01:44 karolherbst: :O furmark gets my gpu to 90°C with the blob
01:44 karolherbst: and 67W
01:44 karolherbst: why is nouveau so competetive in that benchmark :/
01:46 glennk: karolherbst, its not a particularly shader heavy benchmark, just eats fill rate on modern cards
01:47 karolherbst: glennk: yeah, but that power consumption
01:47 karolherbst: never saw it going that high before :D
01:47 karolherbst: well
01:47 karolherbst: it is shader heavy
01:48 karolherbst: I think, because when I disable that thingy in the middle, the fps just skyrockets
01:49 karolherbst: glennk: would be fill rate being eating also result in a high "core" load?
01:50 karolherbst: ohh wait, maybe it does, because I also see a high memory load
01:52 karolherbst: now would be a good time to actually see what the blob does better here
01:53 karolherbst: it seems to deal a tiny higher memory load though
01:55 glennk: no idea what "core" measures
01:57 karolherbst: it's the pdaemon counter
01:57 karolherbst: which means, its just the thing nvidia-settings displays as the core load
01:57 karolherbst: PGRAGH I think
01:57 karolherbst: and PCOPY
01:59 karolherbst: or was it PCOPY and GR?
01:59 karolherbst: well GR is graph, isn't it? :D
02:00 skeggsb: NV_PPWR_PMU_IDLE_MASK_GR_ENABLED | NV_PPWR_PMU_IDLE_MASK_CE_2_ENABLED
02:00 skeggsb: whatever that measures
02:00 karolherbst: yeah
02:00 karolherbst: I get 5% load on the PCOPY thingies though
02:00 karolherbst: so 95% on GR
02:00 karolherbst: mhh 100% though
02:00 karolherbst: so it seems to be always active
02:01 karolherbst: anyway, it's only a perf difference of about 4%
02:01 karolherbst: ohh actually 9%
02:03 karolherbst: I think nouveau is pretty good in everything except when it comes to optimized binaries
02:04 karolherbst: pixmark_piano is 25% less perf and this is just a big shader
02:04 karolherbst: no reported memory load (<0.5%)
02:04 karolherbst: just 100% GR
02:05 karolherbst: and nrealy no PCOPY(<2%)
02:05 karolherbst: but this shader generates like over thousends of instructions, so it is a mess to find optimization opportunities
02:07 karolherbst: I would need those perf counters to further analyze this, but I think improving the compiler might be a thing in the end, and maybe the zcull thing too
03:16 karolherbst: mupuf: so, any ideas about the fsrm ?
03:46 mupuf: karolherbst: well well
03:46 mupuf: there must be a way to select the mode
03:46 mupuf: I doubt the FSRM has been redesigned
03:47 karolherbst: yeah I doubt that too
03:47 karolherbst: the fsrm works the same way on my fermi and on my kepler, just the configuration is different
03:47 karolherbst: even the output clock behaves the same
03:49 karolherbst: mupuf: you know that on my kepler the status reg changes it contents really often even on low temps?
03:50 karolherbst: and somehow I managed to let the status reg behave like on my fermi (which changes the content only when the fsrm triggers)
03:50 karolherbst: couldn't reproduce it in a way where I understood what I did though
03:51 mupuf: yeah, but you disabled the idle clock gating ;D
03:51 karolherbst: :D
03:51 karolherbst: ohhhh
03:51 karolherbst: that's what it was
03:51 karolherbst: well
03:51 karolherbst: I just poked the values from the fermi vbios into my kepler
03:51 karolherbst: I was thinking if I just set all the regs to the same it should behave like the fermi one
03:52 karolherbst: maybe I missed something
03:54 karolherbst: mupuf: maybe the only thing we should do is verifying the thresholds, because I don't see the blob caring much about all of this anyway
03:54 mupuf: karolherbst: one way to detect if you are in the mode two or 3 thresholds could be .... to check the mask of the register. But that is definitely a last resort measure
03:54 karolherbst: well
03:54 karolherbst: I can set the values like on the fermi on my kepler
03:54 mupuf: yeah, we should just reproduce what the blob does
03:55 karolherbst: well the blob does nothing much :/
03:55 karolherbst: totally ignoring those regs usually
03:55 karolherbst: but I would feal uneasy about this. Maybe the blob parses the vbios script and does some writes if it detects something is wrong?
03:56 karolherbst: thing is, we need more traces :/
03:56 karolherbst: we don't have many for fermi/kepler
03:56 karolherbst: ohh wait
03:56 karolherbst: I did my find command wrongly
03:57 karolherbst: most of the traces don't contain the string "trace" :D
03:58 imirkin: but they contain the string 'xz' :)
03:59 karolherbst: yeah
03:59 karolherbst: I know
03:59 karolherbst: just forgot about that
04:01 karolherbst: mupuf: what are those "INPUT_*_DIV" fields by the way?
04:02 mupuf: karolherbst: on some cards, the fan management and temperatur econtrol is done outside the chipm
04:02 mupuf: there are input GPIOs that are connected to these input events
04:02 karolherbst: ahh okay
04:02 mupuf: and the regs you see are the dividers to set in this case
04:03 karolherbst: so that even the fsrm can put the fans at a high speed or something?
04:04 mupuf: nope, not possible for the fan
04:05 karolherbst: ohh then it is the other direction
04:05 karolherbst: mhh in pmoreau nvc0 trace the blob only reads PTHERM.FSRM_CFG_5 :/
04:06 karolherbst: but usually PTHERM.I2C_SLAVE.THRESHOLD_2 is also read on other cards
04:08 karolherbst: mupuf: PTHERM.I2C_SLAVE.THRESHOLD_10 <= 0xc8 :D
04:08 karolherbst: this sounds like disable this friggin threshold, we don't need it
04:08 karolherbst: 200°C, seriously
04:08 mupuf: hehe
04:08 mupuf: at this temperature, the components will fall off :D
04:08 karolherbst: right
04:09 karolherbst: but this threshold seems to be unrelated anyway
04:09 karolherbst: I just added it for completness
04:09 karolherbst: it has a value thre_8+1 for me and I wanted to figure out if 8 or 10 is used
04:09 karolherbst: ohhh
04:10 karolherbst: PTHERM+0x10c => 0x49 and then PTHERM+0x10c <= 0x9
04:10 karolherbst: I think this trace might be helpful
04:10 karolherbst: 1 div for each threshold
04:14 karolherbst: wait......
04:14 karolherbst: that's strange
04:15 karolherbst: mupuf: did I told you that it seems like the ptherm reads the vbios script and executes it?
04:15 mupuf: well, that would be perfext!
04:16 karolherbst: yeah, I mean, on my gpu even on nouveau the values are set according to the vbios
04:16 karolherbst: that's not the point though
04:16 karolherbst: in my trace, the blob sets them again :O
04:16 karolherbst: from the host
04:16 karolherbst: and it is clearly the script it is executing
04:17 karolherbst: crit set to 0x6e, read, read, set to 0x68
04:17 karolherbst: vbios: R[0x020480] = 0x0000006e then ZM_MASK_ADD R[0x020480] & ~0xffffff00 += 0x000000fa
04:18 karolherbst: don't see that in any other trace
04:24 mupuf: see, that makes a lot of sense for it to be in a script
04:24 mupuf: but why is it not run at boot? :s
04:24 mupuf: I guess it could theoritically be a problem
04:25 karolherbst: ohhhh wait
04:25 karolherbst: the default values I see on my gpu looks exactly like those fermi cards
04:27 karolherbst: mupuf: https://gist.github.com/karolherbst/9d7c1e534339f5aa65b2
04:27 karolherbst: mupuf: nouveau doesn't parse that script and executes it, right?
04:27 mupuf: well, it depends
04:28 karolherbst: on what?
04:33 mupuf: on whether the main script references this script or not
04:33 mupuf: but I guess nit
04:33 mupuf: so, we need to find a way to get a pointer to this table
04:33 mupuf: and execute it
04:33 mupuf: does nvbios show the scirpt
04:34 mupuf: ?
