00:47imirkin: this is embarassing. i can't even manage to properly upload data.
00:47imirkin: or maybe i can't download it
02:11imirkin: skeggsb: any idea why i'd get a DST2D_LINEAR_MISMATCH when using 2d blit engine with a *tiled* dst but which is in GART? (and this is G84, so none of those nv50 sysmem problems)
02:12imirkin: same everything but with dst in VRAM works fine. the source is untiled in GART.
02:19imirkin: so ... it looks like when *either* (or both) of the surfaces have a non-0 memtype, 128k byte lines don't work with m2mf
02:20imirkin: additionally it appears that there are some restrictions with m2mf about it working with tiled sysmem buffers
02:20imirkin: not that those would be common, but ... that's news to me
18:27imirkin: wow, m2mf also fails at height > 4k, even without any tiling at all!
18:28imirkin: if there are any offsets
18:57imirkin: nevermind on that. the height > 4k thing was my bad.
19:13imirkin: and double-confirmed that m2mf with 8192-wide 64-bit textures works fine.
19:15imirkin: so at least on G84, m2mf can't handle (a) non-linear in GART and (b) non-linear textures > 64kb "pitch"
19:15imirkin: for the former, you get an error, for the latter the texel at x = 4096 contains the source value from x = 0, and so on
23:38lomb: hey everyone, I'd like to write an Article about the best supported nouveau cards
23:39lomb: unfortunatelly the best supported card that you can still buy brandnew is the GT 710
23:39lomb: are there any other cards you know that people can still get?
23:39lomb: without having to buy secondhand cards?
23:40RSpliet: lomb: you're going to have to do some cross referencing from the Wikipedia website "list of NVIDIA GPUs" or whatever it's called
23:40lomb: RSpliet: you mean I have to look up every single card?
23:40RSpliet: But anything in the Kepler or first gen Maxwell (GM10x) ranges are the good stuff
23:40imirkin: lomb: did a quick check on newegg
23:40lomb: I know
23:40imirkin: and it appears you're right - geforce gt 710 is like the only kepler-era board sold
23:40RSpliet: lomb: journalism is hard work :-P
23:40lomb: I mean, I know which cards are good :P
23:40lomb: I just don't know which ones you can still buy
23:40lomb: on Amazon, etc
23:41imirkin: on ebay, you can buy whatever, but yeah - often second hand
23:41lomb: RSpliet: indeed :P
23:41imirkin: but that's a better deal anyways
23:41lomb: imirkin: true, but I'm trying to find off-the-shelf parts
23:41lomb: although to be honest: for many people a gt 710 is more than enough
23:41imirkin: lomb: ooh, looks like GT 640 is also still sold
23:42imirkin: (again, on newegg)
23:42lomb: especially since there is a version with a VGA port and one with 4 hdmi ports
23:42imirkin: which is basically a faster version of the GT 710
23:42imirkin: probably same chipset -- GK208
23:42lomb: hey, I didn't know that
23:42imirkin: but more better fast RAM
23:42imirkin: e.g. https://www.newegg.com/p/17Z-001W-00DU5
23:43imirkin: otth, that's just part of the newegg marketplace
23:43imirkin: not actually sold by newegg
23:43lomb: I wonder how well this card actually performs
23:43imirkin: so yeah, who knows. (that also feels way overpriced.)
23:43lomb: why the hell is this card so damn expensive?
23:43imirkin: well, $170. so less than that in euro. but yeah, way overpriced
23:43imirkin: probably the price at release, never updated it with the times
23:44lomb: $, yeah
23:44lomb: not € :P
23:44lomb: you can get a gtx 780 for that on ebay
23:44imirkin: GT 710 appears to still be actively produced
23:44imirkin: (or they just have a TON of stock to get rid of)
23:44lomb: imirkin: yes, Asus even produces a version with 4 hdmi ports now
23:44lomb: or that :P
23:44imirkin: it's based on GK208, which is a super-low-power chip
23:45imirkin: (comparatively. not compared to SoC embedded gpu's.)
23:45imirkin: you can often get them fanless/etc
23:45lomb: yes, I saw that
23:45lomb: well, sucks :L
23:45imirkin: of course the population who wants fanless GPUs tends to be completely separate from the one who wants GTX 780 (or RTX 3080)
23:45RSpliet: Yeah they stopped doing low-end GPUs shortly after Kepler I think, with Intel taking that chunk of the market with their IGPs
23:45lomb: but to be honest: most good computer hardware can only be gotten used
23:46imirkin: RSpliet: nah, GP108 is low-end too -- GT 1030
23:46lomb: btw, how does it look like on the Quadro side?
23:46imirkin: i'm actually running on one right now :)
23:46imirkin: lomb: how do you mean?
23:46lomb: imirkin: I have to check if they still make some kepler quadro cards
23:46imirkin: it's the same chips, just more ram, or better fans/etc
23:47RSpliet: imirkin: hmm, fair. That's Pascal though, still 2/3 gens old
23:47imirkin: lomb: you can get them second-hand. i've been eyeying a K4000 or K4200 myself
23:47lomb: how well do they work with nouveau?
