01:27mlankhorst: neither ;-)
01:28mlankhorst: but most code is sucked out of ttm now
01:33mlankhorst: without its custom locking implementation ttm is a lot less scary
04:59nfk: karolherbst, if you found the bug on gentoo, why did you not file a gentoo bug as well?
05:11karolherbst: nfk: mhh totally forgot about this somehow :/
05:14karolherbst: nfk: I hope though, that xproto doesn't get masked
05:15karolherbst: If I look at this: 256 clients isn't that much
05:18nfk: karolherbst, sadly I just filed one :P
05:19nfk: karolherbst, then recommend in that gentoo bug they apply that xorg-server patch
05:21nfk: also what's the point of not masking that xproto version when going above 256 will cause crashes if I understand correctly
05:23nfk: and now that I think about it, having many tabs in chromium is a bit sick
05:23nfk: for one it's not firefox where many tabs are even supported by gui
05:23nfk: and the memory consumption and general lag is stupendous
05:24shakesoda: I don't even want to know what would happen in chrome if I used something like tree style tab
05:25shakesoda: RIP memory
05:56karolherbst: nfk: I already did :D
05:56karolherbst: nfk: having many tabs isn't that bug problem
05:57karolherbst: I usually have more than 10 around
05:58karolherbst: chrome uses like 1,5GB of memory here with 12 tabs and several extensions
05:58karolherbst: this isn't that much
06:00nfk: karolherbst, then how did chromium let you repeat it?
06:02nfk: also i think i need another coffee, because that's still stupid amount, the RES for firefox right now is only 504 MB for me
06:02nfk: and i have had about 30 large resolution image tabs open for the better part of my day
06:03nfk: and that's with the mozilla's hated adblock+ in use
06:11karolherbst: wow, flash plugin 420MB :/ that's bad
06:12karolherbst: okay, with flash disabled and only adblock I am also at 500MB
06:35nfk: for a few times less tabs but okay, mine are not that bad except for the huge images
06:35nfk: i think the largest file today had 5 digit vertical resolution
06:36nfk: and if anyone wonders how is that even possible with firefox, see my leave/quit message
06:51Karlton: surf only uses 49.6mb when I first start it up but it grows everytime I reload a page
06:51Karlton: I think webkitgtk 2.4.9 has memory leak :(
07:02karolherbst: Karlton: or just a pretty optimistic cache
07:12nfk: karolherbst, or it's taking ideas from firefox
07:13nfk: the fucking assholes at mozilla repeatedly told me firefox has no memory leaks despite exactly the same issue
07:13nfk: it turns out (see my leave/quit message) it's indeed not a leak, it's fucking intentional
07:13nfk: obviously no one will tell you that so now i have it as my leave message to tell the masses about the bullshit from mozilla
07:17Karlton: that has been known for a long time: http://harmful.cat-v.org/software/firefox
07:19Karlton: "If your software is so complex you can’t figure out how to fix a memory leak in less than twenty minutes, then you are doing something wrong, if the leak stays around for months, maybe it is best that you give up and start from scratch."
07:22karolherbst: but web browsers aren't simple
07:24nfk: karolherbst, not a leak
07:25nfk: it knows about it, just does not care
07:25Karlton: karolherbst: but they should be
07:25karolherbst: but is there something objective on that harmfull homepage for a change? :D
07:26nfk: oh, sorry, Karlton not karolherbst
07:27nfk: Karlton, firefox by default does not free dirty pages that are left after you close a tab and possibly just reload (not sure about that last one) to reuse them interally but it will also never fragment the memory to avoid complexity/overhead/pahfomance hit so in such case it will just ask for more pages from OS
07:28nfk: and then it repeats, repeats, repeats until its useless heap is like half your whole RAM and even linux is starting to feel it
07:28nfk: yes, it's a feature desu
07:29Karlton: and unfortunately they are trying to make web browsers do everything
07:29nfk: and the memory.free_dirty_pages from the top of my head does exactly what is says on the tin - makes it release those unused pages to OS
07:30nfk: Karlton, again, firefox can be used for days and countless tabs as long as you just enable that switch so it's not all bad
07:30nfk: also i rather have stuff in the cloud than using windows only applications/internet explorer compatible crap
07:30karolherbst: why should linux feel it if half the RAM is used?
07:31Karlton: nfk: maybe webkit has this feature now :(
07:31karolherbst: that would be a big issue inside the kernel if that would be that way
07:31nfk: karolherbst, because you're linking gcc :P
07:31nfk: this is gentoo, obviously
07:31nfk: or are you telling me i can't look for new wallpapers while portage is doing work?
07:31karolherbst: I also compile stuff like libreoffice in ram and hardly notice anything
07:32karolherbst: but thats more because of CPU, not RAM
07:32nfk: Karlton, would be pathetic in that case, i'm prettox "invented" that crap back when XP was a thing
07:32karolherbst: free RAM is wasted RAM
07:32nfk: allegedly it's due to windows having slow allocator
07:33nfk: but i still feel mozilla has feces for brain if their engineers went "yeah, let's turn this on on every platform because windows and mozilla experience"
07:33nfk: *pretty sure firefox
07:33karolherbst: openssl *cough*
07:33nfk: sadly my system is kinda laggy due to recompilation of qt
07:33karolherbst: how much RAM do you have?
07:33nfk: 4 GB
07:34nfk: this system is like 7 years old by now
07:34karolherbst: ever thought about zram?
07:34nfk: if a miracle happens i'll have enough money to get a skylake
07:34karolherbst: you could try to use zram for swap instead of slow old disc swap
07:34nfk: but i also should get a smartphone finally now that my nokia has a loose backplate that's also securing batery and SIM T_T and the car is a bad shape
07:35nfk: rust and now also a skipped timing belt
07:35nfk: at least it's likely due to oil and not engine almost grinding to a halt so crankshaft should be fine and i'll avoid the worst possible case just hopefully
07:36nfk: *oil on timing belt
07:36nfk: karolherbst, i have been using zswap i think for a year or so
07:36nfk: can't say i really feel a difference
07:37nfk: also my swap is on a raid10
07:37nfk: software but still raid
07:37karolherbst: isn't zswap just compressed swap on a disc?
07:37karolherbst: ahh no
07:37karolherbst: its the old zram thing
07:37karolherbst: you need a really big swap with it
07:37karolherbst: and a high swapiness
07:38nfk: no, it tries to compress pages that are about to go to swap and if possible keeps them in ram compressed with lz4 (not default)
07:38karolherbst: otherwise you don't change much
07:38nfk: swappiness is normal, swap is huge though i fail to see how that matters given that it's using ram to store those compressed pages
07:39karolherbst: it needs less ram? :D
07:39nfk: it uses up to 25% of ram by default
07:39nfk: and tha'ts the best difference
07:39karolherbst: who does?
07:39karolherbst: zram doesn't do it
07:40nfk: zram hard allocates that memory iirc while zswap is dynamic afaik
07:40karolherbst: mhh not really
07:40karolherbst: zram only uses as much memory as it currently needs
07:40nfk: also zswap seems to be more compatible with stuff than zswap
07:40karolherbst: I even compile inside a zram fs with portage
07:41karolherbst: if you add the discard mount option it actually frees memory :D
07:41nfk: apart from hibernation being crazy which might or might not be due to zswap everything else seems normal
07:41karolherbst: so you can have a 16GB zram filesystem which uses nearly no memory, if nothing is there
07:41nfk: is discard like trim?
07:42karolherbst: mhh, well it supports it, but pages wich only contains 0 will be marked and not compressed
07:42karolherbst: or better compressed
07:42Karlton: what do you compile that needs more than 2gb of ram?