04:34 karolherbst: yesa
04:34 karolherbst: well
04:34 karolherbst: it gets executed already
04:34 karolherbst: that's why I am asking
04:34 karolherbst: loading nouveau results into the same values than seen with the blob
04:35 karolherbst: mupuf: https://gist.github.com/karolherbst/d2090fcabbd28c1051a1
04:35 karolherbst: ZM_REG_SEQUENCE 0x03
04:35 karolherbst: this is for the thresholds
04:36 karolherbst: ZM_REG_SEQUENCE 0x06 is the low config
04:36 karolherbst: ZM_REG_SEQUENCE 0x02 is CFG_5
04:36 mupuf: well, that's good
04:37 mupuf: it would be nice to see if changing the values after boot and then loading the blob would change anything
04:37 karolherbst: on some tesla cards (and maybe fermi) there are garbage values in the script
04:37 karolherbst: well powering on the gpu doesn't seem to set the values from the vbios
04:37 karolherbst: at least on my gpu
04:41 karolherbst: mupuf: which gpus are in reator currently?
04:41 mupuf: GM206 and GF116
04:42 karolherbst: mhh
04:42 mupuf: karolherbst: the card needs to be POSTed
04:42 karolherbst: and when does that happen?
04:42 mupuf: not sure it gets posted at boot in the optimus case
04:42 karolherbst: by the bios?
04:42 mupuf: yes
04:42 karolherbst: mhh
04:42 mupuf: or by the kernel module, if needed
04:42 karolherbst: can I repost it after that? :D
04:42 mupuf: that's why there are init scripts
04:42 mupuf: yes
04:43 mupuf: vbetracetool can do that IIRC
04:43 karolherbst: ohh vbetool called here
04:43 karolherbst: mupuf: do you think the ACPI methods repost the gpu?
04:44 mupuf: I doubt it
04:44 karolherbst: so does nouveau repost the gpu on every gpu resume then?
04:45 mupuf: likely, yes
04:45 mupuf: as in, someone has to do it anyway :D
04:45 mupuf: so it is likely nouveau and not the bios through the ACPI method
04:46 karolherbst: k
04:59 karolherbst: mupuf: can I disable reposting with nouveau at load time?
05:00 mupuf: not sure you can, you can force a repost though
05:00 karolherbst: yeah that I know :/
05:00 karolherbst: reading source then
05:41 karolherbst: mupuf: yep, after posting the values are changing
05:44 mupuf: not sure what you mean
05:45 mupuf: I wanted to know if, when changing the values after posting but before loading the blob, the values would change
05:45 karolherbst: I mean, after calling init->func->post(init, init->post);, te regs are changing according to the vbios script
05:45 karolherbst: ohh
05:45 karolherbst: well on my gpu yes, because the blob just writes the value
05:45 karolherbst: it doesn't chech the content of those regs
05:45 karolherbst: *check
05:46 mupuf: I see, which suggests it re-runs the script
05:46 mupuf: hmm
05:46 karolherbst: but it only does that on my gpu
05:47 karolherbst: I could check on your gf116 though
05:47 mupuf: well, one is enough to be too much
05:48 karolherbst: yeah, that's what bothers me too
05:49 karolherbst: ohhh
05:49 karolherbst: but maybe that helps us
05:49 karolherbst: because we actually know which regs are important
05:50 karolherbst: mupuf: grep -B15 -A5: https://gist.github.com/karolherbst/e4ad1424a4bcaec6fe27
05:50 karolherbst: it looks so nice :D
05:51 karolherbst: ohh wait
05:51 karolherbst: it is just like the script :/
05:51 karolherbst: seems like the blob just runs the entire script on my gpu
05:51 karolherbst: maybe that's posting but inside the driver or something
05:53 mupuf: yep, sounds very likely
05:54 karolherbst: ho, there is even pcie stuff inside that script
05:54 karolherbst: 0x08c040
05:56 karolherbst: mupuf: any idea what this script could initialize between pdisplay and ppci?
05:57 karolherbst: 0x08d1xx regs
05:58 mupuf: no idea
06:03 karolherbst: mupuf: so what is the plan. Let the stuff be set by posting and just check the values in init and only interfere if the values are insane in any way?
06:15 karolherbst: personally I don't want to solely trust the vbios do contain the right thing, because when something accidentally get stuck in the fan (can happen) and the gpu just overheads for that reason and the gpu breaks, because the fsrm was wrongly configured that would be a really sad case :/
06:26 mupuf: karolherbst: we need to do what the blob does
06:26 mupuf: and mimic it
06:26 karolherbst: yeah well, that would mean reading those values and then do nothing
06:28 karolherbst: except for rare special casees
06:30 karolherbst: mupuf: the other trace where the values were set was from a suspend-resume trace, so I bet this is reposting again
06:30 mupuf: that makes sense
06:30 karolherbst: mine will be for sure reposting, because I am sure the gpu was turned off by bbswitch
06:31 mupuf: well, the good thing is that for once, we just have to read back the state, print that in a debug message
06:31 mupuf: and potentially add a warning if the configuration is busted
06:32 karolherbst: yeah
06:32 karolherbst: but I think we should really set a sane crit value when the set value is too high
06:32 karolherbst: but I didn't see that in any fermi or kepler vbios
06:32 karolherbst: so this might be a tesla only problem
06:33 karolherbst: will check the traces there
06:37 mupuf: ack!
06:37 karolherbst: first trace: crit set to 0x82 :D
06:38 karolherbst: thing is, blob doesn't seem to mind
06:38 karolherbst: or can g84 cards sustain that?
06:39 karolherbst: mhh maybe endure is the better word here
06:40 mupuf: the tesla are insane
06:41 mupuf: the crit value could go up to 132°C
06:41 mupuf: so, it seems about right
06:42 karolherbst: k
06:42 karolherbst: threshold_2 is 114°C, so this looks good then
06:43 karolherbst: mupuf: from one of your gpu: https://gist.github.com/karolherbst/3c2f3bb5077b2e617c31
06:43 karolherbst: what did you do? :D
06:43 karolherbst: g84
06:44 mupuf: hmm, what the heck
06:51 karolherbst: mupuf: but it seems to be common that the blob writes 0xff into this threshold on tesla :/
06:52 mupuf: well, trust what the blob does I would say...
06:52 karolherbst: mupuf: with nva0 it is getting sane
06:52 mupuf: tesla is a mindboggling mess for PM
06:53 karolherbst: well then nva+ it is :D
06:54 karolherbst: ohh well, there is a weird nva3 :/
06:54 karolherbst: mupuf: do you mind if I implement it for nvc0 and newer=
06:54 karolherbst: ?
06:55 karolherbst: well [nvc0,nvf0) actually
06:55 karolherbst: no idea how that stuff works on nvf0 and newer
06:58 mupuf: yes, it is OK not to support every chipset from day 0
06:58 RSpliet: karolherbst: when I get back from vacation I could check a GeForce 2 for you
07:00 karolherbst: :D
07:01 karolherbst: mupuf: yeah well, I was thinking to support tesla would be a good idea because of reclocking
07:01 mupuf: sure but the bios should already be doing the right thing, right?
07:01 karolherbst: right
07:01 mupuf: RSpliet: ah ah
07:01 mupuf: the FSRM got introduced in g80 though
07:06 karolherbst: mupuf: so then 111/113/115 for the mid/high/crit threshold sounds sane enough? It is still a bit higher, but better than having even higher values :D
07:06 karolherbst: 115 is the highest value I saw so far
07:06 karolherbst: on fermi/kepler
07:08 mupuf: 115 is fine on tesla
07:08 mupuf:heated up some tesla until 135°C
07:08 mupuf: well, the g84 you saw, actually
07:08 karolherbst: yeah but I ask about fermi/kepler :D
07:09 karolherbst: most cards have crit set to either 6e or 6d though
07:09 RSpliet: mupuf: oh... so I don't have to check my fathers Riva 128 while i'm in NL?
