver redesign thing
19:25 skeggsb: so, a long time ago
19:25 skeggsb: my agp machine is... not fun
19:25 imirkin: like 3.8 or whatever?
19:25 skeggsb: whenever it was, yeah
19:26 skeggsb: i pretty much stick to >=nv40 now
19:26 imirkin: you should get PCI cards :)
19:26 skeggsb: yes, it's been on my todo list for a while actually
19:27 imirkin: ebay's got tons of 'em
19:27 skeggsb: how exactly does one search for "pci" without getting pcie too :P
19:27 imirkin: hehe
19:27 imirkin: by adding things to the search
19:27 imirkin: like
19:27 imirkin: tnt2 pci
19:28 imirkin: or "geforce2 pci"
19:28 imirkin: etc
19:29 imirkin: unfortunately nv2x is all agp (or xbox)
19:29 imirkin: you _could_ of course also get a new agp mobo
19:54 skeggsb: imirkin: haha, hilarious, the first tnt2 i see is $58.64.. why the fuck would someone pay that now?!
19:55 skeggsb: hrm, they're all pretty expensive actually wtf
19:56 imirkin: hmmm... mine are $10
19:56 imirkin: i'm guessing you're seeing US results and shipping is expensive
19:57 skeggsb: yeah, that price was without the shipping too
19:57 skeggsb: there's a gf2 mx400 pci, AU$6.37 + $40.37 postage haha
19:57 imirkin: http://www.ebay.com.au/itm/Palit-daytona-TNT2-M64-32mb-PCI-video-card-/380377751156?pt=AU_Components&hash=item589047de74
19:58 imirkin: oh, but it's agp
19:58 imirkin: despite the title
19:58 skeggsb: heh
23:45 wolter: I have both nvidia and nouveau modules installed, how can I preselect from the grub menu which to boot with?
23:46 imirkin: some distros parse the kernel cmdline and you can tell them which thing to blacklist
23:46 imirkin: your userspace has to be different too though depending on which you want to use, so it'd be a bit tricky
23:53 wolter: I'm using ubuntu 15.04, with systemd
23:53 wolter: I created a systemd service for this purpose (the service writes a blacklist in modprobe.d and a xorg.conf in xorg.conf.d), but it seems I haven't set it to launch soon enough
23:55 pq: you also need to switch all GL-related libraries - basically everything the nvidia driver package installs.