!eWOErHSaiddIbsUNsJ:nixos.org

NixOS CUDA

289 Members
CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda58 Servers

Load older messages


SenderMessageTime
16 Dec 2025
@sporeray:matrix.orgRobbie BuxtonNice work!!!00:39:33
@arilotter:matrix.orgAri Lotterok will PR soon - doing some christmas baking but this is huge lol00:39:39
@arilotter:matrix.orgAri Lottermmaybe tomorrow00:39:42
@arilotter:matrix.orgAri Lotter

this ez tho:

    cudnn = super.cudaPackages.cudnn.overrideAttrs (old: {
      buildInputs = (old.buildInputs or []) ++ [ super.cudaPackages.cuda_nvrtc ];

      patchelfFlagsArray = (old.patchelfFlagsArray or []) ++ [
        "--set-rpath" "${super.lib.getLib super.cudaPackages.cuda_nvrtc}/lib:\$ORIGIN"
      ];
    });
00:39:51
@sporeray:matrix.orgRobbie BuxtonNow I won’t run into this issue in a week 😁00:40:02
@yorik.sar:matrix.orgyorik.sarWell, as for any locks, I expect a slowdown, that’s why I’m asking about less parallel case :)05:08:40
@yorik.sar:matrix.orgyorik.sarHuh, so a significant portion of that eval is actually parsing stuff, interesting.05:09:24
@yorik.sar:matrix.orgyorik.sarIt could be a good thing to improve that one :)05:10:07
@yorik.sar:matrix.orgyorik.sar connor (burnt/out) (UTC-8) I see in release notes https://github.com/NixOS/nix/pull/14219 that can also improve things. Is it in your base branch? 05:11:40
@connorbaker:matrix.orgconnor (he/him)That’s what the changes do05:13:10
@connorbaker:matrix.orgconnor (he/him)Yeah I pulled master a few days ago05:13:38
@yorik.sar:matrix.orgyorik.sarDo they? It looked like a separate lock implementation for this case.05:14:54
@connorbaker:matrix.orgconnor (he/him)They add functionality to the path lock (among other things)05:16:54
@yorik.sar:matrix.orgyorik.sarHuh, I must’ve missed it. Shouldn’t read code early in the morning on my phone.13:28:27
@arilotter:matrix.orgAri Lotterok wew! did some testing, this seems ti work with only the patchelfFlagsArray and doesn't require the buildInputs https://github.com/NixOS/nixpkgs/pull/47139415:58:51
@connorbaker:matrix.orgconnor (he/him) Kevin Mittman (UTC-7)^^^
Are you able to ask internally when this behavior started and for which architectures it occurs?
16:11:28
@arilotter:matrix.orgAri Lotterfrom my end, i'm not quite sure when it gets triggered / when it was added :/16:12:51
@arilotter:matrix.orgAri Lotterall i know is i can run some code on my local machine but not on an h10016:13:03
@arilotter:matrix.orgAri Lotter* all i know is i can run some code on my local machine with a 4060ti but not on an h10016:13:11
@justbrowsing:matrix.orgKevin Mittman (UTC-7)Will check later, Github seems to require login now16:17:38
18 Dec 2025
@rudyshoes:matrix.orgAtix joined the room.14:58:53
@rudyshoes:matrix.orgAtixhey there! i would like to contribute to this project as i have a nvidia graphic card on my laptop , an RTX 4050 with 6gb VRAM14:59:35
@rudyshoes:matrix.orgAtix changed their display name from Rudy-in to Atix.14:59:59
@sporeray:matrix.orgRobbie Buxton
In reply to @rudyshoes:matrix.org
hey there! i would like to contribute to this project as i have a nvidia graphic card on my laptop , an RTX 4050 with 6gb VRAM
What sort of stuff do you think you’ll want to contribute to!
15:01:52
@rudyshoes:matrix.orgAtix what sort of stuff is there? 15:02:22
@rudyshoes:matrix.orgAtixi can contribute to whatever15:02:40
@sporeray:matrix.orgRobbie BuxtonThere is a ton of stuff! I think I’d start with what I was interested in15:03:40
@sporeray:matrix.orgRobbie Buxton That could be gaming 15:03:48
@sporeray:matrix.orgRobbie Buxton That could be ML 15:03:52
@sporeray:matrix.orgRobbie Buxton It could be physics sims 15:04:03

Show newer messages


Back to Room ListRoom Version: 9