07:09 karolherbst: is the nvc0 somehow special that it is okay with 0x73?
07:09 mupuf: RSpliet: hehe
07:10 mupuf: the nvc0 was heating up like crazy!
07:10 karolherbst: k
07:10 karolherbst: then 0x6e is the max we allow
07:10 karolherbst: 110°C
07:10 karolherbst: still hot
07:10 karolherbst: we don't want to get 110 on any card, right?
07:10 mupuf: let the bios decide, seriously
07:10 karolherbst: yeah, I just mean if we get a really high value or the fsrm isn't configured (for whatever reasons)
07:11 mupuf: it was the people who designed the bios' decision to set the values they set, let it be
07:11 karolherbst: okay
07:11 mupuf: just parse it back and print that as a debug message
07:11 karolherbst: I was just thinking, because this is an area which could actually break cards
07:11 karolherbst: okay
07:11 karolherbst: so printing without fixing
07:11 mupuf: but please make sure we never have to set it ourselves
07:12 mupuf: and for that, we need to check what hte blob does when the card is already posted
07:12 karolherbst: mupuf: nvkm_error for temp == 0 or <90 (because fan/downclock threshold) or >120?
07:12 karolherbst: well for crit that is
07:12 imirkin: my G96 was pretty constantly at like 92 degC
07:13 imirkin: might have had something to do with the fact that i removed the fan
07:13 karolherbst: :D
07:13 imirkin: but it was fine... up to 110
07:13 imirkin: at 120 it would self-destruct
07:13 karolherbst: I doubt that
07:13 imirkin: well, the box rebooted
07:13 imirkin: ;)
07:14 karolherbst: I checked the vbios, and most of the tesla cards have a crit of 125°C
07:14 imirkin: i dunno what happened to cause that, but i suspect running at 120+ might have had something to do with it
07:14 karolherbst: yeah
07:14 karolherbst: crit means => refuse work basically
07:19 damex: hi, i have nvs4200 card along with intel on my t420 laptop. i am trying to use nouveau + intel since i have hardware switch and graphics switchable using vgaswitcheroo. nouveau works as offload just fine but when i try to use it as a secondary display (with or without offload) - my secondary display becomes like that https://www.youtube.com/watch?v=-GWaEo-pEGs can i do something to make it work? i need from nouveau only basic graphics output
07:20 damex: i am on 4.3.3 kernel and 1.0.11 nouveau xorg driver. it is gentoo testing.
07:23 imirkin: damex: wrong channel... nouveau just displays what it's told to. the intel card is sending it crap data =/
07:23 imirkin: damex: you're not the first to hit this, but i haven't the faintest clue what separates the people who see it from the ones who don't
07:23 imirkin: damex: try #intel-gfx
07:24 damex: hm...
07:27 damex: imirkin, are you sure? is there was any workaround?
07:29 imirkin: pretty sure.
07:30 damex: imirkin, i tried to search around for a while.... no success yet ;<
07:52 karolherbst: mupuf: https://github.com/karolherbst/nouveau/commit/95ceffc262a6e5e291867c04f48a56a0ceb0fab8
07:52 karolherbst: okay and now I have to tackle those chipset checks :)
07:57 karolherbst: mupuf: fsrm_validate func poiner inside nvkm_therm_func and then check if the functions is set and call it?
08:00 mupuf: karolherbst: the net is acting up again
08:00 mupuf: it is still loading
08:02 karolherbst: then add .patch :D
08:02 mupuf: thr_crit == 0?
08:02 mupuf: why test this since it is obviously covered by the <= ()
08:02 mupuf: 90
08:03 karolherbst: :D
08:03 karolherbst: right
08:04 mupuf: the checks on the chipsets will not fly either
08:05 karolherbst: I know, but I already said this
08:05 karolherbst: " <karolherbst> mupuf: fsrm_validate func poiner inside nvkm_therm_func and then check if the functions is set and call it?"
08:05 karolherbst: insted of those checks
08:06 mupuf: sounds a much better idea, yes :)
08:07 mupuf: sounds like*
08:07 mupuf: sorry, I did not get your sentence at first
08:07 karolherbst: no worries
08:07 karolherbst: connection issues? :D
08:08 imirkin: get = understand
08:08 karolherbst: ohhh
08:11 imirkin: now you get it? :)
08:28 karolherbst: mupuf: https://github.com/karolherbst/nouveau/commit/5a10b278044bd5428f56e6d3ca7a8b7775769b6d.patch
08:37 karolherbst: mupuf: if you don't have anything to complain about I will send it to the ML
08:49 Dezponia: Tom^: Do I still hold the record for Heaven benchmark score? :P
08:49 Tom^: yes
08:50 Dezponia: Yay!
08:50 Dezponia: Totally worth it!
08:51 karolherbst: would be nice to see how a GeForce GTX Titan Z would do
08:51 Dezponia: karolherbst: I was actually debating getting one of those but the lack of SLI support and the fact that each singel core is still quite a lot slower than 1 TITAN Black core made me skip it
08:51 karolherbst: :D
08:52 karolherbst: this thing is still like 66% faster than the 780 Ti
08:52 karolherbst: also nouveau can't do any SLI so far
08:52 Tom^: can even X?
08:53 imirkin: why would X care?
08:53 karolherbst: wait the titan Z can do SLI
08:53 imirkin: it's a question of automatically splitting up a gpu load across multiple GPUs
08:53 imirkin: or you mean titan x vs titan z? that makes more sense.
08:54 karolherbst: I think he does
08:54 karolherbst: Dezponia: the titan core z seems to be able to do SLI :p
08:54 karolherbst: quad sli by the way
08:54 Dezponia: karolherbst: Yeah but the TITTAN Z is JUST regular SLI, its nothing magic. It'll still require driver and application SLI support to actuall work like "1 card"
08:54 karolherbst: sure?
08:55 Dezponia: karolherbst: yes, its just like regular SLI
08:55 karolherbst: so the one card is a sli card by itself?
08:55 Dezponia: Yepp
08:55 Dezponia: Thats the problem
08:55 karolherbst: then 4 Z would be 8x sli?
08:55 Dezponia: No, its still limited to quad SLI, aka 2 TITAN Z cards
08:56 karolherbst: aha
08:56 Dezponia: Its the same as buying 2 TITAN cards, except hotter and lower clocks. Only time its EVER worth it is if you're limited by space
08:56 karolherbst: and why are there 4x Z steups?
08:56 Dezponia: karolherbst: Which is why "dual gpu cards" are such a poor value proposition even on Windows, you're litteraly paying for two normal GPU's strapped together for a premium price just to take 1 PCI slot less or fit in a smaller case
08:57 Dezponia: karolherbst: There are no 4x Z setups
08:57 Dezponia: karolherbst: And if there are then they're running 2x quad-SLI, not as "one unit" or whatever you would call it
08:58 karolherbst: well there are 4x Z setups
08:58 Dezponia: karolherbst: Got any source on that? Because Nvidias drivers does not support that
08:59 karolherbst: currently I try to find a good source, there are plenty, but most of them look a bit strange though
08:59 Dezponia: karolherbst: Sure you're not just confusing it with people claiming "quad sli" with 2x Z cards?