07:42nfk: a lot
07:42karolherbst: and discard just tells the zram device to overwrite deleted files with 0
07:43nfk: karolherbst, it's easier to list what needs more than 4 GB: pypy (shit), chromium with -g
07:43nfk: there's some more but i can't remember from the top of my head
07:43karolherbst: wine with -g is really painfull :D
07:43nfk: and if you have the guts to try LTO, it's likley 16 GB will not cut it for some programs when debug symbols are in use
07:43karolherbst: ahh I use lto by default for everything here
07:44karolherbst: with gcc 4.9 lto isn't that bad anymore
07:44nfk: i already suspected you had a lot of RAM
07:44nfk: oh, mozilla's rust
07:44karolherbst: yeah makes sense with gentoo
07:44nfk: also 4 GB without debug symbols and ld switches to use less RAM will not cut it
07:45karolherbst: you should try creating a 8GB zram device and add it as swap
07:45karolherbst: and set swapiness to 100
07:45mupuf: karolherbst: lto works great on gcc 5.2 :)
07:45karolherbst: you big problem is file cache
07:45mupuf: Just tested it again today
07:45karolherbst: I usually have a file cache of around 11 or 12 GB
07:45karolherbst: and my discs are idling all the time
07:45mupuf: I get 3.5% more perf on cpu-limited cases
07:46karolherbst: mupuf: with lto?
07:47karolherbst: I thought it would reduce more memory usage
07:47karolherbst: in fact its kind of strange, that it helps with CPU bottlenecked tasks
07:47karolherbst: it makes sense a bit, but still
07:49karolherbst: nfk: how big is your file cache usually?
07:49nfk: right now it's less than 1 GB
07:49karolherbst: thats like nothing
07:49nfk: but usually around half of my RAM
07:49nfk: which is 2 GB
07:49karolherbst: even 2GB is like nothing
07:50nfk: it's half of my RAM which is a lot
07:50karolherbst: do you have a lot of disc IO while doig stuff?
07:50karolherbst: disc cache is 13GB here
07:50karolherbst: and I have 16GB RAM
07:50karolherbst: half of RAM is nothing ;)
07:50nfk: generally lag starts to creep in after total non-cache memory goes above about 2.5 GB
07:51karolherbst: yeah,, probably disc IO
07:51karolherbst: it just doesn't find anything in the cache and has to read a lot from the disc
07:51karolherbst: this slows down everything
07:51nfk: also BFQ is slightly better than CFQ though both are similar and BFQ worst case is worse than CFQ
07:52nfk: karolherbst, what did you expect, it's SATA1 on a 7 yo system
07:52karolherbst: yeah, if BFQ is better then application want to read a lot
07:52mupuf: karolherbst: yes, with lto
07:52karolherbst: usually you want a situation where it doesn't make a difference if you use CFQ or BFQ
07:52nfk: though with raid10 its top speed actually at times exceeds SATA1 theoretical limit
07:52karolherbst: then everything is fine :D
07:52mupuf: and gcc 5.2 added a bit more perf than 5.1
07:52karolherbst: I usually avoid masked compilers
07:52karolherbst: or at least I wanted to
07:52karolherbst: but then I got hsw and tried out lto :D
07:53karolherbst: nfk: really you should try out a 8GB zram swap device
07:53karolherbst: + swapiness above 75
07:53nfk: karolherbst, also i'll point out i'm actually using alias mpv='sudo nice -n -15 ionice -c1 -n0 sudo -u me mpv'
07:54nfk: and portage naturaly also has tuned class and niceness
07:54karolherbst: doesn't have portage itself settings for this?
07:54karolherbst: but niceness won't improve the memory situation
07:54nfk: with that i'm even most of the time able to watch stuff while doing daily updates
07:55nfk: karolherbst, everything by default has the same values so effectively CFQ/BFQ just waste their potential
07:55nfk: in ideal world either metadata or a deamon would be fine tuning them per process
07:56karolherbst: how fast is your kernel timer?
07:56nfk: 300 Hz, i think
07:56nfk: 100 Hz has a terrible performance hit
07:56karolherbst: of course
07:56karolherbst: faster is better for desktops :D
07:56nfk: and 1000 is too slow
07:56karolherbst: it shouldn't
07:56karolherbst: why is it too slow?
07:57tobijk: how is it slow?
07:57nfk: also this is a quadcore so it's actually 300/4, iirc
07:57nfk: oh wait
07:57karolherbst: I have 8 cores
07:57karolherbst: and so I have 1000*8
07:57karolherbst: no problem
07:57tobijk: 1000*4 -> no problems
07:57karolherbst: full-prempt kernel?
07:57tobijk: runnung on 1000 for years
07:58karolherbst: for deskopt usage: 1000Hz timer + full-prempt is like the thing you need to have
07:58nfk: karolherbst, tobijk, sorry, i got it wrong, basically 1000 is too fast and 100 is too slow
07:58karolherbst: what means "too fast"?
07:58nfk: i guess i should take the third coffee but i'll rather not
07:58tobijk: nfk: you fear the overhead?
07:59nfk: karolherbst, tried forced-preemption but that had a very serious performance hit so i'm on voluntary preemption
07:59karolherbst: ahh tickelss idle
07:59karolherbst: also important
07:59nfk: it's surprisingly good
07:59nfk: i can even play a keyboard piano over PulseAudio no less
07:59karolherbst: yeah pulse :/
07:59nfk: does not even need jack
07:59karolherbst: okay, so pulse is better with vol prempt?
07:59nfk: if everything is right pulse is running in soft-realtime anyway
08:00nfk: and with very negative niceness which you can make even more negative, iirc
08:00nfk: which is what i have done
08:00nfk: basically i have fine tuned both pulse and mpv and it took multiple evenings to balance them out
08:00tobijk: for pulse: https://bbs.archlinux.org/viewtopic.php?id=170640
08:00nfk: it's easy to make pulse so nice it will hurt mpv
08:01nfk: arch forums? i dread what i'm gonna see there
08:01nfk: their wiki is good but users are...
08:01tobijk: i can give you another source :P
08:01nfk: tobijk, do note that i wrote some of the gentoo docs on PA
08:04nfk: karolherbst, is tickless idle even useful? i don't remember seeing notable changes with it
08:05karolherbst: power consumption
08:05karolherbst: less ticks on idle cores
08:05karolherbst: which means less wakeups
08:05nfk: which of course is nice to have but i actually prefer to have responsiveness over that
08:05karolherbst: doesn't have much todo with it
08:07karolherbst: you should try to think about why a ondemand goveror actually reduces battery life compared to performance ;)
08:07karolherbst: on laptops
08:08nfk: no way
08:08karolherbst: it does
08:08nfk: unless it of course did things too often or maybe did not scale voltage but i'm sure that's not the case
08:09karolherbst: short version: longer active cpu cores need more power then high voltaged short active ones
08:09nfk: but isn't ondemand doing exactly that?
08:09nfk: i'm pretty sure i'm using ondemand on this desktop as well
08:10chithead: ondemand is a little too slow for some interactive workloads
08:10nfk: it should be aggresively switching between the states especially in the past
08:11nfk: nowadays it's tweaked and should linger in mis well sometimes
08:11nfk: *middle states
08:11nfk: *as well
08:11nfk: i'm a bit impressed how it can miss some input and then just carry on
08:11nfk: in the past it would lead to issues
08:12karolherbst: this is for cpu freq
08:12karolherbst: this has nothing todo with sleep states
08:12nfk: yes, i'm talking about that
08:12karolherbst: for a laptop you always want to have max freq
08:13tobijk: uhm nope, i doubt that
08:13karolherbst: because if a task finishes early the core can sleep longer and earlier
08:13nfk: are you sure you're not trying to troll me?
08:13karolherbst: sleep states save more power than lower clocks
08:13karolherbst: currently my CPU uses like 0,01W
08:13karolherbst: because the cores are sleeping all the time
08:14karolherbst: max W is 47
08:14karolherbst: tell me how a task wich needs double times on half clock can save power?
08:14chithead: this is not completely true http://www.phoronix.com/scan.php?page=article&item=intel_pstate_linux315&num=6
08:14karolherbst: chithead: this is software
08:14karolherbst: not theory ;)
08:14nfk: karolherbst, what if it does not need all that speed in the first place?
08:14chithead: running on max cpu frequency and entering sleep states more often is sometimes better and sometimes worse
08:15karolherbst: yeah it depends a lot of the actual stuff you are doing
08:15karolherbst: but mostly its worse when you are at high load anyway
08:15nfk: also note that it's less about frequency and more about voltages
08:15karolherbst: and then battery life sucks anyway
08:15nfk: lower frequencies means lower voltage will keep the circuit stable
08:15chithead: android uses the interactive governor because the other methods did not meet their requirements
08:15nfk: the actual frequency changes power consumption very minimally
08:16karolherbst: can all cores go to sleep on a phone anyway?
