00:31theodrim: Hello, is it possible to disable some nouveau message spam in dmesg/syslog (related to vdpau, I'm guessing)?
00:33imirkin: what do the lines look like?
00:34theodrim: nouveau 0000:01:00.0: DRM: skipped size 0
00:34imirkin: that's really bad
00:34imirkin: why is something trying to allocate a zero-sized buffer... ugh
00:35theodrim: cause is running mpv with --vo vdpau, no issues on performance side or anywhere else.
00:36imirkin: hm =/
00:36imirkin: anyways, find the relevant log message and comment it out
00:36theodrim: find it in sources you mean? welp, was hoping to not build anything from source on this box :)
00:38imirkin: depending on how it's being printed you could adjust log levels
00:38gnarface: theodrim: got disk space but don't wanna make a mess? debootstrap a build environment into a chroot, or use qemu-kvm
00:39gnarface: then you can just delete the directory/disk image after you're done
00:40theodrim: gnarface, yeah, sure, but I've came to ask, maybe there is some settings to disable logging or something simmilar. imirkin, adjust via kernel boot options or something else? I didn't find any relevant ones (other than debug, but i don't see empty one there).
00:42imirkin: dpeending on the message... i think that one might be harder
01:12gnarface: theodrim: (don't look at me like i might know the answer)
01:12gnarface: debootstrap i might be able to help with
01:12gnarface: driver patching... not so much
11:25anEpiov: any improvements in 4.13, except for the lack of opencl
12:38pmoreau: anEpiov: “Nouveau for open-source NVIDIA has HDMI 3D / stereoscopic support” according to Phoronix. It’s possible it also contains some bug fixes, though I can’t remember which ones off the top of my head.
12:45karolherbst: pmoreau: by the way, do you think you will find some time you take a look at some of my patches? Really want to mainline most of it before starting new things now
12:46pmoreau: Yes, definitely going to take some time to do that during the week, both for reviewing and testing them.
12:46karolherbst: :) nice
12:47pmoreau: Would be awesome to get as many as possible mainlined. I am getting a bit lost between the ones I have reviewed and the ones I haven’t, as you repost them from time to time (as they have been sitting on the ML ignored), along with other patches. :-)
12:48karolherbst: but usually I add all the r-by tags
12:48karolherbst: thing is they mostly depend on my clk subdev rework and I didn't want to port them over to the current code
12:49karolherbst: and most can't be ported anyhow
12:49pmoreau: Right, but sometimes I have reviewed them without sending the email for it, as I was waiting to review the whole series, but got stuck on something and forgot to finish reviewing the series. xD
12:49karolherbst: pro tipp: just send one mail after you are done with one patch
12:50pmoreau: Yeah, I should do that.
12:51karolherbst: I also need to go through all those and work on the things I notice there, but at least I am done with what I wanted to land next feature wise
14:29imirkin: anEpiov: kernel 4.13 (and for many many kernels prior) have everything needed for opencl to be supported (at least prior to some of the HSA measures). you won't find support in a kernel update.
14:33anEpiov: imirkin: in an nvidia card?
14:33imirkin: sure. there's just not much kernel support required for opencl on top of what you'd need for opengl
14:33imirkin: (in fact, probably less than what you'd need for opengl...)
14:35imirkin: that said, i doubt you'll see the user-space end of opencl supported in the next 12 months.
14:35imirkin: so asking every week about it is, at best, annoying.
14:35anEpiov: 12 months!!
14:36karolherbst: next 12 months sounds like a good estimation, maybe faster. Depending on what will happen the next few months
14:36imirkin: no one's working on it
14:36anEpiov: ok, knowing that the release time is 12 months there's no need to find out sooner.
14:36karolherbst: imirkin: right, currently not
14:37imirkin: some parts of some of it are being addressed, but there are lots and lots and lots of bits needed to have a fully workign opencl 1.2 impl, which is kinda the min most people need
14:37karolherbst: not saying it would be a small thing to do when somebody starts to work on it
14:37imirkin: i'd guesstimate 4-6 months of full-time effort from a developer who didn't need to learn all about nvidia to start with.
14:37karolherbst: that's optimistic
14:38imirkin: i'm an optimist :)
14:38karolherbst: well, we'll see what the future brings
14:38imirkin: a lot can be done in 6 months, by the right person
14:39karolherbst: yeah, that's true though
14:39imirkin: but ... no one's working on it in such a manner, and there's no prospect that anyone will.
14:39xvilka_: since we are speaking about optimistic things - asking again today - how I can help with G107 firmware loading problem?
