06:21 fltrz: karolherbst: I was wondering: arch package repository states "split packages: [...], opencl-mesa, [...]" cf: https://archlinux.org/packages/extra/x86_64/mesa/
06:22 fltrz: should I install it for nouveau opencl on NVidia GT 740M ? the opencl-mesa package description states "OpenCL support for AMD/ATI Radeon mesa drivers
06:23 fltrz: so I think not, but on the other hand I don't know
06:24 fltrz: also there is some weird discrepancy between lshw and WikiPedia regarding bus width / GPU chip
06:26 fltrz: lshw: description: 3D controller; product: GK107M [GeForce GT 740M]; vendor: NVIDIA Corporation; physical id: 0; bus info: pci@0000:01:00.0; version: a1; width: 64 bits; clock: 33MHz; capabilities: pm msi pciexpress bus_master cap_list rom; configuration: driver=nouveau latency=0; resources: irq:28 memory:f6000000-f6ffffff memory:e0000000-efffffff memory:f0000000-f1ffffff ioport:e000(size=128)
06:26 fltrz: memory:f7000000-f707ffff
06:31 fltrz: the pci bus one up in the tree states width: 32; the 3D controller states GK107M and its width 64 bits according to lshw; but on wikipedia we have either: 1) GK107 128bit bus or 2) GK208 64bit bus; or is there some factor of 2 of the PCI Bus (like double or quad data rate?)
06:34 fltrz: I'm guessing the device driver developers know better than some market enthousiast on wikipedia though
08:22 hell__: fltrz: it could also be that your GT 740M doesn't have the max amount of RAM. AIUI, if there's only half of the RAM chips, the bus width will be halved.
08:25 hell__: for example, see the bottom left quarter of this picture: https://i.imgur.com/4kArhez.png
08:26 hell__: the chip with the whiteish plastic rectangle on it is the Nvidia GPU
08:26 hell__: there are two black rectangles above the GPU (two RAM chips), and there's space for another two RAM chips to the left of the GPU
08:27 hell__: (I imagine the other side of the board has more RAM chips / spaces)
08:56 fltrz: hell__: oh I though the bus width was PCI bus width, its direct to RAM?
08:57 fltrz: i.e. I didn't understand why the lshw reported value of the GPU bus width was twice that of the PCI bus width
09:28 hell__: ah wait, PCI bus width
12:37 karolherbst: fltrz: the problem is, that model names are random and don't really reflect "one" thing. It's more of a market segmentation thing than actually following any hardware details.
12:37 karolherbst: but lshw I think only prints PCI information and not neccassarily what the VRAM is all about
13:28 fltrz: karolherbst: I was just curious, sorry to be asking questions regarding nouveau, could you take a look at the pinned comment at https://aur.archlinux.org/packages/mesa-git and its packagebuild at https://aur.archlinux.org/cgit/aur.git/tree/PKGBUILD?h=mesa-git it states some things about gallium so I'm not sure how compatible it will be with the changes you suggested yesterday. if I screw up will I still be
13:28 fltrz: able to boot to tty (I manually start sway at each boot, so I presume I can recover without booting sway)
13:29 fltrz: should I go for the default option 4, or would you recommend following the preferred option 1?
13:29 fltrz: I'm sorry for being such a scaredycat pita
13:36 phomes_: I thought that it could be a fun weekend project to try something for the nvk driver. Make use of the new common physical device management, or something like that.
13:37 phomes_: I built the driver but it errors out in nouveau_ws_device_new(). The drm version is too small. Do I need a specific kernel branch? I tried with 6.0rc3, but it still fails. The gpu is a rtx 2080.
14:15 mynacol: fltrz: I am running mesa-git on my computer because of !10752 (just merged). I just chose option 4. Updates to stable llvm are rare, but when they happen you have to revert to stable mesa temporarily or ignore some dependencies while updating llvm and then mesa-git.
14:29 karolherbst: fltrz: well.. easy solution is to not use llvm-git
14:30 karolherbst: phomes_: if the drm version is too small, I suspect something really odd is going on
14:31 karolherbst: mhh probably not using nouveau
14:31 karolherbst: but that would be weird
14:32 karolherbst: though we might want to validate on the driver
14:35 phomes_: maybe I am not running it as intended. I build and install it. Then run 'VK_ICD_FILENAMES=/home/phomes/nouveau/install/share/vulkan/icd.d/nouveau_icd.x86_64.json vulkaninfo'
14:39 karolherbst: you need to also set LD_LIBRARY_PATH
14:39 karolherbst: but then I am even more confused how you get this drm error
14:43 karolherbst: anyway, the only reasonable explanation would be, that you are not using nouveau on the kernel side
14:47 powerman2206: Does waifu2x-ncnn-vulkan work for anyone?
14:47 powerman2206: I get a core dump and am not sure if it's Nouveau's fault or something else
14:47 karolherbst: probably nouveaus faul
14:47 karolherbst: t
14:48 karolherbst: vulkan is very much work in progress at this point
14:50 phomes_: karolherbst: thanks. I will set LD_LIBRARY_PATH and try with a mainline kernel again. This computer is on ubuntu. Perhaps their kernel is not configured as neccesary
14:53 karolherbst: phomes_: sure it's using nouveau though? Because there is no way to mess this up
14:56 powerman2206: karolherbst: hm, shit
14:56 karolherbst: but I think we should add a check to see the correct driver is loaded, so the error message is more helpful in the future (as I suspect the nvidia driver is used on your system)
14:58 phomes_: karolherbst: it is on nvidia right now, but in earlier attempts I had nouveau. But not with LD_LIBRARY_PATH set of course. Let me try it out. Thanks for the help