08:16chithead: higher frequency requires higher voltage
08:16nfk: chithead, as in a daemon that controls stuff from userland?
08:16karolherbst: I thought one is always needed for celluar stuff
08:16chithead: no, interactive governor is part of the android kernel patchset
08:17nfk: karolherbst, cellular stuff would have its own chip in older designs and a programmable core in modern ones, i think
08:17tobijk: karolherbst: if something arrives at the modems, wake the cpu? :>
08:17karolherbst: yeah I think
08:17karolherbst: internet reduces battery life a lot too
08:17nfk: and if wakeup is handled by a physical key there's no reason why it could't shut down cpu and wait for an auxilary I/O chip to wake it up
08:17tobijk: how is that? ;-)
08:17karolherbst: if you want your phone to have enough batteries for days: disable internet
08:18karolherbst: even while doing nothing
08:18tobijk: you mean 3G+4G modes
08:18nfk: karolherbst, i think you're talking about wifi not internet
08:18karolherbst: nope, internet
08:18karolherbst: wifi just make the problem worse
08:19chithead: applications which poll the internet cause the cpu to wake up
08:19tobijk: well then you have apps waking everything up and doing god knows what
08:19nfk: wifi is a known battery drain and it's usually on even when you're not connected to anything for various purposes that tinfoilers do not like
08:19karolherbst: nfk: I highly doubt that
08:19karolherbst: its just the wrong victim
08:19karolherbst: of course it drains more powr
08:19karolherbst: but not as much as all the applications with internet stuff
08:19nfk: how? android will explicitly keep track of nearby wifis to calculate the rough position
08:20karolherbst: yeah, but thats not the fault of wifi
08:20nfk: you can't even have the high accuracy gpc mode without also having wifi on
08:20karolherbst: I can
08:20karolherbst: you just use the wrong android
08:20nfk: or non-stock android?
08:20karolherbst: yeah :D
08:21karolherbst: cm has three location things: high, battery and device
08:21nfk: on the proprietary android turning on high precission mode always turns on wifi even if it was off
08:21karolherbst: where device is gps only
08:21karolherbst: nad battery is wifi and mobile network :D
08:21karolherbst: yeah, its a mess
08:22karolherbst: but wifi on while having location turned is good enough
08:22nfk: and how's the signal quality without A-GPS?
08:22karolherbst: 20% power suckage over the day
08:22karolherbst: does it matter?
08:22nfk: i think i had grandma's phone set to GPS only and it couldn't get a lock even after minutes of trying
08:22karolherbst: gps is required for exact position anyway
08:23karolherbst: inside the house?
08:23nfk: yeah but walking around it
08:23karolherbst: gps works better outside ;)
08:23karolherbst: for obvious reasons
08:23nfk: the apperture should have been reasonable
08:23karolherbst: wifi and mobile are used for a rough pinpoint
08:23karolherbst: its a lot of guessing
08:24karolherbst: and smart algorithms
08:24nfk: but gets a lock in seconds
08:24nfk: even in a basement
08:24karolherbst: but wifi and mobile are mostly table lookupts
08:24karolherbst: wifi name + mobile towers => location
08:24nfk: be happy then
08:25nfk: it could also be sending received gps data for calculations on servers
08:26karolherbst: this phoronix chart is wrong
08:26karolherbst: by the way
08:26nfk: i think you'll find moronix is wrong
08:26karolherbst: chithead: he should have added total consumption as well
08:26karolherbst: what matters 10% less power consumption in avarage if you need double the time?
08:27chithead: the test method was not very smart indeed
08:27nfk: moronix is not very smart indeed
08:27nfk: fixed it for you
08:27karolherbst: acpi ondemand seems to be the winner there
08:28chithead: should have used a website and reloaded once a minute for some hours, then measured total kWh
08:28nfk: karolherbst, also i hope you have irqbalance running
08:28karolherbst: conservative and pstate ondemand have to long runtimes
08:28karolherbst: but pstate performance is pretty good
08:28nfk: though i guess for power saving you might actually want to have a different distribtution than what it gives
08:28karolherbst: my battery lif is down to 40% anyway
08:29karolherbst: so I don't care that much
08:29karolherbst: bat life sucks always
08:29karolherbst: I mean like tot capacity
08:29nfk: that's what i thought
08:30karolherbst: but regarding power usage there is a lot of BS going around
08:30karolherbst: mostly of them seems right at first
08:31karolherbst: but if you dig deeper, not so much anymore
08:33nfk: tobijk, i finally remembered to look at that URL and all i can say is - arch users are even worse in the head than i thought was possible
08:33nfk: at least the initial poster
08:33nfk: 1) uses VLC
08:33nfk: 2) has skype with message sounds
08:33nfk: 3) has skype
08:34tobijk: well it seems to happen with other workloads as well
08:34nfk: 4) complains about Skype abusing PA (probably)
08:35nfk: workloads? what you linked sounds like PA applying cork when skype emits a sound which is either due to skype telling it to do so or because PA does it always which might be because skype is too cool to care about PA features so the developers had to hack around
08:35nfk: also 4) uses VLC
08:35tobijk: i kinda like vlc
08:35nfk: see, whole 4 things wrong with that arch user
08:35shakesoda: what's the beef with VLC
08:35nfk: vlc is fucking shit, has always been, will always be
08:35shakesoda: skype sucks, I'll give you that
08:35tobijk: works fine here
08:35nfk: their developers have outright admited they have no idea what they're doing
08:36shakesoda: and they've still got a better media player than most... ouch.
08:36nfk: it never has worked fine
08:36nfk: it has epic bugs
08:36nfk: there were even files produced back in the day that exploited them to annoy vlc users
08:36tobijk: well then i'm lucky, never encountered those bugs :)
08:36nfk: which was the only way to force vlc to fix them
08:37nfk: some of those files also bsod windows
08:37nfk: overall vlc is made by people who do not understand multimedia
08:37nfk: on windows you should be using something else like mpc and whatnot, on linux you have Mplayer and mpv
08:37nfk: heck, i know people who use mplayer/mpv on windows
08:38shakesoda: and every other media player is apparently made by people who don't understand it either, since VLC consistently sucks less somehow.
08:38shakesoda: I use MPC-HC on windows, though, it is definitely better.
08:38nfk: you have not tried MPlayer/mpv
08:38shakesoda: mplayer sucked major ass at subtitle styles when I used it, haven't tried in a couple years.
08:39shakesoda: vlc at one point also sucked at that, granted.
08:39nfk: yeah, mpc-hc was for a while better than MPlayer/mpv but nowadays they should be comparable
08:40nfk: shakesoda, you're a bit wrong there, ASS was implemented on non-Windows by libass project which is related but independent from MPlayer and mpv
08:40nfk: vlc naturally implemented libass badly too because it's vlc
08:41shakesoda: yes, I remember it being implemented badly (but unusably so in mplayer at the time)
08:41nfk: and vlc decoders are obviously from ffmpeg which, surprise, also comes from MPlayer project or maybe related as I"m not sure which of the two was older
08:42shakesoda: my timelines are very likely skewed from outdated packages when I was doing this stuff though
08:42shakesoda: it's kinda moot now
08:42nfk: shakesoda, again, you are probably wrong or comparing different versions, MPlayer had libass before VLC
08:42shakesoda: I don't doubt it.
08:42nfk: because libass primary devs (all 2 back then) worked with MPlayer project
08:43shakesoda: this was several years ago and I was definitely using debian packages which are guaranteed to be a ridiculous mix of random dates.
08:44shakesoda: years ago :P
08:44nfk: what several years ago? libass is almost 10 years old by now
08:44shakesoda: old habits die hard
08:44nfk: granted it was very basic at first and most of the cool work was done by the second lead developer and all that but it's been pretty much in its current state since some 2009
08:46nfk: i think it was pretty usable bar some bugs in around 2008 possibly 2007
08:46nfk: probably 2008
08:51tobijk: karolherbst: btw how is it going with pcie link reclocking? :>
08:53nfk: tobijk, btw, mpv project lead is a debian user so debian has remarkably new versions at least in testing, iirc
08:53nfk: not bleeding edge but possibly from even this year which is incredibly fresh for debian
08:53nfk: so do give it a try
08:53nfk: or compile from source to get the latest git version
08:54tobijk: yeah why not, i will
08:54tobijk: i'm lazy, so lets see if i find a repo :)
08:54nfk: not sure about the stable branch but the main development branch has featurs only found in mpc-hc and maybe MPlayer
08:54nfk: tobijk, should be as easy as sudo apt-get install mpv
08:55nfk: provided you're on at least testing
08:55nfk: or know how to get packages from testing on stable
08:56tobijk: you know the latest stable version by heart?