14:40imirkin: [not to belittle pmoreau's efforts on getting a spirv -> nvir converter, but that's hardly all of it, he's still pretty far away from it, and working at maybe 1/20th time, rather than full time]
14:40karolherbst: imirkin: yeah, would be nice if somebody could support him on this
14:40imirkin: xvilka_: unfortunately that's a tough one.
14:41pmoreau: karolherbst: wink wink wink
14:41imirkin: xvilka_: did you see my comments yesterday?
14:41anEpiov: from a not-technical person I think shouldn't be the hard, intel/amd opencl laid out, just mapped whatever functions are similar to nvidia hardware.
14:41karolherbst: pmoreau: :D
14:41xvilka_: imirkin: oh, probably not, let me search in logs
14:41anEpiov: no need to create algorithms from ground up
14:41imirkin: anEpiov: technical people always have a hard time coming up with time estimates
14:42imirkin: because there are always unexpected things that come up, which defeat even the best laid plans
14:42pmoreau: imirkin: For 1.2, definitely. For 1.0, not entirely sure, though it will probably takes some before I can use upstream LLVM to generate SPIR-V. :-/
14:42imirkin: non-technical people have no hope of being able to estimate the level of difficulty, as they have no idea what's even involved in the first place.
14:42imirkin: pmoreau: do you properly handle loops?
14:43pmoreau: I should make a map of OpenCL 1.0 support, rather than just a SPIR-V support one. That could help.
14:43karolherbst: pmoreau: I guess focusing on 1.0 would be the best bet for now.
14:43karolherbst: pmoreau: mind creating trello cards for this?
14:43pmoreau: imirkin: I do, but the RA does not. ;-) Still need to look into that.
14:43xvilka_: so from all you said we have to draw at least 2 conclusions: 1) there is no hope 2) future is not bright
14:43imirkin: pmoreau: and non-structured control flow?
14:43imirkin: xvilka_: yeah, it's not great.
14:43pmoreau: Not entirely sure about non-structured control flow.
14:44imirkin: dunno if CL 1.0 had it, but i definitely know it's a thing in CL
14:44pmoreau: karolherbst: Trello cards about? I have some here https://trello.com/b/YyQi5Wy2/nouveau
14:44pmoreau: I would guess it was already there in 1.0
14:44karolherbst: pmoreau: ohh right
14:45imirkin: anEpiov: for comparison, intel and amd both have separate teams that do their opencl impls. those teams involve multiple full-time employees.
14:45imirkin: anEpiov: none of the work those teams have done is in any way reusable for nouveau
14:45anEpiov: how's that possible?
14:45anEpiov: opencl uses hardware.
14:45imirkin: you can just take my word for it.
14:46karolherbst: anEpiov: none of those are based on mesa or anything which could be shared across vendors... sadly
14:46karolherbst: it's a bit stupid, but that's how it is
14:46imirkin: there's an open-source frontend for opencl in mesa, called 'clover', but it's far from feature complete, and does not "connect" to nouveau
14:47imirkin: it's largely been abandoned
14:47karolherbst: imirkin: is there any big technical reason for this by the way?
14:47imirkin: for what?
14:47karolherbst: or was it rather like "meh, don't want to use clover"
14:48imirkin: for not using clover? lots of functionality would have to be built, esp for OpenCL 2.0 support which is something AMD has championned
14:48karolherbst: well, that AMD didn't focus on clover? Or was it more like due to the fact they already had something they would rather open source than putting effort into clover
14:48imirkin: they had a perfectly good existing driver that they could just reuse.
14:49karolherbst: well, they also had an OpenGL driver they could have reused
14:49imirkin: and they still are
14:49karolherbst: well, but they also put efforts into making a proper mesa based driver
14:50imirkin: not really. they hired 2 guys.
14:50karolherbst: well, that's more than 0
14:50karolherbst: but I thought there are a bit more than just 2?
14:50imirkin: how many people do you think worked on ROCm?
14:50karolherbst: too many
14:50imirkin: probably 20+
14:56karolherbst: most likely, yes
15:04imirkin: xvilka_: you have the issue where the GR unit in your GP107 is not coming up right?
15:07xvilka_: imirkin: yes
15:09xvilka_: imirkin: as I mentioned I have some RE, C and assembly experience, no knowledge of GPU though
15:09xvilka_: so hope I can be helpful
15:09imirkin: xvilka_: ok, well grab a copy of envytools
15:12xvilka_: imirkin: moreover, I have an access to exactly the same laptop with Windows, if this of any help
15:17xvilka_: imirkin: ok, will try to play into the Record&Replay game
16:17anEpiov: karolherbst: holy!! you're famous!