09:00 karolherbst: Dezponia: one "strange" source: http://wccftech.com/monstrous-custom-rig-4-gtx-titan-gpus-powered-super-bowl-time-show/
09:00 karolherbst: "Thats a resounding total of 8 GK110 cores and 48GB of GDDR5 memory"
09:00 karolherbst: so yeah, I am sure
09:01 imirkin: doesn't say they're running in SLI... just that there are 4 of them
09:01 Dezponia: karolherbst: Dont think thats really using SLI, just GPU Compute with CUDA accross them
09:02 karolherbst: could be, right
09:05 Tanure: Hi, I have a destop with arch linux, Nvidia gtx 970, using nvidia drivers, weston built from sources. I compiled everything, and go to a second tty and start weston-launch, but I got :
09:05 Tanure: [15:00:02.645] failed to create output for /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0/drm/card0
09:05 Tanure: [15:00:02.646] fatal: failed to create compositor backend
09:06 imirkin: weston needs 3d accel
09:06 Tanure: I'm part of video group
09:06 karolherbst: blame nvidia
09:06 imirkin: nouveau doesn't do accel on GM20x
09:06 Tom^: as far as i know the blob doesnt support wayland yet either.
09:06 Tanure: Hi karolherbst
09:07 karolherbst: Tom^: we can still blame nvidia right?
09:07 Tanure: man, I starting to get angry with nvidia
09:08 Tom^: nvidia ╭∩╮
09:08 damex_: nouveau should do proper acceleration on gt520m/nvs4200m ?
09:08 Tanure: =p
09:08 karolherbst: Tanure: well if you want perf with nouveau you should use a kepler
09:08 imirkin: Tanure: that sounds like GF108 or GF119 or something? should be fine.
09:08 imirkin: er, damex_ --^
09:09 karolherbst: well despite that reclocking doens't work on fermi
09:09 imirkin: not sure what that has to do with correctness of rendering
09:09 karolherbst: but who cares about that on a 520m :D
09:09 karolherbst: right
09:09 Tom^: he should get an 780ti :p
09:09 karolherbst: yeah, 780ti is a beast
09:09 Tanure: the best thing is go back to nouveau, at least works
09:10 damex_: hm... okay
09:10 damex_: fermi it is
09:10 karolherbst: Tanure: well you won't get any accell on maxwell gen2 cards, because there are signed firmwares needed by nvidia
09:10 karolherbst: and some patches
09:10 imirkin: damex_: if you're looking for rendering speed, the intel gpu is probably going to be faster... could go either way.
09:11 Tanure: remove my nvidia ? =/
09:11 damex_: imirkin, some outputs dedicated to nvidia gpu. probably nvidia will handle that stuff ;p
09:11 karolherbst: Tanure: well, do you have any other gpus?
09:11 Tanure: no
09:11 Tanure: just this nvidia-crap
09:11 imirkin: damex_: but nouveau will currently (a) support GL 4.1, and (b) allow you to use st/nine, which you can't do with the intel driver
09:11 karolherbst: Tanure: did you just bought it?
09:11 Tanure: yes
09:11 damex_: hm...
09:11 Tanure: =/
09:11 damex_: imirkin, nice ;D
09:11 Tom^: Tanure: if you need 3d you wont be running nouveau on the 970, and the nvidia blob doesnt support wayland yet. so you are a bit locked to X :/
09:12 karolherbst: Tanure: you could give it bakc and take a 780 ti instead
09:12 karolherbst: or maybe a 770 something
09:12 Tanure: karolherbst : I can't.
09:13 karolherbst: then you are out of luck with nouveau sadly
09:13 Tanure: how is AMD with linux and open source?
09:13 Tanure: better than NVida?
09:13 imirkin: Tanure: much better
09:13 Tom^: thats a philosophical question.
09:13 imirkin: Tanure: they have multiple people, with documentation, employed by amd, contributing to mesa and linux kernel
09:14 karolherbst: nouveau has to figure out the documentation by itself
09:14 Tom^: yes but what do you do when mesa doesnt suit your needs? catalyst is a no go. :p
09:14 karolherbst: well
09:14 karolherbst: with amd it isn't a big problem though
09:14 Tanure: so, I can use the proprietary driver and no kernel taint?
09:14 imirkin: more importantly, nouveau mostly relies on volunteers
09:15 karolherbst: Tanure: you will get one for sure
09:15 imirkin: Tanure: huh?
09:15 Tanure: amd + proprietary driver == kernel taint , or it's open ?
09:15 karolherbst: all non free modules taint the kernel I think
09:15 karolherbst: or at least should
09:15 Tanure: if there is a byte close == kernel taint
09:16 Dezponia: Tanure: AMD used propritary firmware
09:16 imirkin: there's an open driver, and a closed driver
09:16 imirkin: they are separate
09:16 Tanure: not a problem a closed drivers
09:16 imirkin: closed driver is called catalyst
09:16 Tanure: ops
09:16 Tanure: closed firmware.....
09:17 Dezponia: imirkin: The driver is free software but the firmware is not, even for the free driver
09:17 Tanure: firmware it's not a problem if you at least have the blob
09:17 Dezponia: So its "good" but for freedom is still not a good idea
09:17 imirkin: Dezponia: does the firmware run on my cpu?
09:17 Dezponia: imirkin: Pretty sure it does yet
09:17 Tanure: runs inside amd card
09:17 Dezponia: imirkin: Loaded by the kernel at least
09:17 imirkin: Dezponia: time to educate yourself then
09:17 imirkin: firmware does *not* run on the cpu
09:18 imirkin: it's basically "hey, we were too cheap to add a rom on the card"
09:18 Dezponia: imirkin: I was under the impression it was loaded from the kernel as a blob?
09:18 imirkin: in that it's a sequence of bytes, and the kernel reads it? yes
09:18 imirkin: but it's not x86 instructions or antyhing
09:18 imirkin: it's command processor instructions, to be executed by the gpu's... command processor.
09:19 imirkin: (and some other, largely related, stuff)
09:19 chillfan: if that's the case, isn't it a good thing for nouveau? surely free software blob reverse engineered better than something you can't control on the card?
09:19 Dezponia: imirkin: Ah, well yeah I guess I was unclear then, I ment it comes from the kernel as a blob not that it runs on the main CPU. I dont want any blobs in my kernel :)
09:19 karolherbst: chillfan: nouveau has the firmware on the gpu
09:19 chillfan: when the blobs aren't blobs i mean
09:19 karolherbst: its basically the same
09:19 imirkin: Dezponia: why do you care where the blob comes from? hard disk, or on-card rom?
09:20 karolherbst: (nouveau also reads the firmwarae blob, huhu)
09:20 imirkin: chillfan: unfortunately that stuff is a lot harder to reverse-engineer
09:20 Dezponia: imirkin: Because I want to be able to throw a FSF approved distro on a system and not have to track down blobs (since they're stripped from the linux-libre kernel)
09:20 Tanure: About firmware, please read : https://wiki.ubuntu.com/Kernel/Firmware
09:20 imirkin: chillfan: and also GM20x GPUs only load signed firmware anyways, so anything we produce won't be loadable on there anyways
09:21 chillfan: ah alright
09:21 imirkin: Dezponia: ok, but that's just a dumb policy. you're giving it more power by subscribing to it...
09:21 chillfan: is there may be any fix for this build problem for karolherbst's branch? http://www.pasteall.org/62848
09:22 chillfan: (using stable reclocking rekpler)
09:22 karolherbst: chillfan: use a 4.3 kernel
09:23 chillfan: you mean I shouldn't use a later one?
09:23 karolherbst: chillfan: well by default that branch will cap your clock to the base clock
09:23 karolherbst: chillfan: no idea, I test my stuff with 4.3
09:23 chillfan: oh okay, i can try with 4.3 :)
09:24 Tanure: there is any news about Nvidia release the signed firmware?
09:24 chillfan: if users collectively badger nvidia, might this help?
09:25 chillfan: just wondering if i should mail them and complain when zyx feature is not in nouveau :p
09:26 Tom^: its funny how unigine heaven isnt affected a single score or fps by compiling mesa etc with -O3 running with bfs, all that makes any change what so ever is my core clock :P
09:26 Tom^: ricing just isnt working here ;_;
09:27 Dezponia: Tom^: What about CPU clock? Perhaps that could squeeze out a few more FPS if you have a overclockable CPU?