08:57nfk: i'm using the live ebuild on gentoo because the snapshots are outdated due to lack of a maintainer
08:57nfk: but a volunteer is keeping the live ebuild alive
08:57tobijk: i have 0.9.2 now, lets test it :)
08:57nfk: anything above 0.6 should be usable though the cool stuff might not even have yet been released, not sure
08:58nfk: the easiest way to test if you have the cool stuff is by opening man mpv and searching for interpolation
08:58nfk: if you have that under OpenGL options then you're golden
09:05tobijk: nfk: works fine on a fist glance and i even have the interpolation below opengl :)
09:06nfk: then try turning it on
09:07nfk: should be as easy as -vo opengl-hq=interpolation on the command line
09:07nfk: or vo=opengl-hq:interpolation in ~/.mpv/config
09:08nfk: note that you can configure vo_opengl such that it will lag without a well clocked and powerful GPU
09:11tobijk: nfk: works fine here, seems my system is fine
09:12nfk: tobijk, my config: https://paste.kde.org/ppzgupqxi/ibpfhr
09:12nfk: the paste will be there for a week
09:13nfk: in particular scales can easily tear your GPU a new one
09:13nfk: try stuff like ewa_lanczossharp for scale, cscale and if that does not do it tscale which certainly will pwn most gpus
09:14nfk: note that it might look not very well if work at all for tscale
09:14nfk: you can also try color management which also adds extra load on GPU
09:15tobijk: heh i think i'm fine with most efaults for now :)
09:17nfk: also if you wonder where the gui like overlay went, it's probably due to osc=no in my config
09:18nfk: i'm too oldschool for gui over my video
09:20nfk: tobijk, oh, and use-filedir-conf=yes is a potential security hazard as mpv will load parse and use mpv.conf files in the same directory as the file it's playing back
09:20nfk: and you're pretty explicitly required to ensure there's nothing weird there
09:20tobijk: that is...stupid :O
09:21nfk: i can't recall how but in a very convulated way you could have it load unsafe stuff from internet, iirc
09:21nfk: it's off by default but i have it on for "reason"
09:23tobijk: nfk: vo=drm:devpath=/dev/dri/card1 seems dodgy
09:24nfk: works for me :P
09:24tobijk: on my system it can easily be anotherone
09:24nfk: also it only works if you do -vo drm
09:25nfk: for some fucked up reason the vo=...,drm, will not load [vo.drm] section even though [vo.opengl-hq] and [vo.vdpau] do actually work
09:25nfk: tobijk, also obviously requires kms for resolution and drm for SPEED and better do not try it from X11 anyway
09:26nfk: but, yes, it's surprisingly usable from tty
09:27nfk: and despite being called drm it's almost certainly KMS dependent as well
09:27karolherbst: tobijk: well, I wait for skeggsb to land his cleanups
09:28karolherbst: he wants to add a pci subdev I want to use
09:28tobijk: who does not hae kms nowadays?!
09:28tobijk: karolherbst: ah i thought he already did
09:28karolherbst: also here is everything you need: https://gist.github.com/karolherbst/5fdd4a543d20916bc362
09:28karolherbst: only width change is missing, but imirkin knows how to do that
09:29nfk: tobijk, nvidia
09:29tobijk: karolherbst: it works fine for me, the biggest downfall is the link reset on going to sleep
09:29tobijk: which it does really frequently
09:30tobijk: nfk: who uses nvidia anyway? :D
09:30imirkin: karolherbst: well, only on pre-kepler
09:30imirkin: karolherbst: and i've long forgotten by now. but hopefully logs remember.
09:30nfk: tobijk, oh and a word of caution, while mpv's VDPAU has openg interop it's at best experimental on nvidia and known to do very bad things on nouveau
09:31nfk: naturally it's off in my config but if you do try it, be ready to see a hard lockup
09:31imirkin: yeah. just use mplayer. mplayer works. everything else doesn't.
09:31nfk: even the opengl interop?
09:32nfk: and does mplayer have frame interpolation?
09:32imirkin: i didn't say mplayer had more features
09:32nfk: and color management? (might actually have that one nowadays but mplayer2/mpv was first)
09:32imirkin: i said that mplayer is the thing that actually successfully always plays every video
09:32nfk: so does mpv
09:32imirkin: everything else has more features but always misses one big one "playing videos"
09:32nfk: which is a fork of mplayer2 which is a fork of MPlayer
09:33imirkin: and yet the managed to screw it up
09:33nfk: imirkin, i can't recall a file that didn't play
09:33imirkin: i have lots of files
09:33imirkin: some of which even have corruption
09:33nfk: specifically for mpv?
09:33imirkin: i don't go re-testing everything on every $player release
09:33nfk: which is weird given both use ffmpeg/libav with preference of ffmpeg
09:34imirkin: iirc mplayer has tons of local patches, but perhaps those have been upstreamed by now
09:34nfk: are you sure you weren't using libav for mpv and ffmpeg for mplayer?
09:34imirkin: i used whatever it used.
09:34nfk: libav is notorious for "not invented here"
09:34nfk: which also brought down mplayer2
09:35nfk: ffmpeg has tons of patches that libav hadn't reinvented back when i last heard about state of things regarding libav
09:36imirkin: libav is a semi-recent thing though, no?
09:36nfk: and at least mpv can use either and so might MPlayer
09:36imirkin: (i always confuse it with libavformat, which has been around forever)
09:36nfk: if half a decade is recent to you, sure
09:36imirkin: yeah, 5 years is recent
09:36imirkin: compared to mplayer's timeline :)
09:36nfk: libav forked from ffmpeg over things and then both sides went to sue each other
09:37nfk: under french law
09:37tobijk: sounds fun
09:37nfk: so it's trebian, iirc
09:37imirkin: well, last i checked (which admittedly was a while ago), mplayer kept an in-tree ffmpeg
09:38nfk: and mpv provides mpv-build repo which has a script that fetches a select ffmpeg version but the mpv itself will build with whatever the system has
09:38imirkin: hmmm... my last mplayer contribution might have been improved xinerama support... forget if they merged it. ~04 or so.
09:38nfk: and distros thave have libav developers in them switched over to libav
09:39imirkin: yeah, i don't really care about the full story :p
09:39nfk: wait, weren't you around 11 back then?
09:39imirkin: that would put me at 22 today?
09:39nfk: too much?
09:40imirkin: closer to 220 :p
09:40nfk: wait, isn't nouveau also under french law?
09:41imirkin: there is no nouveau legal entity
09:41nfk: which was the problem with ffmpeg, iirc
09:41imirkin: but 'nouveau' is a french word :)
09:41nfk: someone oned the domain, someone else owned the logo
09:41imirkin: nouveau has no domain, no logo
09:41imirkin: problem solved!
09:42nfk: no one owned the project title but domain owner just went "ffmpeg iz mine"
09:43tobijk: just name it ffmpeg2 and distribute that ffmpeg is dead
09:43tobijk: problem solved
09:43nfk: trademarked :P
09:43tobijk: or whatever you want
09:43nfk: i don't recall the exact reasoning but they couldn't use ffmpeg2 i think
09:47tobijk: imirkin: any preferences/ideas for a pass in the glsl linker where we sort in cull behind clip? as marek proposed last
09:49imirkin: tobijk: yeah so basically you'd have a lowering pass to combine gl_ClipDistance/gl_CullDistance into a gl_ClipCullDistanceMESA + shader level clip (or cull) mask that tells you which is which.