16:17anEpiov: karolherbst: how can I send you some flowers and a good bottle of red wine?
16:20pmoreau: Good, after all those changes to the SPIR-V linker, I only had two small bugs when trying to run it on my OpenCL samples.
16:21pmoreau: Time to rebase the Mesa work to the latest master.
16:31karolherbst: anEpiov: huh?
16:31anEpiov: karolherbst: phoronix article :D
16:31karolherbst: happens from time to time. I doubt there is one nouveau dev without a phoronix article
16:51NSA: hi there
16:51NSA: i'm using a nvidia gtx750ti
16:51NSA: 3 monitors connected to it
16:51NSA: 2x fullhd, 1x 4k
16:52NSA: the 4k monitor is this model: http://swedx.se/product_info.php?products_id=993&language=en
16:52NSA: it's connected via hdmi
16:53NSA: i'm using wayland with sway right now as WM
16:53NSA: the 4k monitor is forced to use edid 2.0 to be able to do 4k@60hz
16:53NSA: its native resolution would be 3840x2160@60hz
16:53NSA: right now it's running at 4096x2160@24hz
16:53imirkin: we clamp HDMI to 297MHz right now
16:54imirkin: we don't have HDMI 2.0 support
16:54imirkin: [which is needed to go to get above 340MHz TMDS]
16:54NSA: i have no idea what those MHz numbers mean
16:55imirkin: iirc the GTX 750 Ti does not support HDMI 2.0, so the only way you'll get 4k @60Hz is by using DisplayPort
16:55imirkin: the first number in the modeline is the pixel clock in mhz.
16:55imirkin: a reduced-blanking 3840x2160@60Hz mode requires a 533MHz pixel clock
16:56imirkin: HDMI 1.4 only allows up to 340MHz, and most actual implementations restrict that further to about 300MHz (which is what's needed for 2560x1440 @60hz)
16:57imirkin: er no - i lied about that - 2560x1440 only needs 241.5MHz
16:57karolherbst: imirkin: how much is missing for HDMI 2.0 support?
16:57karolherbst: or rather what exactly
16:57imirkin: karolherbst: someone to figure out what, if anything, we need to do differently
16:58imirkin: and i suspect there's some protocol for avoiding doing bad things if you e.g. have a bad cable. dunno.
16:58karolherbst: I see
16:58imirkin: (or does HDMI 2.0 not need better cables? i have no clue.)
16:58imirkin: i am in possession of an HDMI 2.0 monitor nowadays
16:58NSA: i just remember that i might not have used the same gpu when i used the screen with windows
16:59imirkin: but I don't think i have a HDMI 2.0 GPU
16:59NSA: is there a way to check if the 750ti can do hdmi 2.0?
16:59imirkin: NSA: i believe the GTX 9xx series and above support HDMI 2.0
16:59karolherbst: imirkin: sounds about right
17:00imirkin: NSA: https://devtalk.nvidia.com/default/topic/766346/geforce-750-ti-maxwell-600-mhz-tmds-/?offset=6
17:00NSA: my 750ti only has hdmi/dvi/vga
17:00imirkin: GM10x - no hdmi 2.0. GM20x - hdmi 2.0.
17:00imirkin: really? i thought most came with DP =/
17:03NSA: i must have only tested it on windows with my 1070 then
17:03NSA: that one has 3xDP, HDMI and DVI
17:03imirkin: yeah, the 1070 definitely supports HDMI 2.0
17:04NSA: i guess i'll have to completely rework my desktop then :|
17:04imirkin: i'd recommend getting a board with DP
17:05imirkin: and if your monitor only has HDMI 2.0, you can get a DP -> HDMI 2.0 active adapter
17:05imirkin: that will work well with nouveau, if that's your target.
17:05NSA: i have a few of them lying around
17:05NSA: if i buy a new gpu it won't be nvidia
17:05imirkin: wise call.
17:05NSA: too many problems with linux
17:05NSA: i've been trying to get this stuff running for like 3 days already
17:06NSA: xorg doesn't have per display dpi
17:06imirkin: should have asked here earlier :p
17:06NSA: wayland doesn't play well with nvidia
17:06NSA: it wasn't just the resolution
17:06imirkin: ah yeah
17:06imirkin: xorg barely supports multi-head :)
17:06NSA: i'm having a dual gpu setup
17:06NSA: with one gpu disabled for vm passthough
17:06imirkin: it's all hack-on-top-of-hack
17:07NSA: xorg has fluxbox though
17:07NSA: there doesn't seem to be a similar thing for wayland
17:07imirkin: hrm. GTX 950 still way too expensive to get for something silly like trying to get HDMI 2.0 going.