09:27 Tom^: i got the non K 3770 so cant do much about that. :P
09:28 chillfan: using voluntary preempt has helped me with the nvidia drivers
09:28 Dezponia: Tom^: Ah... hmmm... enabled XMP on your RAM at least? :P
09:28 karolherbst: :D
09:28 Tom^: Dezponia: yes
09:28 karolherbst: what are you talking about
09:28 karolherbst: CPU will only help with shader compilations
09:28 chillfan: we're talking marginal gains in frames per second, but differences in frame latency are bigger
09:28 karolherbst: and maybe reduce gl call overhead
09:28 karolherbst: yeah, but this isn't the issue with unigine
09:29 Tom^: karolherbst: i need uh ~4 more score in unigine, not a single optimization kernel or mesa wise has affected it =D
09:29 karolherbst: :D
09:29 karolherbst: Tom^: well you could OC your core
09:29 karolherbst: it is a single nvapoke then you got more core clock
09:29 Dezponia: imirkin: Also just to be clear, what firmware does nouveau use, stuff written by nouveau, stuff loaded from the card written by Nvidia :) I'm suspecting the latter but just to get all my facts straight if you have the time :)
09:29 karolherbst: Dezponia: no, firmware located on the nvidia gpu
09:30 imirkin: Dezponia: nouveau makes use of, among other things, the vbios, which comes from the gpu or acpi
09:30 Dezponia: karolherbst: So firmware written by Nvidia on the GPU? Gotcha :)
09:30 karolherbst: you can get it with nvagetbios
09:30 karolherbst: and try ot read it with nvbios
09:30 karolherbst: but there are parts not understood
09:30 Tanure: Dezponia: for non signed drivers nouveau has a way to extract the firmware from closed drivers from nvidia
09:30 imirkin: Tanure: that's a whole separate matter
09:30 imirkin: and not required for regular operations
09:30 Dezponia: Tanure: What imirkin said
09:31 Tanure: why?
09:31 Dezponia: imirkin: Would that even be allowed for regular operation?
09:31 chillfan: i thought nvidia were planning to move that stuff to vdpau only
09:31 imirkin: Dezponia: sure, you can tell nouveau to load pgraph firmware off disk instead of the in-kernel ones written by the nouveau team
09:31 Dezponia: imirkin: Well I guess what I mean, nouveau would not be allowed to extact it and then bundle it and ship it with the driver :)
09:32 chillfan: at least their installer keeps promising that I'll have to do that in the future
09:32 imirkin: there's also video decoding firmware... that's separate... no open version of that exists at present.
09:33 imirkin: Dezponia: no. but nothing preventing one from shipping the extraction script along with the redistributable complete driver...
09:33 imirkin: but either way... why would one want that
09:33 Dezponia: imirkin: Efficent! :P
09:33 imirkin: except for the vdec firmware
09:35 Dezponia: imirkin: So whats the fuzz about Maxwells signed firmware. I guess thats preventing you from loading it from the card and making it work with nouveau?
09:36 imirkin: Dezponia: signed firmware means that the firmware we load has to be... signed
09:36 imirkin: Dezponia: i.e. we can't load our regular firmware
09:36 imirkin: Dezponia: also there are additional technical restrictions which make loading any firmware a very tricky proposition
09:36 Dezponia: I'm so confused right now :P
09:36 imirkin: Dezponia: and lastly, they've changed things around in their driver which makes their firmware not actually show up in traces
09:37 Dezponia: Oh, how handy
09:37 Dezponia: And totally necessary
09:37 imirkin: i doubt it was done on purpose
09:37 imirkin: the loading process is entirely different for these gpu's...
09:38 Dezponia: imirkin: Anyway it sounds like you're waiting for Nvidia to hand you a redistributable license to use a signed blob or something for maxwell and newer?
09:38 imirkin: along with the code to load it in, yes
09:38 imirkin: not all maxwell, just GM20x
09:38 imirkin: GM10x has no such restrictions
09:40 imirkin: the nvidia employee who's working on nouveau + tegra had to deal with this to get nouveau loading on the Tegra X1, and has gone the extra step of also making things work on desktop gpu's. i believe he's waiting on various internal approvals now.
09:41 chillfan: karolherbst: thanks, it works with 4.3
09:42 chillfan: or well it compiled with 4.3, is it on the max clocks by default?
09:43 Dezponia: imirkin: Will the GM20x cards require a blob in the kernel or remain blobless like kepler etc?
09:44 imirkin: Dezponia: GM20x can only load signed firmware. and it's unlikely that nvidia will ever hand us a key or sign things for us
09:44 tanure: imirkin : what's your motivation to work with nouveau , dealing with nvidia ?
09:44 imirkin: tanure: well, up until recently, dealing with nvidia wasn't part of it
09:45 imirkin: tanure: it's fun, i get to work on puzzles, learn GL, write compilers, etc
09:45 chillfan: karolherbst: doing echo 0f > to the pstate interface gives this in dmesg: http://www.pasteall.org/62849
09:45 tanure: well, I have a nvidia and I would like to just get thing working and do my code, but I'm thinking of sell my nvidia and buy a AMD
09:45 chillfan: not sure if that's expected, so just mentioning :)
09:46 imirkin: tanure: definitely AMD is the way to go if you want to avoid proprietary drivers
09:46 imirkin: (or intel, but they don't make powerful GPUs)
09:48 Dezponia: I'm waiting for AMDGPU to get a bit more established, perhaps in the 400 series of cards or whatever is on the horizon (also sounds like the 400 series will be a massive upgrade in terms of power efficency etc) and then I'll likely try them as well :)
09:50 chillfan: oops, shouldn't have tried setting 0d heh
09:53 chillfan: ok 0f messes stuff up for me too, glitchy flickering stuff happens
09:55 Dezponia: Humm, does the GM20x series support 3D acceleration yet?
09:56 imirkin: Dezponia: not with nouveau
09:57 imirkin: Dezponia: the gpu's themselves, however, are quite capable of doing 3d accel :)
09:57 Dezponia: imirkin: Figured as much, this guy must be running software 3D accleration or something: https://h-node.org/videocards/view/en/1663/NVIDIA-Corporation-GM204--GeForce-GTX-970---rev-a1-/1/1/NVIDIA-Corporation/undef/undef/undef/undef/undef
09:57 Tom^: kernel 3.13?
09:57 Tom^: O_o
09:57 chillfan: ok erm, if i have deblobbed kernel I guess I can't use vdpau with the external firmware?
09:58 imirkin: Dezponia: note "VESA" :)
09:58 imirkin: that kernel is from WAY before the maxwell series was announced
09:59 Dezponia: imirkin: Yepp, just weird that he would claim that then. Even if its running VESA why claim it for the hardware?
09:59 Dezponia: Oh well whatever, people are weird
10:03 chillfan: should "0f" be expected to work on the stable reclocking branch yet?
10:03 Tom^: depends, on my card its using to low volt for that clock.
10:03 chillfan: i think the same happens here
10:04 chillfan: [ 31.934363] nouveau 0000:01:00.0: clk: failed to raise voltage: -22
10:04 chillfan: [ 31.934508] nouveau 0000:01:00.0: clk: error setting pstate 3: -22
10:05 chillfan: anyway will try whenever something changes in stable reclocking kepler :)
10:06 chillfan: Tom^: does your card use reference clocks?
10:06 Tom^: nope
10:06 chillfan: just curious if it might be why voltage is different, mine is one of the superclocked version
10:07 imirkin: chillfan: does karolherbst have a copy of your vbios?
10:07 chillfan: imirkin: i sent it before, i think someone said they were uploading it somewhere
10:08 chillfan: anyway the mail i sent over ought to start with my nick here
10:09 chillfan: i hope all is ok with the vbios as it seemed a little small (64K)
10:10 imirkin: chillfan: 64K is the right size
10:10 chillfan: alright :)
10:14 chillfan: trying to find the voltages for it, i think they aren't standard in order to allow for the higher clock speeds, but not beyond the limit set by nvidia
10:14 karolherbst: chillfan: k, back
10:14 karolherbst: chillfan: what gpu was it again?