09:49nfk: tobijk, oh, btw, you can have your own shaders with mpv
09:49nfk: but i'm sure mplayer also has that feature
09:50tobijk: nfk: i dont even want to go there :>
09:50nfk: but you could
09:50tobijk: imirkin: well yeah the basic was made clear, but i'm not pretty firm when it comes to the glsl
09:50nfk: now imagine somehow making mpv load a shader from internet via that filedir-config stuff
09:50tobijk: just do somthing and see waht comes out? :D
09:51imirkin: tobijk: take a look at the lower_clip_distance pass
09:51imirkin: tobijk: it lowers int gl_ClipDistance into ivec2 gl_ClipDistanceMESA
09:51imirkin: [or something along those lines]
09:52tobijk: imirkin: yeah i have an lower_cull_distance as well, so best would be to combine those and do the masking while in the way
09:52imirkin: tobijk: i would create a whole separate pass that was run in place of those
09:53tobijk: mh fine
09:53nfk: tobijk, oh, and one last correction: it's -profile vo.drm if you want to make it use [vo.drm] settings from command line, has to be a bug but whatever
09:54tobijk: nfk: as the vo.drm just had the dodgy line anyway, i removed it for now
09:55nfk: as you can see generally stuff like [ao.pulse] and [vo.opengl-hq] actually autoload when ao_pulse and vo_opengl-hq are in use but it's either due to them being first in the respective lists or something else
09:55nfk: conceptually it should not mattter so it's a bug somewhere but i can't be bothered to report it
10:06nfk: and for your information i only have one GPU :P
10:14karolherbst: imirkin: yeah, kepler :/
10:14karolherbst: I tried several, but ...
10:14karolherbst: we really need somebody with the blob you can do this
10:14karolherbst: nothing helps more than a mmiotrace with changed widths on kepler+
10:14karolherbst: if its actually possible at all
10:15karolherbst: I was actually thinking if a hotplugged card could trigger a width change somehow if too many lanes are used
10:54imirkin_: karolherbst: what was that in answer to?
10:54imirkin_: oh lane width changes on kepler?
10:54imirkin_: i got the blob going on my gk208, but i couldn't find a way to reduce lane width
10:54imirkin_: it does switch between 2.5GT/s and 8GT/s under load
10:55specing: why are these cards attached to x16 anyway? Isn't x8 more than enough even for the biggest of cards?
10:56imirkin_: specing: latency i assume
10:58specing: they are x16 in laptops as well
10:58specing: the mxm2 connector on my 8600m gt is ... full of pins
10:58specing: 64 for 16 lanes and who knows how many for control signals and power/ground
10:59imirkin_: gk208 is only x8
10:59specing: I have g86
10:59imirkin_: ah, that didn't even do pcie 2.0
10:59specing: no, G84M
10:59specing: goddamn marketing codenames and whatnot
10:59imirkin_: i.e. you're stuck at 2.5GT/s
11:00imirkin_: yeah, that's why i stick to chip numbers
11:00imirkin_: chip id's don't lie
11:14karolherbst: specing: I have games where I get a 25% speed bump just by going from 2.5GT/s to 8.0GT/s pcie speed
11:15specing: karolherbst: (irrelevant) are those games open source?
11:16karolherbst: only saw that in talos and serious sam 3, so maybe it has something todo with their engine
11:22tobijk: specing: not everyone is using only open source ;-)
11:23AnarquistaLibre: What's the most modern card that works with nouveau drivers I can buy that doesn't use any non-free software?
11:23AnarquistaLibre: or other bits
11:23imirkin_: AnarquistaLibre: do you have a good definition of "non-free software"?
11:23tobijk: is the kernel in its standard version free?
11:23AnarquistaLibre: The GNU/FSF definition.
11:24imirkin_: AnarquistaLibre: for example, do you include the board-specific vbios which includes instructions for how to init the board as "non-free"?
11:24AnarquistaLibre: Hmm Idk, lol.
11:24AnarquistaLibre: I'm just looking for a card that will work with all free software. Can't find the answer online.
11:25imirkin_: if so, i think only the oldest Riva TNT's didn't have a vbios with init instructions
11:25tobijk: well but that are the little differences :)
11:25imirkin_: like the diamond one probably
11:25tobijk: imirkin_: btw i have an old riva tnt2
11:25imirkin_: tobijk: that def has a vbios
11:26tobijk: found it last wekk :)
11:26AnarquistaLibre: Well does that make it nonfree?
11:26AnarquistaLibre: What's the most modern card I can buy that would work?
11:26imirkin_: tobijk: should work pretty well with nouveau
11:26imirkin_: AnarquistaLibre: it's all a question of what one means by non-free
11:26imirkin_: and don't point me to the GNU/FSF definition, since i don't understand it
11:26AnarquistaLibre: The accepted definition.
11:26AnarquistaLibre: What is your definition?
11:26tobijk: well i wont run something on it, only for testing purpose, if there are open tests
11:27imirkin_: my definition is code running on the cpu is open == free
11:27tobijk: for bugs etc.
11:27AnarquistaLibre: Okay, what's the most modern card, going by your definition, that I could use?
11:27imirkin_: any gpu should do fine
11:27imirkin_: i'd definitely recommend amd gpu's over nvidia
11:28specing: AnarquistaLibre: if you plan to run any games that interact with the internet on it, then you may as well go blobbed as said games are all hilariously insecure
11:28specing: AnarquistaLibre: if you have no intention of running games, then core2duo GMA4500HD iGPU is perfectly fine
11:28imirkin_: AnarquistaLibre: stay away from nvidia maxwell gpu's for now though
11:28imirkin_: those don't work very well with nouveau
11:29AnarquistaLibre: I do wanna run games, but I wanna be as FOSS as possible.
11:29specing: I wouldn't recommend radeon to anyone
11:29specing: I had to patch the kernel to fix a hang on a 7 year old card!
11:29tobijk: AnarquistaLibre: then go for an amd one
11:29imirkin_: specing: intel is probably best, but they only go so far wrt performance
11:29specing: AnarquistaLibre: get a seperate computer for games and only use it for insecure stuff
11:29RSpliet: imirkin_: I think some, if not all, Riva 128's have a VBIOS
11:30specing: imirkin_: Enemy territory runs on GMA 4500 HD :)
11:30imirkin_: RSpliet: sure, but does that vbios include code tables
11:30imirkin_: RSpliet: i mean init scripts
11:30AnarquistaLibre: So open source isn't good for security?
11:30imirkin_: AnarquistaLibre: no, internet isn't good for security.
11:31specing: AnarquistaLibre: the problem is that game devs simply do not care about security
11:31imirkin_: specing: hard enough to get the damn thing to work, you want it to be secure too??
11:31specing: I tried pointing out obvious flaws (downloading unsigned code and executing it) and got called paranoid
11:31specing: imirkin_: yes.
11:32RSpliet: imirkin_: good question... hasn't that always been the hack to get output on the card before the machine was POSTed?
11:32specing: whois AnarquistaLibre
11:32RSpliet: so that you could actually see errors during POST :-P
11:32imirkin_: specing: you'd greatly enjoy the imageio library someone pointed out to me recently -- if you don't have something locally, it just downloads it.
11:32specing: (stalker mode activated)
11:33imirkin_: RSpliet: that's the x86 code, and all vbios's have it
11:33specing: imirkin_: there are more such things yes, e.g. GNU Octave replacement for MATLAB
11:33imirkin_: RSpliet: i'm talking about init scripts that are called by e.g. nouveau to bring the thing up
11:34specing: seriously, the more I know the more pissed off I get
11:34imirkin_: specing: just use R. it's perfect in every way.
11:34imirkin_: except you can't use it to computation. other than that it's great.
11:34specing: I need to get my logger's license so I can have a source of income that doesen't involve computers
11:34AnarquistaLibre: Meh, a totally free desktop sounds impossible. Guess I'll just stick to my Lenovo X200 with libreboot.
11:34imirkin_: talk about security... heh.
11:35tobijk: your stiehl chainsaw may have internet :P
11:35specing: AnarquistaLibre: it is not impossible, just get a different computer for insecure stuff :)
11:35imirkin_: AnarquistaLibre: but really you're just asking the wrong questions, and misinterpreting our answers.
11:35specing: AnarquistaLibre: I have another T400 with radeon for games
11:35AnarquistaLibre: How? All I want to know is what will give me the most free computer.
11:35imirkin_: AnarquistaLibre: for example specing is just talking about security which is one reason one might want a free-only install, but far from the only one.
11:35AnarquistaLibre: I don't want another computer really
11:36AnarquistaLibre: Just want one, as free as it can be.