17:08NSA: i think for now i might change my setup to only use the fullhd screens on linux
17:08NSA: and not the 4k screen
17:08NSA: i had planned to have the 4k screen connected to both gpus to either use it for gaming on windows or extend the linux desktop when i don't play
17:08imirkin: if you do end up using nouveau, come back and ask how to reclock the gpu so it's not mega-slow.
17:09NSA: well i do use nouveau now
17:09NSA: and that's not gonna change for at least a few months
17:09NSA: unless i switch back to xorg
17:09karolherbst: imirkin: we could send you one
17:09NSA: which i might just do if i don't need the per screen dpi
17:10NSA: either way until next weekend i guess i'll stay on nouveau
17:12NSA: imirkin: what do i want/need to know about reclocking?
17:12karolherbst: NSA: can't toolkits handle that properly now? I thought that's why they all added support for dpi
17:13NSA: karolherbst: idk how i'd have to configure them
17:13NSA: appearently there's per-application dpi aswell
17:13NSA: but that won't work if i have an application with windows on multiple screens
17:13karolherbst: that's what GTK/QT5 will most likely do
17:13NSA: especially the browser
17:13karolherbst: maybe somebody should work on making this nice and smooth
17:14NSA: wayland has per screen dpi
17:14karolherbst: what happens if an application ends up on both?
17:15NSA: i think whichever screen has the center of the application
17:15karolherbst: so, it's broken currently
17:15NSA: at least that's how it works on windows
17:15NSA: i'd guess it's similar on wayland
17:15karolherbst: then windows is broken as well
17:16karolherbst: I know that on mac os x this is well implemented
17:16NSA: it's far better than not having per screen dpi though
17:16NSA: haven't used osx in years and last time i used it i had no highdpi hardware
17:16karolherbst: well I don't know about if it's "well implemented", but at least there it matters where it is displayed
17:16karolherbst: not where the center is
17:17karolherbst: I have one from work
17:17NSA: it's probably less about center and more about % of the window
17:17NSA: like whichever screen holds the majority of the window
17:17NSA: i guess
17:17karolherbst: still broken if you ask me
17:18NSA: i don't really mind that though
17:18NSA: i rarely have windows spanning over multiple screens
17:18NSA: if i have i don't care about the part that's "off-screen"
17:18NSA: i mostly see it when moving windows between screens
17:19karolherbst: and while moving it should resize automatically, no idea if it's done already
17:19NSA: yeah that works
17:19NSA: on windows
17:19karolherbst: at least
17:19NSA: can't talk about linux since i haven't tested that yet
17:20NSA: wayland doesn't have fractions for dpi scaling
17:21karolherbst: with mac os x they did the smart step to only have floating point pixels
17:21NSA: they say the compositor sendind ceil(scale) to the client and downscaling to the fraction
17:21karolherbst: or rather to introduce the concept of floating point pixels
17:21NSA: on windows i had the screen configured for 150% input scaling
17:21NSA: which is really nice on a 28" 4k screen
17:22NSA: with the WMs i tested so far i couldn't figure out whether or how i'd configure that stuff though
17:22NSA: gnome supposedly has per-display dpi but a config for that is nowhere to be found
17:23NSA: sway seems to just let me pass the scale directly to the wayland scaling stuff
17:23imirkin: NSA: with the GTX 750 Ti, grab kernel 4.12 or later. you can then change perf levels in /sys/kernel/debug/dri/0/pstate.
17:23imirkin: karolherbst: 'we'?
17:23NSA: got 4.12.13
17:24imirkin: karolherbst: i don't really want individuals paying to send me stuff - if you can get a company or other deep pocket to sponsor it - fine.
17:24NSA: imirkin: currently set to http://ix.io/zZC
17:24imirkin: so ... 2 perf levels
17:24imirkin: you can echo 0f in there to move it up, or 07 to move it back down
17:25NSA: thanks, works a lot better now
17:25NSA: mpv can play a 1080p video without stuttering again
17:25karolherbst: imirkin: we as in somebody with a lot og nvidia gpus
17:26NSA: imirkin: is there a way to set that permanently or do i have to create some post-boot script for that?
17:26karolherbst: imirkin: well if shipping isn't too expensive, I am sure mupuf could send you one over or find a different solution or whatevr
17:26imirkin: NSA: you can boot with nouveau.config=NvClkMode=15
17:27imirkin: which will perform the reclock on boot
17:27imirkin: [yeah, not confusing at all that that param is decimal, while pstate uses hex.]