23:47imirkin: RSpliet: well, volta doesn't count ... i guess yeah. ampere/pascal
23:47imirkin: lomb: as well as non-quadro
23:47imirkin: it's the same everything
23:47imirkin: some quadros may have extra-fancy features, which will generally not be supported by nouveau
23:48imirkin: e.g. some have frame-locking, stuff like that
23:48lomb: hey, the K4000 is relatively cheap
23:48imirkin: (across multiple outputs)
23:48lomb: around 8ß€
23:48imirkin: hm, i thought it was like $60
23:48lomb: but yeah, not that useful with nouveau then I guess :L
23:48imirkin: the nice thing about a lot of those quadro's is that they're single-slot
23:48lomb: so consumer cards it is :P
23:49imirkin: so i actually prefer them
23:49lomb: you can often get a gtx 770 for the same price
23:49RSpliet: lomb: with nouveau they'd work as if they are GeForce
23:49lomb: RGspliet: OH
23:49imirkin: the K4000 is probably on-par with GTX 770 in terms of perf though -- i'd check
23:49lomb: because nouveau ignores the bios, right?
23:49RSpliet: lomb: not quite no
23:49imirkin: lomb: there is no difference between quadro and non-quadro from a driver perspective
23:49lomb: I heard that often those cards are just using a different bios
23:50lomb: that's what I meant
23:50imirkin: it's marketing, as well as just different availability
23:50lomb: and nouveau ignores that
23:50imirkin: i.e. you can't get a single-slot GTX 770
23:50lomb: so the cards work the same
23:50imirkin: for any amount of money
23:50imirkin: but you can get a K4000 which is single-slot iirc
23:50lomb: hmm, I think I'll eventually buy a K4000
23:50lomb: to see how it performs
23:50lomb: would be amazing to attract more people to use nouveau
23:51imirkin: nouveau has serious usability issues for the average person
23:51imirkin: i would strongly recommend AMD (or Intel)
23:51lomb: because right now when you go on the internet you only get videos and articles that show you how bad nouveau compares to the proprietary driver
23:51RSpliet: lomb: ^ we've been doing the opposite in this channel :-P
23:51lomb: because they are using bad cards that you can't reclock
23:51imirkin: even on kepler, we're way behind
23:51imirkin: with full reclocking
23:52lomb: imirkin: it works well enough for me though
23:52RSpliet: lomb: as long as NVIDIA is unwilling to provide redistributable firmwares, there's no hope for nouveau becoming a serious contender on newer GPUs
23:52lomb: I can do anything I want with nouveau
23:52lomb: except using vulkan
23:52imirkin: turns out having a 1k-person team with full access to docs can produce a better compiler than like 1/2 a person
23:52imirkin: lomb: try running plasma
23:52lomb: and not having to rely on scripts that reclock my card for certain applications
23:52lomb: imirkin: I used it only once and I didn't like it :L
23:52lomb: I'm using Mate
23:52imirkin: lomb: yeah, but plasma + nouveau = death
23:53imirkin: also if you try to use e.g. mpv with GL output -- death
23:53lomb: imirkin: hmm, which card are you using?
23:53imirkin: lomb: i have a drawerful of NVIDIA GPUs
23:53imirkin: you name it, i got it
23:53RSpliet: Riva 128!
23:53imirkin: except if it's new, or high-end
23:53imirkin: RSpliet: bleah
23:53RSpliet: My dad has one, I should pinch it once he retires his PC
23:53lomb: maybe my usecases just work very well with nouveau :L
23:54imirkin: RSpliet: otoh, here's my currently-plugged in selection: https://hastebin.com/agahoqezut.yaml
23:54imirkin: lomb: yeah, it works well for me too :)
23:54imirkin: but it doesn't sound like you are the average end user
23:54RSpliet: TNT2, nice!
23:54lomb: which one is your favourite card btw?
23:54lomb: oh god xD
23:54lomb: is the TNT2 even supported by nouveau? xD
23:55lomb: i'm sure that even Mate = death on this card
23:55imirkin: i mean, don't expect GL4.5 on it
23:55lomb: or is it?
23:55imirkin: but you get (a) full modesetting and (b) most of GL 1.2 or 1.3
23:55imirkin: MATE could not make use of its acceleration
23:56imirkin: so you'd just get the modesetting bits
23:56imirkin: even xf86-video-nouveau's EXA impl only implements a handful of acceleration things on pre-NV10 hardware
23:56imirkin: (NV10 = GeForce 256)
23:56RSpliet: imirkin: I meant to ask, out of curiousity: what's your current interest in the various upload mechanisms?
23:57lomb: I don't have many older nvidia cards
23:57RSpliet: which itch are you trying to scratch? :-)
23:57lomb: but I could include a full list of how well my old cards work
23:57imirkin: RSpliet: i was trying to figure out where m2mf was broken
23:57lomb: thanks for the info on the TNT2 btw
23:57imirkin: but to do that, you can't really use m2mf to do the upload in the first place
23:57RSpliet: imirkin: for G84?
23:57imirkin: RSpliet: yes
23:57imirkin: RSpliet: https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8396/diffs?commit_id=97a5e0dd8f7b5c7a19077e84371d5a430147605d
23:58imirkin: various GLES3 "accuracy" tests were failing for some weird reason. turns out we couldn't read out the surface back to the CPU where the accuracy data lived.
23:58lomb: it sucks that there seems to be only one person sending patches to nouveau right now :(
23:59RSpliet: imirkin: ah! That makes sense :-)
23:59lomb: I wish I could help
23:59ccr: developers don't grow on trees
23:59imirkin: lomb: if you see something broken, send patches :)
23:59lomb: but I don't know anything regarding the subject
23:59RSpliet: lomb: it's not quite that dire. There's imirkin doing a lot of good stuff, Red Hat employs three contributors, and every now and then some people at NVIDIA throw some patches at us