11:36RSpliet: then think of the trees and birds, and don't buy a new one ;-)
11:37RSpliet: that Thinkpad X200 has Intel X4500HD, which is about as libre as it gets as far as GPUs go
11:38specing: AnarquistaLibre: I don't think you can just add a GPU to the X200
11:39specing: AnarquistaLibre: it is possible with the T400 and 2503 dock, but I don't know if the X200 has something similar
11:39RSpliet: or... is that pre-i915?
11:39AnarquistaLibre: I was going to build a desktop.
11:39AnarquistaLibre: But I don't know how to go about it.
11:39specing: AnarquistaLibre: < AnarquistaLibre> I don't want another computer really
11:40AnarquistaLibre: They also do make express PCI slots that connect through the Expressport or whatever it's called on laptops.
11:40specing: AnarquistaLibre: libreboot now supports a server board
11:40specing: AnarquistaLibre: yes, they do, briefly forgot about it
11:41specing: AnarquistaLibre: but it is 250 MB/s max. It is going to bottleneck every single post 2008 card
11:41AnarquistaLibre: I don't want two desktops. I should clarify. I don't mind having a free laptop and desktop. Just don't want one free desktop and one non-free desktop.
11:42AnarquistaLibre: So don't buy something really new.
11:42specing: AnarquistaLibre: ok. Then as I said, libreboot now supports an old server board and one newer
11:42specing: AnarquistaLibre: ask in #libreboot for the desktop build
11:43AnarquistaLibre: What GPU should I get for it?
11:43specing: as for cards, I think imirkin_ told me that maxwell GM0* is ok (e.g. Geforce 250 GTX)
11:43AnarquistaLibre: Looking to get one as free as possible if I can
11:43specing: imirkin_: would it be possible to publish a list of "recommended" cards
11:44imirkin_: specing: GM10x you mean? Which is GTX 750
11:44specing: There are many of us who would like to know
11:44AnarquistaLibre: Any of you nouveau devs?
11:44imirkin_: specing: only in terms of chips, not in terms of marketing names
11:44shakesoda: AnarquistaLibre: multiple people talking right now are
11:45shakesoda: (not me, though!)
11:45imirkin_: specing: maybe i should make a FeatureMatrix2 page which just talks about high-level things per-chip
11:46specing: imirkin_: perhaps also how they perform on the open source games using nouveau to drive it
11:47specing: (really high level)
11:47RSpliet: specing: no
11:47RSpliet: performance is too much dependent on the marketed name (clock speeds, RAM type differences)
11:47RSpliet: you can't do that high level
11:47imirkin_: specing: i'm thinking the list of things people are interested in are... kms, ddx, 3d (e.g. max GL version), yuv overlay (for older gpu's), xvmc, vdpau, reclocking
11:48imirkin_: RSpliet: did i miss anything obvious?
11:48imirkin_: i guess yuv overlay is supported everywhere that has a useful one so i can drop it
11:48imirkin_: [but not used anywhere. heh.]
11:48specing: imirkin_: I don't know what half the things you listed mean
11:48imirkin_: specing: which half
11:49specing: 'ddx', 'yuv overlay', 'xvmc'
11:49RSpliet: imirkin_: compute (OpenCL) perhaps?
11:49imirkin_: XvMC = mpeg1/2 acceleration thing
11:49huehner: imirkin_: maybe even higher level, 2d works (modesetting, maybe accell), 3d technically works, 3d works resonably fast (reclocking)
11:50imirkin_: huehner: too subjective i think. there's no more "2d" anyways.
11:50imirkin_: everyone has crazy compositors and other bs
11:50RSpliet: 2d nowadays is just flat 3d
11:50specing: imirkin_: does anyone still use mpeg1/2? (I know they are used in mpeg3 and 4, but I'm just pointing out that not everyone is a gfx driver dev)
11:50specing: I hate it how I have to spend a month researching my options before purchase
11:51huehner: imirkin_: true probably for the audience of the simplified page most likely they'll have compositors and stuff
11:51specing: it would be much better if I could just navigate to nouveau.org and know withing 5 minutes what to buy
11:51imirkin_: specing: easy -- buy amd
11:51specing: and after 1 month of research I bought an ivy bridge desktop only to find out a year later that it is backdoored!
11:52specing: imirkin_: but not all AMD is okay
11:52RSpliet: isn't every Intel CPU unsafe?
11:52imirkin_: all amd is a lot more ok than nouveau is
11:52specing: RSpliet: yes, but noone told me that 2 years ago
11:52specing: RSpliet: well, every post core2duo
11:53RSpliet: pre- too... their microcode is a "don't ask, we don't tell" beast
11:53specing: imirkin_: I have very bad experience with radeon
11:53imirkin_: specing: how's your experience with nouveau?
11:53specing: imirkin_: both with the closed driver and open one
11:53specing: imirkin_: 1/1 would buy again
11:53specing: radeon is 0/3 piece of trash
11:54chithead: it is known for several years that intel amt can be used to run malware that has dma access to your system https://events.ccc.de/congress/2013/Fahrplan/events/5380.html
11:54imirkin_: specing: heh ok. well i just see the people that file bugs :)
11:55RSpliet: chithead: juicy
11:55specing: the only problem I see with nvidia is that they die a fiery death
11:55specing: but then I can whip out the heatgun and fix it
11:56specing: with radeon there are just software issues everywhere
11:56specing: and in that case the only thing to do is join #radeon and start crying
11:56imirkin_: wait, for your 1/1 you mean with nvidia blob? or nouveau?
11:56specing: imirkin_: both
11:56imirkin_: huh, ok
11:57specing: I have used the nvidia blob for over 5 years
11:57imirkin_: nvidia blob tends to be quite good
11:57specing: and nouveau the last year
11:57specing: imirkin_: yes, it is excellent
11:57imirkin_: nouveau can fail if you start doing things that the developers don't do
11:58specing: imirkin_: do you test with certain games/ 3D programs before release?
11:58imirkin_: what release?
11:58imirkin_: i do use the piglit test suite
11:58imirkin_: there is no "nouveau" software package.
11:59imirkin_: and on occasion i'll load up a game or two
11:59imirkin_: plus there's my daily use
11:59imirkin_: i.e. no-compositor no-gl no-nothing desktop :)
11:59imirkin_: [actually chrome probably gives it a bit of a workout]
12:04specing: would nouveau work on cards branded as "Geforce 2 MX MAX" and "Quadro 4"?
12:04specing: both are AGP
12:04specing: just asking, they are in my drawer atm
12:04imirkin_: specing: yeah, but you only get GL 1.2
12:05imirkin_: + a few exts
12:05imirkin_: curro didn't add cubemap support, and i was too lazy to do it too.
12:06imirkin_: hm, also missing border clamp for GL 1.3
12:06imirkin_: i wonder how nvidia did it.
12:06imirkin_: iirc they provided as much as GL 1.4
12:07imirkin_: maybe they just fell off the hw path for that
12:07specing: well, not really looking for GL, more like just a basic desktop if I ever wanted to power on some of the ancient hardware I hoarded
12:08imirkin_: yeah, should work fine. xv will suck though because we don't make use of the overlays (even though they're exposed at the kms level)
12:08specing: s/hoarded/never thrashed/
12:08imirkin_: is there a difference? :p
12:10specing: but I'd still like to see gpu recommendations
12:11imirkin_: specing: the problem is that there's little correlation between what's on the outside of the box and what's inside
12:11imirkin_: you can't recommend someone to buy a particular chip -- that info is just not provided on the packaging
12:11specing: and if, like you said, marketing is to be avoided due to different speeds / ram size, then perhaps list required minimum and an example of a finished card that satisfies it?
12:12imirkin_: also it appears that people will tend to have more issues with certain ways of using nouveau more than with specific gpu's
12:13specing: but chip series is roughly correlated with marketing model name? And sometimes this model/chip pair is given, e.g. on review sites
12:13imirkin_: definitely not
12:13imirkin_: GT 630 can be a fermi or a kepler or a kepler2 for example
12:13specing: this is so hopeless :(
12:14imirkin_: there are 2 different GK208's running around... same pci id space, but one identifies as NV108 and one as NV106
12:14imirkin_: until we saw the NV106's and added support for them, they plain didn't work in nouveau
12:15specing: I assume the radeon situation is not much better?