10:14 chillfan: 780 ti
10:14 karolherbst: chillfan: full dmesg pls by the way
10:15 chillfan: ok will post it, 1 sec :)
10:16 chillfan: http://www.pasteall.org/62850
10:16 karolherbst: chillfan: how did you send the vbios? because I don't see any link in the log
10:16 chillfan: i mailed it to mmio dumps
10:16 karolherbst: ohh I don't have any access there
10:16 karolherbst: also I doubt that you use my version of nouveau
10:16 chillfan: i think then imirkin linked you to it and someone else chimed in to say they'd upload it?
10:17 Tom^: imirkin: this doesnt sound good https://gist.github.com/anonymous/ee888969c0e2f255bd08
10:17 karolherbst: because you should see something like that: nouveau 0000:01:00.0: clk: base: 705 MHz, boost: 797 MHz
10:17 chillfan: hm alright, how can I be sure? I did build from your tree
10:17 karolherbst: chillfan: did you rebuild initramfs after installing?
10:18 chillfan: I don't have initramfs
10:18 karolherbst: did you overwrote the old nouveau module?
10:18 karolherbst: *overwrite :D
10:18 chillfan: well i did make install
10:18 karolherbst: mhhh
10:18 karolherbst: that doesn't seem to work most of the cases
10:18 imirkin: Tom^: that's pretty common for wine
10:18 chillfan: i guess that overwrites it?
10:18 chillfan: oh ok
10:18 karolherbst: chillfan: /lib/modules/whatver
10:18 imirkin: Tom^: they do that crap for basically no good reason
10:18 karolherbst: there you have a new folder called extras
10:18 karolherbst: or something
10:18 Tom^: imirkin: heh ok
10:18 imirkin: Tom^: i mentioned it to them, but fixing it was low priority
10:19 imirkin: no harm done
10:19 chillfan: yeah the nouveau module is in extra
10:19 karolherbst: right
10:19 chillfan: copy it elsewhere? :)
10:19 karolherbst: and the old one is in kernel/drivers/gpu/drm/nouveau/
10:19 karolherbst: remove the old one and move the extras one there
10:19 imirkin: they just go through all the formats, and accidentally hit some illegal combinations of things
10:19 chillfan: ok that's moved
10:20 chillfan: i'll should reboot?
10:20 chillfan: or stop X? i'm not sure with nouveau cause of the console stuff
10:20 karolherbst: reboot
10:21 chillfan: ok brb :)
10:21 karolherbst: mupuf: what do we tackle after fsrm, power consumption or proper volting?
10:23 Tom^: power consumption.
10:23 karolherbst: Tom^: and with power consumption I mean power sensors :p
10:23 Tom^: yes
10:23 karolherbst: and staying within power budget
10:23 Tom^: indeed
10:25 karolherbst: chillfan: dmesg pls :D
10:25 chillfan: ok erm, not sure if it worked.. after moving module it doesn't load, unknown parameter, did a make install (ok will post dmesg again) and it worked then
10:25 karolherbst: chillfan: did you rename the module to ko.xz?
10:25 karolherbst: with the move
10:26 chillfan: no
10:26 karolherbst: mhh anyway, dmesg :D
10:26 chillfan: http://www.pasteall.org/62851
10:27 karolherbst: well it worked: "nouveau 0000:01:00.0: clk: base: 1006 MHz, boost: 1071 MHz"
10:27 karolherbst: but with default options you only get 1006MHz max now
10:27 karolherbst: on 0f
10:27 karolherbst: but it should reclock without issues
10:27 chillfan: ok so if i set 0f should be ok? :)
10:27 karolherbst: that's why it is stable, the branch doesn't say something about optimal reclocking, so it's not as fast as it could be :D
10:27 karolherbst: chillfan: yeah for now
10:27 karolherbst: it could be though
10:28 karolherbst: that the voltage is still too low
10:28 Tom^: i doubt it will, since he is still on to low volt.
10:28 karolherbst: and your gpu messes up
10:28 mupuf: karolherbst: yes, power consumption
10:28 chillfan: okay i will watch sensors in case
10:28 karolherbst: Tom^: well your gpu is the only one that's need a higher voltage actually
10:28 chillfan: it's working no dmesg error about voltage now
10:28 karolherbst: :D
10:28 Tom^: didnt Dezponia need it too?
10:28 chillfan: karolherbst: thank you :)
10:28 mupuf: I can do experiments with my GM206
10:28 karolherbst: chillfan: well benchmark with unigine
10:28 karolherbst: chillfan: and tell me if that's stable
10:28 chillfan: okay i will try it :)
10:28 karolherbst: chillfan: if yes, boot with nouveau.config=NvBoost=2
10:28 karolherbst: and benchmark again
10:29 karolherbst: if this also works, nouveau.config=NvBoost=3
10:29 mupuf: but ... I need to investigate this issue with nvafakebios
10:29 karolherbst: mupuf: right
10:29 karolherbst: and then I can also nvafakebios :)
10:29 chillfan: I guess it'll let me know if it's not stable?
10:29 karolherbst: (which actually removes one of me excuse not doing stuff)
10:30 chillfan: or it'll just crash i guess heh
10:30 karolherbst: chillfan: you will notice
10:30 karolherbst: Tom^: well my gpu can even OC by 135MHz without even touching the voltage
10:30 chillfan: alright, should i use the one from phoronix test suite?
10:30 karolherbst: Tom^: so my gpu can run far below the "min voltage"
10:31 karolherbst: chillfan: you can just torrent unigine benchmark, they have a link somewhere
10:31 karolherbst: heaven
10:31 karolherbst: chillfan: https://unigine.com/products/benchmarks/heaven/
10:31 karolherbst: linux fast torrent
10:32 chillfan: alright :)
10:32 Dezponia: Tom^: karolherbst: I dont think I had to mess with voltage to much, just needed karolherbst's nouveau for 0f to work with my card if I recall?
10:33 karolherbst: Tom^: see, your gpu quality is just not that good :p
10:33 karolherbst: or just too much OCed
10:33 chillfan: this is the pstate output: http://www.pasteall.org/62852 (whilst i wait to set up unigine)
10:33 Tom^: (╯°□°)╯︵ ┻━┻
10:33 Dezponia: Tom^: Clearly you should buy a titan black
10:35 karolherbst: well
10:35 chillfan: ok i ought to run extreme or just basic? any more options i should enable?
10:36 karolherbst: the 780 ti is pretty much the fastest you can get
10:36 karolherbst: chillfan: all maxed out, except 4x msaa and tess to normal
10:37 chillfan: at my best res?
10:37 karolherbst: chillfan: 8x msaa has a weird issue, has to be investigated
10:37 karolherbst: chillfan: if you wish
10:37 karolherbst: chillfan: what is your best res?
10:37 chillfan: 1920x1080
10:37 karolherbst: piece of cake
10:37 chillfan: full screen matters?
10:37 karolherbst: 4k would be a different issue though :D
10:37 karolherbst: mhh
10:37 karolherbst: I would do fullscreen
10:37 karolherbst: but I don't like it
10:37 karolherbst: because unigine doesn't let you tab out
10:38 karolherbst: worst kind of application
10:38 chillfan: ok will leave that one off :)
10:38 karolherbst: you will be suprised
10:39 karolherbst: and den you can add +10% and that*s what nouveau can currently do with your gpu
10:39 Dezponia: A third challanger appears for the Heaven top score?
10:39 karolherbst: well
10:39 karolherbst: that went well
10:40 Tom^: to low volt.