17:27NSA: that's what i was just about to ask
17:27karolherbst: NSA: you will see the last line not hitting 1280 MHz after setting to 0f
17:28NSA: karolherbst: indeed, it's 1063 MHz
17:28karolherbst: but you could boot with nouveau.config=NvBoost=1 or nouveau.config=NvBoost=2 if you need more perf, or "nouveau.config=NvBoost=2,NvClkMode=15" if you mix both
17:28imirkin: karolherbst: i'd rather not take mupuf's GPUs
17:28NSA: what's wrong with mupuf's GPUs?
17:28karolherbst: NSA: the NvBoost things may put too much heat on the GPU in certain workloads though, that's why we default to 0
17:29NSA: how can i access gpu temps?
17:29karolherbst: and some GPUs tend to be unstable with higher clocks as well
17:29karolherbst: NSA: sensors
17:29karolherbst: imirkin: why not?
17:31NSA: can i change the NvBoost without reboot?
17:31imirkin: coz it's not like i'm poor and can't afford it - i just don't want to afford it. and i hate the idea that he's essentially willing to afford it and i'm not, and thus he's sending me shit which i could very well just get myself.
17:31karolherbst: not yet
17:31karolherbst: imirkin: doesn't make sense to me. buy him 2/3 beers the next time you meet to cover the shipping costs or whatever
17:34NSA: makes sense to me
17:34NSA: imirkin's point
17:35anEpiov: would FMV, Function Multivesioning, be a solution for opencl in different hardware?
17:35karolherbst: NSA: well it makes sense to me in a sense of understanding what he says, but the other part doesn't
17:35anEpiov: that would still need hand coded the hardware part for each card/generation.
17:36NSA: karolherbst: i get his point though
17:37NSA: not sure how to better explain it though
17:37karolherbst: sure, but I wouldn't worry about something like this
17:37NSA: different personality
17:37karolherbst: and maybe we simply need somebody with a brutally high upload, so that we could have a remote machine to do nouveau stuff with a camera recording a screen or so
17:38NSA: does he have brutally high upload?
17:38karolherbst: nobody in NSA has except coorperations
17:38NSA: what do you consider brutally high though?
17:38karolherbst: and maybe people with google fiber
17:38NSA: i'd consider 100mbit brutally high already for german standards
17:39karolherbst: german standards are like shit if it comes to internet speed
17:39NSA: that's my point
17:39karolherbst: only the US is more shitty here
17:40NSA: isn't the problem in US just that they don't have same service everywhere?
17:40karolherbst: the problem is, that ISP can "buy" monopoly regions
17:40NSA: not just that
17:40NSA: also that they can decide whether or not to upgrade locations
17:40karolherbst: ohh, that's pretty much the main reason. Having no competition makes some coorperations super lazy
17:41NSA: you could just replace that with "all"
17:41karolherbst: ohh, I know a hand full where this isn't true
17:41NSA: probably localized ones?
17:41karolherbst: something like that
17:42NSA: my internet connection is 20up
17:42NSA: guess how much downstream i have
17:42NSA: to get that much upstream
17:43NSA: how did you know?
17:43NSA: are you german?
17:44NSA: might need to fix my weechat geoip script
17:44NSA: doesn't lookup ipv6 yet
17:45RSpliet: 1/10 is a pretty common ratio for consumer broadband
17:45karolherbst: RSpliet: I have 1/5
17:45NSA: it's 1/20
17:45karolherbst: RSpliet: I could get 2/5
17:45karolherbst: and it depends on the kind of broadband you get
17:46NSA: if i don't take cable i can only get up to 50mbit down
17:46karolherbst: for fiber it is bs to not give you 1/1
17:46karolherbst: for "cable" you get a crappy split
17:46karolherbst: for DSL the split is better
17:46RSpliet: karolherbst: there's limited spectrum, you can only allocate once, most consumers don't really want a high uplink. Makes perfect sense to me
17:47karolherbst: it makes no difference for fiber
17:47karolherbst: it makes a difference if you do 10GBit up on fiber
17:47karolherbst: but we are far away from that
17:49RSpliet: There's more to fibre than just the last kilometre...
17:50karolherbst: well, you don't guarentee 1/1 at all time if you are smart
17:50karolherbst: but technically it doesn't matter if you limit it to 100Mbit or give the straight 1/1
17:50karolherbst: because it makes no difference whatsoever
17:50karolherbst: sure, if the connection to the core is limited, that's what everybody together is getting at most
17:50RSpliet: It matters at the peering side, where you pay for exchanging that data with other networks
17:50karolherbst: but this is a different story
17:51RSpliet: the simple inconvenient truth is that 99% of consumers that want a higher uplink would use it for exchanging copyrighted materials, which is to be discouraged. If you really really need synchronous speeds, you pay business rates.
