| 9 Jul 2025 |
 Charlotte 🦝 (it/its) | In reply to @aloisw:julia0815.de
The memory limit is only checked after each evaluation, so it can definitely happen. ah | 11:14:10 |
| Charlotte 🦝 (it/its) | so reducing it to 1GiB or less might work better? | 11:14:37 |
 aloisw | It's not going to help much if the numbers from your post are accurate. You need to reduce the number of workers or the memory usage per evaluation. | 15:17:30 |
 raitobezarius | Grimmauld (any/all) i pulled hyperfine and am currently running the benchmark you said | 16:48:13 |
| raitobezarius | and i don't think i'm reproducing your results at all | 16:48:21 |
| raitobezarius | nixpkgs on master
❯ hyperfine --warmup 2 "./2_91/bin/nix-eval-jobs --workers 6 /home/raito/dev/github.com/NixOS/nixpkgs --system x86_64-linux"
Benchmark 1: ./2_91/bin/nix-eval-jobs --workers 6 /home/raito/dev/github.com/NixOS/nixpkgs --system x86_64-linux
Time (mean ± σ): 110.343 s ± 3.017 s [User: 354.990 s, System: 111.729 s]
Range (min … max): 103.929 s … 114.064 s 10 runs
nixpkgs on master took 25m43s
❯ hyperfine --warmup 2 "./2_92/bin/nix-eval-jobs --workers 6 /home/raito/dev/github.com/NixOS/nixpkgs --system x86_64-linux"
Benchmark 1: ./2_92/bin/nix-eval-jobs --workers 6 /home/raito/dev/github.com/NixOS/nixpkgs --system x86_64-linux
Time (mean ± σ): 109.001 s ± 3.871 s [User: 348.020 s, System: 109.610 s]
Range (min … max): 103.484 s … 117.563 s 10 runs
| 17:00:48 |
 K900 | What the fuck hardware are you running this on holy shit | 17:01:45 |
| raitobezarius | the same as you lol | 17:01:53 |
| K900 | 2 minute nej on all of nixpkgs??? | 17:01:54 |
| raitobezarius | this is my Ryzen | 17:01:56 |
| raitobezarius | on all of nixpkgs on x86_64-linux | 17:02:04 |
| K900 | Well yes | 17:02:09 |
| raitobezarius | i always had 2 minutes nej all the time | 17:02:10 |
| K900 | But like | 17:02:11 |
| K900 | Huh | 17:02:13 |
| K900 | Really? | 17:02:16 |
| raitobezarius | (on the EPYC 7763, it goes even faster) | 17:02:19 |
| raitobezarius | yep! | 17:02:20 |
| K900 | OK so either my closure thing is dog slow | 17:02:29 |
| K900 | Or | 17:02:30 |
| raitobezarius | nej is absurdly fast on Zen ≥ 3 platforms with a lot of cores | 17:02:31 |
| K900 | WAIT. | 17:02:31 |
| K900 | @Grimmauld (any/all) are you testing on the closure thing | 17:02:40 |
| raitobezarius | i mean if your entrypoint is something different than nixpkgs, then yes :D | 17:02:52 |
| raitobezarius | i'm not evaluating the same thing | 17:02:55 |
| K900 | No not that | 17:03:01 |
| K900 | The closure thing parses nej output | 17:03:08 |
| raitobezarius | ye | 17:03:16 |
| raitobezarius | he's indeed testing the rust software called closure-thingy | 17:03:26 |
| K900 | That's my thing | 17:03:34 |