12:15imirkin_: specing: it is... the model numbers map precisely to chips (although not always as sequentially as you might hope)
12:16imirkin_: specing: search for "radeon decoder ring"
12:16nfk: [21:34] <specing> I need to get my logger's license so I can have a source of income that doesen't involve computers // i intend to eventually get a farmer's certification so that i have a backup in case someone proves me right and brings down most of the worlds computer systems
12:16karolherbst: imirkin_: do you think you will find some time checking my metro 3GB RAM trace, just to confirm the issue? I would like to know on which kernels and cards this bug occur and maybe I could bisect it with the right hardware
12:16imirkin_: karolherbst: not looking forward to crashing my system
12:16karolherbst: it only uses 3GB RAM top
12:17imirkin_: karolherbst: so... unlikely. at least directly. could check it out on reator.
12:17karolherbst: I stoped the application pretty early
12:17specing: imirkin_: so basically nouveau's recommendation is to go with radeon?
12:17karolherbst: at least on my system nothing happens
12:17karolherbst: but radeon is less free :O
12:17specing: imirkin_: that reduces the buying search space by 1/2
12:17specing: I hate radeon GPUs
12:17specing: total software clusterfuck
12:17nfk: Karlton, btw, many hours later that same firefox instance after more browsing is now at less VIRT and RES than back then
12:18specing: I had to patch 4.0.8 kernel for HD3470 hangs a week ago
12:18imirkin_: specing: ok, well tbh, my last radeon gpu was a Radeon 7200 (or 7000).
12:18specing: that cards is like 8 years old
12:18imirkin_: aka r100. so quite some time ago.
12:18karolherbst: in theory: if you reimplement some reg writes like in a non free firmware in GPL code, does it make the non understood reg writes free?
12:18imirkin_: try 15 :p
12:18imirkin_: it was the VIVO version. and the VIVO eventually worked on linux, which was highly surprising.
12:19nfk: specing, btw, amd might get bought out by microsoft so i'd wait for intel skylake, it should have at least an updated gpu compared to prevous stuff
12:19specing: imirkin_: I still have a radeon 9000
12:19specing: imirkin_: I think
12:19specing: imirkin_: and I still remember the software clustefuck around it
12:19nfk: though god knows how well it works, at least you might need ubuntu 15.10 to get majority out of it, i expect
12:19imirkin_: nfk: and of course after skylake is out, you should wait until the next one, since it'll have an updated gpu.
12:19specing: imirkin_: just like on HD3xxx desktop card and HD3470 mobility one
12:20nfk: imirkin_, stop reading my mind
12:20nfk: and skylake will be out in a month or less
12:20specing: nfk: 1) it will be backdoored 2) skylage igpu will require firmware blobs
12:20nfk: though god knows how long till it's on the shelves near specing
12:20karolherbst: was a reason stated at all?
12:20specing: my budged for backdoored stuff is exactly $0
12:21nfk: it will have new memory protection instructions that will basically make it a no op to have array bounds checking
12:21karolherbst: specing: do you own a phone?
12:21karolherbst: how did you get it?
12:21nfk: and as long as your os is compiled with something like gcc 5.0 it should be there since glibc got patched last year, iirc
12:21specing: karolherbst: I own a mobile backdoored real-time tracking device, yes
12:22nfk: specing, and you really think amd is that much better? at least you should have complained about intel being evil
12:22specing: nfk: I don't buy AMD because noone makes quality AMD computers
12:22specing: nfk: I buy core2duo thinkpads
12:22nfk: because while they are not as evil as oracle they're certainly with very questionable ethics
12:22nfk: made by lenovo?
12:22specing: nfk: btw we have a project to make AMD laptops at openlunchbox.com
12:23specing: nfk: made by Wistron (Lenovo doesen't make motherboards) and its bios replaced with libreboot
12:24nfk: and you checked the source code of both that libreboot and the toolchain that built it or can otherwise verify it was built from the real sources?
12:25nfk: or at least someone you trust with your life did it and is also technically capable to do it? thought as much
12:25nfk: i'd love to run everything open source but the idea that it's automatically secure is a fallacy that openssl proved wrong nicely
12:25nfk: yet people still buy that shit
12:26specing: I wanted to get rid of the ME. libreboot accomplishes that
12:26RSpliet: nfk: thanks... now I can't sleep anymore tonight :-(
12:26specing: nfk: the backdoor
12:26nfk: you will, don't worry
12:26nfk: RSpliet, if it's too hard, watch some animu and you'll eventually not give a damn
12:26specing: nfk: http://libreboot.org/faq/#intelme
12:26nfk: if that does not work, try some strong alcohol
12:28specing: nfk: however, just because I cannot audit all the tens of millions of lines of C (disaster) and C++ (disaster v2.0) code that runs on my system doesen't mean I should just give up and install windows
12:28nfk: specing, sounds a bit blown up, for one my system would fall into the "definitely has it" but it's actually ICH9
12:28nfk: also i was under impression only server mobos had it
12:29specing: nfk: all core2duo gen3 and later have it, with the exception of Intel Quark
12:29nfk: specing, yeah, but the idea that you're any safer unless a trusted party has audited it and reported or fixed the bugs and given you the fixed version is worth jack shit
12:29nfk: and now consider that something like 0.1 million a year are spent on auditing open source code by white house alone
12:30specing: nfk: however, just because I cannot audit all the tens of millions of lines of C (disaster) and C++ (disaster v2.0) code that runs on my system doesen't mean I should just give up and install windows
12:30nfk: chances are nsa/pentagon has something like a thousand bugs they can mix together for a good acid trip
12:30specing: I don't doubt it
12:30nfk: then stop claiming open source makes you safe
12:30specing: its their job
12:31specing: nfk: I never did
12:31nfk: and better start learning logging, you might need it
12:31specing: you are the one that went on a rampage with a purpose that is still not clear to me
12:32nfk: from time to time i wonder why no one has yet tried taking down majority of it systems, it's without any doubt possible and if done swiftly enough by the time a reaction starts most systems will be out of order
12:32nfk: sure, world will not end overnight as internet and military systems and some other weirdness will keep running but most civilian western systems will be out of commision as if nuked
12:33nfk: and with internet of things it could be as effective as emp even
12:39imirkin_: mupuf: btw last i checked neither the G80 nor the GM107 were plugged into reator...
12:39imirkin_: mupuf: would it be possible to get one of those in there?
13:05hakzsam: imirkin_, hey, it's my fault because I'm checking perf counters configuration on fermi
13:05imirkin_: ah ok. go ahead then :)
13:06imirkin_: mupuf: you need an auto-loader robot that will switch gpu's on demand out of your library ;)
13:06hakzsam: hehe :)
13:06imirkin_: like the old cd tray thingies
13:06hakzsam: or buy reator2 ;)
13:06imirkin_: you'd have to buy reator20 to make all of his gpu's available...
13:36RSpliet: imirkin_: this would be a good start: http://www.trustedsec.com/files/cuda_1.jpg
13:37imirkin_: yeah, i pointed out many-pcie x16 boards before
13:37imirkin_: trouble is, they're very expensive
13:38RSpliet: and they require a pocket power plant to run
13:38mupuf: imirkin_: yeah, I thought about that :D
13:41imirkin_: 7 x16 (but really x8) slots: http://www.supermicro.com/products/motherboard/QPI/5500/X8DTH-i.cfm
13:44tobijk: thats a beast :)
13:47RSpliet: hmm yes, only... it doesn't really look like you can *actually* plug in 7 GPUs
13:47RSpliet: unless you use extension cords or the like... I don't think it fits physically otherwise
13:48tobijk: not the 2x formfactors, thats for sure :)
13:48specing: how about a riser card?
13:48specing: like those coin miners used
13:48RSpliet: you'd have to
13:51mupuf: I would have problems finding a case for that!
13:52RSpliet: got a laser cutter?
13:52imirkin_: cases are for sissies anyways
13:53specing: mount it on the wall
13:53RSpliet: the only case I need is a case of beer
13:54mupuf: ah ah
13:54imirkin_: RSpliet: weaning yourself off the kegs i guess?