17:51karolherbst: pro tipp: shit on the telecom and don't exchange with them
17:51karolherbst: then it's for free
17:52karolherbst: and "peering" is for free
17:52karolherbst: you don't pay to peer
17:53karolherbst: RSpliet: any better arguments than this "exchange copyrighted bs" bs?
17:54karolherbst: sure some would do this
17:54karolherbst: but they would do it without anyhow
17:54karolherbst: and most wouldn't do it more, jsut because they have a faster uplink
17:55NSA: i know several people that transfer several TB of copyright infringements per month without high upload at home
17:55NSA: people don't need high upload at home for piracy, they just rent servers for that
17:55karolherbst: it's inconvenient at most
17:56NSA: not even that bad though
17:56NSA: not much more effort to set up than when setting it up at home
17:56karolherbst: and you could do interesting things with a high upload: online backups
17:56NSA: considering the downstream available these days it's also easy to stream content from servers on demand
17:56RSpliet: karolherbst: well... name the home user use-cases for high uplink
17:57NSA: as he just said, backups
17:57NSA: i have unlimited storage at google drive
17:57NSA: which allows uploading 800GB per day
17:57NSA: which means in theory i could backup my whole system every day
17:57karolherbst: you could properly stream videos as well
17:57NSA: (excluding longterm data storage)
17:58karolherbst: high res and everything
17:58NSA: yeah, like all those twitch / youtube streamers
17:58karolherbst: there is tons of things I could come up with you could do with a high upload
17:58NSA: they could do 4k 60fps live streams
17:58NSA: (assuming they get the right hardware)
17:58RSpliet: karolherbst: yes, use-cases exist. But we're talking about the *average* clientelle?
17:58RSpliet: If you have special needs, you pay special prices
17:59NSA: except there are no consumer offers for "special needs"
17:59NSA: even if i wanted to i couldn't buy that
17:59karolherbst: online backups would be a hell of a average clientelle use case
18:00karolherbst: compuer breaks down and you can just setup a new one without loosing anything?
18:00RSpliet: yes, but most users back-up their photos and a very infrequent video. There's no time pressure
18:00NSA: i'm sure my mother would be happy about weekly backups of her pc
18:00karolherbst: whith todays uploads you couldn't even do a normal backup properly
18:00NSA: without buying a new hdd every week
18:01karolherbst: anyway, it's a shame we don't get a proper internet connection
18:01karolherbst: most countries have
18:01karolherbst: the "most advanced ones" don't
18:02RSpliet: karolherbst: you're overreacting. Most people don't have more than 10-20GiB to back-up (we're not talking about copyrighted materials right?). You only back up your diffs, so after an initial 3Hr upload at 20mbit you can do a weekly sync in under half an hour. And that happens in the background...
18:03karolherbst: no I am not
18:03NSA: right now my mother has 2mbit upload iirc
18:03NSA: for 100gb backup that'd be 111 hours
18:03RSpliet: NSA: okay 2mbit is poor
18:04karolherbst: I know it's possible to do it properly
18:04karolherbst: there is simply no will here to do that
18:04NSA: depending on how you do backups it might not be possible to do incremental backups
18:04NSA: like the way i do backups
18:04NSA: i'm sending encrypted archives to google drive
18:04karolherbst: you can get 1Gbit for 12€ a month in romania and even if you take account of difference in salary, it's brutally cheap
18:05NSA: iirc you can get like 300mbit in both directions in spain for something like €40
18:06karolherbst: and it has 500MBit upload
18:06NSA: i'd get that
18:06NSA: do they deliver to germany?
18:06karolherbst: me as well
18:06karolherbst: no :D
18:06RSpliet: But they're a member of the EU, so you could just move there...
18:06karolherbst: I could
18:06NSA: "just move there"
18:06karolherbst: have other plans
18:07NSA: like going to a different country is that easy
18:07RSpliet: It is within the EU... no more difficult than moving states in the US
18:07RSpliet: Well, okay maybe except for the language barrier
18:07NSA: might be different if you don't have any social contacts and work from home
18:07karolherbst: RSpliet: anyhow, with your reasoning we could have stayed with the same hardware as 10 years ago, because "most poeple" only X
18:08NSA: karolherbst: and now some people also wayland?