10:40 Tom^: xD
10:40 chillfan: ok that locked up
10:40 chillfan: lol
10:40 karolherbst: yeah
10:40 karolherbst: same issue as Tom^ has
10:40 Tom^: to low volt bro. :P
10:40 chillfan: hehe,
10:40 karolherbst: the min voltage stated in the vbios (maybe nouveau missunderstands it though) is just too low
10:40 chillfan: Tom^: what's your specific card btw?
10:41 Tom^: 780ti
10:41 chillfan: no i mean brand
10:41 Tom^: *shrug* ages ago, MSI i think.
10:41 Dezponia: So wait... am I the only one who could run Heaven without messing with voltages?
10:41 karolherbst: chillfan: well we could hack nouveau to use the highest voltage, but this may let your card get hot a bit
10:41 karolherbst: still okay with base clocks
10:41 karolherbst: but well
10:42 chillfan: hm
10:42 karolherbst: Tom^ has a patch for this :p
10:42 chillfan: i think nvidia set a safe limit on this card, i remember some people complained about it lol
10:42 karolherbst: yeah
10:42 karolherbst: 1.215V or something
10:42 chillfan: sounds familiar
10:42 karolherbst: but
10:42 karolherbst: your card will run at a lower volt
10:43 karolherbst: ohh right
10:43 karolherbst: your vbios :D
10:43 chillfan: the cooler my card fits is intended for the job of it
10:43 Tom^: drm/nouveau/nvkm/subdev/volt/base.c change line 82 to info.max += ret; instead of info.min, and change line 84 to return info.max; instead of info.min
10:43 chillfan: so i think i'd be ok
10:43 chillfan: karolherbst: you should have it there somewhere :)
10:43 karolherbst: I think not :/
10:43 chillfan: or i can send it again
10:43 chillfan: Tom^: ok noted
10:44 karolherbst: chillfan: well please upload it :D
10:44 chillfan: where am i looking for it again?
10:44 karolherbst: I don't have access to all that fancy email stuff
10:44 karolherbst: chillfan: envytools installed?
10:44 chillfan: alright i'll find a file upload
10:45 chillfan: nope
10:45 chillfan: debugfs?
10:45 karolherbst: then /sys/kernel/debug/dri/0/vbios.rom
10:45 chillfan: hm mount -t tmpfs?
10:45 imirkin: karolherbst: do you not have access to mmio.dumps?
10:45 imirkin: chillfan: debugfs
10:46 chillfan: mount -t debugfs /sys/kernel/debug .. says it can't find it in fstab
10:46 imirkin: mount -t debugfs none /sys/kernel/debug
10:46 chillfan: ty :)
10:47 chillfan: using cat vbios.rom > /path/vbios.rom?
10:50 imirkin: chillfan: yes
10:50 chillfan: surprisingly the size is now 200K
10:52 karolherbst: imirkin: don't think so, nobody gave me access so far :D
10:52 chillfan: https://www.speedyshare.com/yJtdZ/evga-gtx780ti-superclocked.rom
10:52 karolherbst: chillfan: seems fine
10:52 chillfan: alright :)
10:53 karolherbst: chillfan: now install envytools :D
10:53 karolherbst: I need something
10:54 chillfan: ok one sec :)
10:55 chillfan: ok they're installed
10:56 karolherbst: as root: nvapeek 101000
10:56 chillfan: 00101000: 8040689a
10:57 karolherbst: mhhh 0.925V on 1000MHz
10:57 karolherbst: that sounds a bit low :D
10:57 chillfan: oh a sec, should i be on 0f?
10:57 karolherbst: nonono
10:57 karolherbst: that is unrelated
10:57 chillfan: okay
10:57 karolherbst: I got the volt from the vbios
10:58 chillfan: i took the bios whilst on standard, no change of reclock value
10:58 chillfan: does that change the result, or doesn't matter?
10:58 Tom^: doesnt matter
10:59 Tom^: your vbios is a static thing until you actually reflash it.
10:59 karolherbst: oh crap
10:59 karolherbst: there are two power budget which seems valid
10:59 karolherbst: 4 actually
10:59 karolherbst: mupuf: you will have a lot of fun with that
10:59 Tom^: =D
11:00 chillfan: will it help if i see if i can find the max voltage for the card?
11:00 chillfan: the one set by the manufacturer
11:02 karolherbst: chillfan: actually 1.2125V is the max voltage :D
11:02 karolherbst: chillfan: its all in the vbios
11:02 chillfan: alright, fair enough :)
11:03 chillfan: http://www.evga.com/articles/00795/#2884 if needed that shows the clocks it uses
11:04 chillfan: could i might be able to extract the exact info from nvidia driver?
11:04 karolherbst: chillfan: thanks for confirming :D
11:04 chillfan: I guess the 'power mizer' shows that?
11:04 karolherbst: "nouveau 0000:01:00.0: clk: base: 1006 MHz, boost: 1071 MHz"
11:04 karolherbst: in your dmesg
11:04 chillfan: aha lol
11:06 chillfan: i could ask to see if i can get the stock voltage off some enthusiasts too if that doesn't work
11:06 karolherbst: well I think mupuf knows already what is the "right" voltage
11:06 chillfan: i'm sure some geeks on a forum know it by heart :p
11:06 chillfan: oh okay
11:07 karolherbst: only needs some verification or something
11:07 karolherbst: but there are more important tasks for now
11:07 chillfan: okay cool, i will check with nvidia driver see if it tells me
11:07 chillfan: sure it has something in there for that
11:09 chillfan: so will let you know if i find that out to confirm it
11:09 chillfan: brb :)
11:10 karolherbst: chillfan: well you can't verify it, you can only tell us if that works stable
11:10 karolherbst: or at least tell us if it's the same as the blob
11:11 chillfan: oh ok, will wait for whenever you need more info then
11:14 chillfan: could try to send some mmio dumps from nvidia / piglit tests from nouveau in the meantime if that's any help at all
11:22 chillfan: brb :)
11:22 karolherbst: chillfan: not randomly
11:24 karolherbst: ohh saints row IV released for Linux :O
11:26 karolherbst: imirkin: issues incoming :p
11:27 imirkin: so.... i just noticed that GF108 is a lot more broken with talos now :(
11:27 karolherbst: mupuf: so are you fine with my patch now by the way?
11:27 imirkin: everything's red, including the intro even
11:28 imirkin: it works *way* better on nv50 btw
11:28 karolherbst: mhh
11:28 karolherbst: imirkin: latest mesa=
11:28 karolherbst: ?
11:28 karolherbst: really?
11:28 karolherbst: no flickering and no issues?
11:30 imirkin: karolherbst: on nv50 it's fine... maybe always has been
11:30 imirkin: karolherbst: on nvc0 it's still just as bad, but now it got worse
11:31 karolherbst: that's bad
11:33 imirkin: i don't have time to look into it now :(
11:34 imirkin: looks like it's something i did since 11.1 though
11:35 imirkin: if you or someone could figure out which change, that'd be great
11:35 imirkin: my suspicion is that it's something that's not actually a bug, but will better point out what we're getting wrong in those shaders
11:35 imirkin: i.e. i just caused some code to get reorganized a bit which made the bug manifest itself differently
11:38 karolherbst: k, maybe I will bistect it then
11:39 imirkin: well first make sure you see the same issue
11:39 chillfan: looks like i can't install nvidia anyway, but thanks for helping me build the module and getting the info to support my card :)
11:48 imirkin: karolherbst: actually i think it might be a local change that's buggered
11:49 chillfan: I have as well an NV118 (940m) that I can try to get info from later if you like
11:49 chillfan: probably just after christmas
11:51 imirkin: karolherbst: yep. it was a local change.
11:51 karolherbst: k
11:54 chillfan: alright later on :)
12:03 damex_: NVIDIA Quadro FX 880M support is bad too ? :<
12:05 imirkin: damex_: can you elaborate what you might mean by 'bad'?
12:06 damex_: imirkin, would it support power management? would it work completely?
12:06 imirkin: what gpu is FX 880M?