13:54mupuf: well, I need to have it flat so as I do not need to secure the gpu with a screw
13:56tobijk: if you dont need a case: http://www.supermicro.com/products/motherboard/Xeon/C600/X9DRG-OTF-CPU.cfm :D
13:58imirkin_: tobijk: with a case: http://www.supermicro.com/products/system/4U/4027/SYS-4027GR-TRT.cfm
13:58imirkin_: 24x 2.5" SAS/SATA drives :)
13:59tobijk: i really like the wall-mounted idea :)
14:01glennk:imagines repurposing a tape robot for swapping around cards
14:02RSpliet: I thought you could hire minions for that...
14:02imirkin_: glennk: yeah, that's exactly what i had in mind when i was making the gpu swap suggestion
14:03tobijk: plugging in the card is easy, find the right position for the power connectors, not that easy :D
14:03RSpliet: like these https://www.youtube.com/watch?v=IiX09kiodfs&t=115s ?
14:05tobijk: he is typing really s l o w, darn it :D
14:06specing: how about having an array of inexpensive desktop mainboards bought used from ebay and all attached to a central master and running coreboot (for bios over serial line)?
14:06tobijk: specing: well you need cpus for it as well + power supply
14:07specing: they can all be suspended to ram when not in use
14:07specing: tobijk: used...
14:07mupuf: anything mechanical is doomed to failure in my appartment :s
14:08mupuf: the approach mwk took was to have a ton of mobos
14:08imirkin_: mupuf: and electrical works much better?
14:08mupuf: well, it is usually small enough that I can keep it contained :D
14:08tobijk: make it electrical and loud, if it fails, mupuf will fix it instantly
14:10specing: or he'll go deaf
14:10mupuf: pretend-deaf will likely be my condition :D
14:11tobijk: make it loud enough the people next door come knocking :P
14:12specing: when it fails, just play some skrillex
14:12mupuf: I\m in finland, they would never do that :D
14:12specing: meaning his nearest neighbour is 10km away
14:13tobijk: specing: but he only has an appartment? :D
14:13specing: tobijk: *shrug*
14:35mupuf: specing: you got to give points to tobijk for following :p
14:40karolherbst: mupuf: got some time to spend? :p
14:40mupuf: a tad
14:40mupuf: I put the children to bed .... err, parents
14:40karolherbst: so you put the parents to bed? how nice of you
14:41karolherbst: its okay though, we could try stuff out in a week or somethig, because before that I won't have my laptop to play with
14:43mupuf: as you like
14:44mupuf: still need to find the damn regs
14:44karolherbst: but imirkin_ mentioned sometimes that I should get access to reator ;)
14:44karolherbst: ahh no, I was talking about my metro memory issue this time :/
14:44karolherbst: but regarding this: I think I will try to get mmiotrace working here
14:44karolherbst: and try to capture only reclocking
14:44karolherbst: should be fairly easy here
14:48mupuf: oh, do I have your ssh key?
14:49mupuf: yeah, got it
14:49karolherbst: yeah, but we shouldn't do that now
14:49karolherbst: time is a bit limited on my side
14:50mupuf: well, I added you already
14:53mupuf: now you know everything
14:53mupuf: just don;t destroy it
14:53mupuf: that would be spoiling the fun of others
14:54karolherbst: I know
14:54karolherbst: I just play around on my local machine with nvidia-smi
14:54karolherbst: maybe I get stuff changed with it
14:54karolherbst: like performance mode
14:55imirkin_: mupuf: what's plugged in? still fermi?
14:55mupuf: imirkin_: yes, hakzsam is still working on both the nvc0 and c1
14:55imirkin_: ah ok
14:56mupuf: too bad the i7 died, it would have made a very good second reator!
14:56imirkin_: heh, talk about a weird gpu matchup... gf100 and gf108
14:56karolherbst: nvidia-smi reports my Performance State
14:57glennk: imirkin_, sounds like a chipset gpu vs discrete, except both are discrete :-)
14:57RSpliet: well, I wouldn't consider the gf100 to be discrete
14:59RSpliet: obtrusive rather :-P
14:59karolherbst: did anybody toyed aroung with NVML?
15:00mupuf: oh my flying spaghetti monster, reator just stopped and it feels goooooood
15:00karolherbst: current vram usage and performance state is like the most I get out of nvidia-smi
15:00mupuf: the nvc1 is just like a jet engine
15:00hakzsam: mupuf, reator just crashed :D
15:01mupuf: wait a sec! I had completely forgotten about this fan issue!
15:01hakzsam: I'll work at least one hour before leaving the place :)
15:01mupuf: I should really talk to nvidia about this!
15:01mupuf: well, going to bed
15:01mupuf: so, sorry, no other gpu will be plugged for the night
15:02karolherbst: does anybody know a nvidia-smi hack to get it work on like all cards?
15:02mupuf: karolherbst: how about you use the ioctls?
15:02mupuf: nvidia-settings is open source
15:03karolherbst: good idea!
15:03mupuf: so you can see which calls you need to make to read the state back
15:03karolherbst: I just want to try changing performance state without starting X
15:03mupuf: you won't be able to
15:03karolherbst: why not?
15:03mupuf: the nvidia driver does not do anything without an x server iirc
15:04imirkin_: mupuf: you mean gf100 is a jet engine?
15:04mupuf: it may change soon though, since it will become a KMS driver
15:04imirkin_: the gf108 should be relatively quiet unless it has a broken fan
15:04mupuf: imirkin_: no, I meant the gf108
15:04mupuf: not broken, but the weird pwm controller
15:04imirkin_: that one :)
15:04mupuf: I had *completely* forgotten about this!
15:05karolherbst: mupuf: I can read memory usage without having X running
15:05karolherbst: and temperature
15:05mupuf: well, you will see
15:05mupuf: but I am pretty sure you won't be able to change the perf level
15:05mupuf: oh, and you may be limited to forcing it to max
15:06mupuf: and then back to adaptative
15:06karolherbst: I think I can only set to max perf or adaptive
15:06karolherbst: but changing max clock is possible
15:06mupuf: maybe you will find more methods soon :)
15:06karolherbst: max perf => increase max clock by 135MHz
15:06karolherbst: should trigger a reclock
15:07karolherbst: I can reclock my memory with +4008 MHz :O
15:08karolherbst: that's a bit drastic
15:09glennk: gddr5 goes up to around 7Ghz afaik
15:10karolherbst: yeah, and this would be 8GHz for me
15:10karolherbst: but my core clock won't go above 705MHz
15:11karolherbst: power source: battery which isn't true
15:11glennk: wonder how high one has to push the memory before it melts itself off the pcb
15:13mupuf: glennk: it won't do that
15:13mupuf: well, actually, it is possible for the memory
15:13mupuf: not for the core
15:14glennk: i meant the memory chips, the core has sensors and will shut itself off
15:33nfk: [23:52] <RSpliet> got a laser cutter? // most cases are made of so shit so soft a dremel with appropriate tip will route it just fine
15:33nfk: just be careful with getting it stuck, i broke my router bit last week that way
15:40nfk: mupuf, are you sure? assuming they're using solder and not hot glue it could very well withstand something like 400 to 500°C with ease provided there was nothing laminated that would tear itself apart under mechanical stress like a CPU would
15:41nfk: in fact, i think the silicon would hits its limit before solder if you jammed that heat through it
16:15karolherbst: its awesome, that stuff like glxgears is just pcie bandwith limited
16:16karolherbst: on 8.0 GT/s I get more then doubles framerates compared with 2.5GT/s with glxgears and glxspheres and blob driver
16:23glennk: glxgears is something like 3 GL api calls per frame
16:23karolherbst: yeah, still
16:23karolherbst: remeber these phoronix benchmarks where glmark is just bad on nouveau?
16:24glennk: the triangles one?
16:24karolherbst: yeah for example
16:24glennk: thats just a measure of draw call overhead basically
16:24karolherbst: it also get a massive speedup with the pcie relink
16:24karolherbst: I think I got more like 4 times the speed compared to stock
16:24karolherbst: or points
16:25karolherbst: ahh gputest is the name, not glmark
16:26karolherbst: glennk: but if a higher pcie link state boosts points that much in these test, how can they test driver overhead?
16:28glennk: driver overhead on mesa, blob has those calls optimized to death
16:29karolherbst: I mean with mesa now