18:08karolherbst: what I try to say is, the only reason we have this crappy internet situation is due to the unwillingness of our ISP
18:08karolherbst: no other reason
18:09RSpliet: karolherbst: and the people putting up with it because 99% don't truly care
18:09NSA: they can make more money that way
18:09NSA: that's why
18:09karolherbst: mhh doubtful
18:09karolherbst: because it will reduce their income in long term
18:09NSA: how would it reduce that long term?
18:09karolherbst: RSpliet: not true
18:09NSA: if people don't have alternatives?
18:10karolherbst: I know enough non technical peopel also complain about their shitty internet
18:10karolherbst: most are aware of the situation
18:10RSpliet: karolherbst: then why is there in all of Germany not a single entrepreneur who comes to challenge the market?
18:10karolherbst: RSpliet: you know the answer
18:10karolherbst: it isn't like you can buy you into this with a lot of money. Well it would help, but it wouldn't change much
18:12karolherbst: but we are slowly getting fiber by smaller competitors
18:12karolherbst: just taking some time
18:15RSpliet: (oh and ftr: "we could have stayed with the same hardware as 10 years ago, because "most poeple" only X" is a bs projection on things I've never said. Consumption of internet has changed drastically w/ all social media sharing going on, but down/up ratios have been sufficient to accomodate that. Hell, they do that mostly on 4G speeds)
18:16NSA: now don't say mobile internet would be any better
18:16NSA: limiting the connection to 56kbps after 1gb traffic?
18:16NSA: surely must be because people do copyright infringements otherwise
18:16RSpliet: NSA: you seem to employ the same discussion tactics as karolherbst... that's not what I said.
18:16NSA: with their downloads
18:17karolherbst: RSpliet: okay true, you never said it like that. What I just refered to that only because there is no use case today, doesn't mean there wouldn't be one if the situation would be any different
18:19RSpliet: That's a valid point. Unfortunately, commercial business don't care about "what could have been" unless it could earn them more money. I bet it takes those smaller optic fibre challengers to change their ways, big companies are conservative because they are allowed to
18:22karolherbst: NSA: well, mobile internet is a different story. Capacity is a bigger problem there and costs much much more than having more fiber
18:23NSA: karolherbst: yet other countries can deliver unlimited offers
18:23NSA: it's just retarded that they keep upgrading speed but not capacity
18:23karolherbst: NSA: which one and at which rate?
18:23karolherbst: ohh my capacity increases over time
18:24NSA: i'd much rather be able to have a steady 1mbps mobile line than being able to use all my 3gb in less and less time
18:24karolherbst: I started 2 years ago with 500MB and now I am at 1.25GB with same costs
18:24NSA: i only ever had the amount change when switching contracts
18:24karolherbst: pro tip: don't have a contract, but only prepaid
18:24karolherbst: it's cheaper anyhow
18:25NSA: i have a contract that i can cancel monthly
18:25NSA: €14 per month
18:25NSA: allnet flat
18:25NSA: €25 for 21mbit unlimited
18:26karolherbst: yeah, but it doesn't make any difference if you get a prepaid or contract
18:26karolherbst: prepaid has the advantage you can just move to a new one without any problems
18:26karolherbst: contracts you need to quit and they follow up with annoying calls and mails
18:27karolherbst: and usually "prepaid contracts" upgrade themselves automatically
18:27NSA: i've only taken contracts i can cancel every month for mobile
18:27NSA: which worked fine for me so far
18:27karolherbst: where contracts you need to quit to get an upgrade
18:27karolherbst: or maybe the mobile ISPs changed that by now
18:28NSA: haven't had any problems with that yet
18:28NSA: i mean obviously you have to pay if you want to keep your number
18:28NSA: when changing providers
18:28NSA: but that's no different to prepaid
18:28karolherbst: NSA: what I mean is, if the initial terms are 500MB, but the same "option" provides 1GB 6 months later, I just get it automatically. I doubt it's true for a contract
18:28NSA: guess not
18:31NSA: what i meant though
18:31NSA: 500mb is a joke considering speeds like 300mbit
18:31NSA: via lte
18:32NSA: few gb isn't much better imo
18:32karolherbst: I only have 21mbit though, but currently I don't even use all the 1GB
18:32karolherbst: everywhere I go there is usually wifi I can use
18:33karolherbst: ohh right, 1.25GB for a few days
18:34NSA: i had 16gb traffic on wifi on my phone in the last 30 days
18:35NSA: i guess if i wouldn't need to be careful with how much i use to still keep some highspeed traffic for when i need it i'd easily use 10gb per month at least on lte
18:35karolherbst: well I simply don't do much on my phone
18:36NSA: i just started a new job that is 45 min of DB train fun away
18:36NSA: not much else to do there
18:39karolherbst: fun, I have 21.6/8.6 split for mobile. Thought it is much worse
18:40NSA: i don't even know what speeds i have on mobile
18:40NSA: other than that i'm using lte
18:40karolherbst: well, I also have LTE
18:41karolherbst: but lte can be everything really
18:41NSA: well it's at least 21mbit
19:29imirkin: karolherbst: i have gbit fiber :)
19:29tobijk: imirkin: *triggered* :D
19:29tobijk: don't make us mad ;-)
19:31tobijk: and i should feel comfortable with my 50 mbit :D
19:31NSA: tobijk: that's what the ISPs want you to think
19:32tobijk: magenta T, and its the only isp around
19:32tobijk: others have 6 mbit *yay*
19:32NSA: they want you to think 50mbit is fine so they don't have to pay the upgrades
19:33tobijk: NSA: well after all these years with 6 mbit, i'm happy about the current situation
19:33tobijk: next county the have stll dsl-light (386kbit down) :o
19:33NSA: happy or just less annoyed by it?