12:07 imirkin: also what do you mean by 'power management'
12:07 imirkin: as for working completely, it's unlikely that any piece of hardware ever made ever works "completely" with any set of drivers
12:07 damex_: > It is based on the GeForce GT 330M graphics card with slower clock rates and therefore is a DirectX 10.1 chip that is produced in 40nm.
12:07 imirkin: i know these aren't the answers you're looking for, but you probably have something very specific in mind, and i have no idea what it is
12:08 damex_: imirkin, hm...
12:08 imirkin: pci.ids suggests that it's a GT216 aka NVA5
12:08 imirkin: which in turn means that it should have working reclocking
12:09 imirkin: however all that stuff's pretty experimental, so i certainly wouldn't guarantee it
12:09 damex_: okay ;<
12:09 imirkin: and the nv50 mesa driver supports most of the functionality offered by the gpu
12:09 imirkin: however it'd be impossible to say that the support were "complete"
12:10 imirkin: and there are def bugs i'm aware of
12:10 imirkin: but it should be mostly good
12:11 damex_: imirkin, okay. thanks.
12:11 damex_: i guess it will be okay to pick thinkpad w510 then
12:13 imirkin: well, again... note the asterisk -- no guarantees
12:13 imirkin: that reclocking will work
12:13 imirkin: it does work ok for a bunch of nva* owners, but some, esp iirc nva5 owners, have reported flickering
12:13 imirkin: at higher speeds
12:13 MichaelLong: damex_, what nvidia is in this w510?
12:14 MichaelLong: ah ok see GT330
12:15 damex_: MichaelLong, yep ;p
12:15 damex_: quadro equivalent
12:16 MichaelLong: my NVS 5400 is doing great an the one before NVS 4xxx was also very good with nouveau
12:17 damex_: MichaelLong, right now i have nvs4200m and i want laptop with more memory ;p
12:18 MichaelLong: more means 32GB?
12:18 damex_: yep
12:18 imirkin: it's having trouble remembering?
12:18 MichaelLong: damex_, nice :)
12:18 damex_: imirkin, more memory for virtual machines ;p
12:19 MichaelLong: mein thinkpad is limited to 16GB :(
12:19 karolherbst: damex_: why not just go with a laptop with a modern intel cpu like hsw or broadwell and just put in it a lot of RAM?
12:19 MichaelLong: thats my plan
12:19 damex_: karolherbst, price; sturdiness; terrible keyboard/ergonomics;
12:20 damex_: karolherbst, that is why i prefer older thinkpads
12:20 imirkin: damex_: T42's are getting difficult to find :p
12:20 damex_: imirkin, ;)
12:20 karolherbst: well if you want to run many vms, did you alreads enabled ksm or uksm?
12:20 damex_: imirkin, people sell it here for 25$ :)
12:21 damex_: karolherbst, ksm. does not really help
12:21 karolherbst: then check uksm
12:21 Tom^: i got a t60p right next to me with 3gb ram and a ssd in it.
12:21 Tom^: <3 =D
12:21 karolherbst: :D
12:21 karolherbst: well
12:21 karolherbst: I don't know your budget
12:21 karolherbst: but with 600$ you can do a lot already
12:21 damex_: karolherbst, i don't really have a budget for that. i just do not want to spend extra for something i don't like.
12:21 karolherbst: right
12:22 damex_: karolherbst, i don't like modern machines ;( last machine i liked is *20 series of thinkpads.
12:22 karolherbst: and why not?
12:22 damex_: why not what?
12:22 karolherbst: why you don't like modern machines
12:24 damex_: bad ergonomics, terrible keyboards, much worse serviceability, relatively pricy compared to older thinkpads ;)
12:24 karolherbst: well I doubt that all "new" machines are that bad ;)
12:24 karolherbst: ohh you want a thinkpad
12:24 damex_: it is like i can pay 1.5k$ for x250 (company paid for that) or i can get 200$ x220. wich is much better.
12:24 karolherbst: well if you want to run vms, you should really plan to have 16GB ram
12:25 damex_: karolherbst, my t420p already have 16gb ram :)
12:25 karolherbst: well zram would be also a solution
12:25 karolherbst: ohh i see
12:25 damex_: and 4 core cpu with 8 threads ;)
12:25 damex_: i just lack memory for vms
12:25 damex_: that is why i am aiming for w510
12:25 MichaelLong: damex_, that good old T420 with 32 or more ram, usb 3.0 and a high-res-screen. thats it.
12:26 damex_: MichaelLong, i can install 1920x1080 screen and get usb3 via expresscard :)
12:26 MichaelLong: not the same :)
12:26 damex_: 1920x1080 screen from t440 ;)
12:26 karolherbst: maybe that sounds sutpid, but my laptop of choice for semi professional stuff are asus :D
12:26 damex_: MichaelLong, but i won't get 32gb ram there
12:26 damex_: and modern machine with 32gb ram will cost much MUCH more
12:27 karolherbst: my old asus is unbeatable in every aspect which is important for a laptop
12:27 karolherbst: it was a N56VM though, so it is way out of your budget :D
12:28 karolherbst: it has the best cooling system I saw in a laptop for a while :/
12:29 damex_: karolherbst, what is so special?
12:29 karolherbst: i7 maxed out at 3.4GHz ivy bridge and nearly silent
12:29 karolherbst: can hold boost forever at 3.4GHz
12:29 damex_: karolherbst, my opinion is that asus is the worst company out there.
12:29 karolherbst: well my experience with that asus is really good
12:30 karolherbst: but yeah, I know
12:30 karolherbst: it sounds like the worst idea :D
12:30 damex_: it is after having to deal with their laptops and desktop/server motherboards
12:30 Tom^: ibm laptops > *
12:30 karolherbst: damex_: also mid/high end stuff?
12:30 karolherbst: I can imagine the cheap stuff is shit
12:30 damex_: karolherbst, yep
12:30 karolherbst: maybe I was lucky
12:30 damex_: karolherbst, their bios/uefi is a garbage.
12:30 karolherbst: but I was suprised how quiet that one is
12:30 karolherbst: ohh
12:31 karolherbst: yeah, that is true for most laptops anyway
12:31 damex_: karolherbst, and that is why their hardware sometime is not even functioning
12:31 damex_: even under windows
12:31 karolherbst: yeah I can totaly believe that, maybe mine is just randomly very good :D
12:31 damex_: i had to deal with 'asus support'
12:31 damex_: lets say that it does not exist.
12:31 karolherbst: damex_: but did you see any laptop, 8 cores at 100%, max turbo, i7 be silent?
12:32 damex_: karolherbst, nothing will be silent.
12:32 karolherbst: this one is :D
12:32 damex_: karolherbst, except full passive desktop/
12:32 karolherbst: well silent compared to "normal2 laptops
12:32 karolherbst: it just sounds like it is idling around
12:33 damex_: good for you ;p
12:33 damex_: i will probably hear it clearly ;p
12:33 karolherbst: yeah I think I was just lucky :D
12:34 karolherbst: well now I have a clevo p150sm and this one is just loud :D
12:35 damex_: well i tried w541 thinkpad and returned it;
12:36 damex_: keyboard just bad;
12:36 urjaman: i see
12:37 damex_: never does not mean better ;(
12:37 damex_: s/never/newer/
12:38 karolherbst: I now
12:38 karolherbst: but it also doesn't mean worse;)
12:46 Yoshimo: well some newer stuiff might feel worse because it lacks support
14:30 karolherbst: looking good so far (saints row 4)
14:40 karolherbst: and it runs pretty bad with nouveau :/
14:45 imirkin: file bugs :)
14:45 imirkin: unless you just mean slow
14:46 imirkin: i'm out for a while... probably won't be on irc, but i'll be on email
14:47 karolherbst: ohh wow is that bad
14:47 karolherbst: only 33% blob perf
14:50 karolherbst: maybe they use tesseleation or something