19:34tobijk: NSA: well its mostly sufficent for what i want to do, so less annoyed is a bit extreme :>
19:34tobijk: if i want to transfer data, i go to the university i study at (1gbit up/down)
19:35NSA: i'm running a homeserver for some stuff and it wouldn't be that unusual for me to have all my 20mbit up used by that if i didn't ratelimit it
19:35NSA: didn't get around to set up a more advanced network at home for prioritizing certain traffic
19:37tobijk: NSA: well thats not a standard usecase
19:37NSA: i never said it'd be standard
19:37tobijk: most people will not have a local server around
19:37NSA: i know several people that do
19:37NSA: in RL
19:38NSA: main reason for me is getting more hdds attached to the server i'm renting in a datacenter is quite expensive with my provider and i can't get a gpu added for cheap either
19:45imirkin: gbit is uncommon where i live... i got lucky that the building i'm in is lit up
19:47tobijk: imirkin: so having it is of course nice, but if yu are honest: do you really need it? (i have a gbit around in university and barely use it at all)
19:47imirkin: absolutely not
19:47imirkin: but they also dropped prices, so i got it
19:47NSA: i'm curious whether you can download from steam with gbit
19:47imirkin: still expensive, but ... not that expensive.
19:47NSA: as in getting gbit
19:47imirkin: i get gbit speeds downloading from google, that's about it
19:47imirkin: steam was weird
19:47imirkin: it'd be fast, and then pause for a while
19:47imirkin: then fast, then pause
19:47NSA: it's always fun when there's a new patch for some game and a friend of mine with 16mbit needs like 30 min to download
19:48NSA: imirkin: might be disk activity
19:48imirkin: yeah. not sure if i put it on the encrypted partition
19:48NSA: afaik they do diffs for updates
19:48NSA: so you don't download full files
19:48NSA: but you have more processing / disk activity
19:48imirkin: this was full games though
19:49NSA: idk then
19:50tobijk: NSA: i tested it and got ~980mbit steam download speed once
19:50NSA: i think highest speeds i've had so far where i noticed the speed was from google drive and steam
19:50MichaelLong: just recently my village got FTTH, I also did not pick the 1 gbit offer, I chose the 200/200 Mbits option, more than enought for now
19:50tobijk: yet it might not be a standard thing, dunno where the dfn peers really
19:51tobijk: google :>
20:47karolherbst: imirkin: steam does a DOS on your DNS
20:47karolherbst: maybe not that harsh
20:47karolherbst: but whenever I disable my local DNS cache, steam stops downloading as well
20:47karolherbst: there are some reports regarding this
20:48imirkin: could be, dunno
20:48karolherbst: I had the same issue kind of
20:49karolherbst: but sometimes it is worse than on other days
20:49karolherbst: but with those speeds disk activity is a valid reason as well
21:04karolherbst: nvidia not only ioremaps 0x200 big regions, but also 0x2040
21:07imirkin: (44.6 MB/s) - '/usr/portage/distfiles/google-chrome-stable_61.0.3163.91-1_amd64.deb' saved [63101682/63101682]
21:07imirkin: not even gbit. but not bad.
21:08imirkin: doesn't have time for TCP to adjust window sizes, i think
21:11karolherbst: imirkin: you want to install those cuda tools, 1.4G package ;)
21:12karolherbst: or chromium, those files are also like 550MB
21:14karolherbst: and then to rule out the drive being to slow: aggressive file caching
23:03karolherbst: fun, on reator 62c000 isn't touched as well
23:03karolherbst: well, the first write