Post AdKclSpFjD7IijEWZc by lanodan@queer.hacktivis.me
(DIR) More posts by lanodan@queer.hacktivis.me
(DIR) Post #AdKcT9k1wbXgJr3pM8 by lanodan@queer.hacktivis.me
2023-12-30T09:58:53.298280Z
0 likes, 0 repeats
@niconiconi Makes me wonder how well it is on Erlang side of things.
(DIR) Post #AdKclSpFjD7IijEWZc by lanodan@queer.hacktivis.me
2023-12-30T10:02:46.546121Z
0 likes, 0 repeats
@niconiconi Seems like it supports it including configuration possibilities.https://erlang.org/pipermail/erlang-questions/2017-January/091309.htmlAlso "hell-banshee cooling fans" is an absolute description of DC fans.
(DIR) Post #AdKgcUhBo2S3jLRLXM by lanodan@queer.hacktivis.me
2023-12-30T10:45:50.809461Z
0 likes, 0 repeats
@niconiconi I'm not entirely sure… like implementation-wise do you end up with one-worker-per-compute kind of scheduling to avoid task-switching overhead or do you end up with a humongous number of tasks (something which is very Erlang) that a scheduler then sorts out, possibly with a pool?
(DIR) Post #AdKrKnZpxwAIZblXlY by lawik@fosstodon.org
2023-12-30T12:40:00Z
1 likes, 0 repeats
@lanodan @niconiconi so Erlang is good at doing parallel work but if the work is number crunching you get into a more nuanced situation. If you want to call C-level code you can do a NIF and get much of the scheduling fanciness. There is some overhead if you need to pass things across that boundary. So depends on the workload.Elixir/Nx has tackled GPU from Erlang but uses XLA so you don't get the scheduling of Erlang.Also more NIF variants exist with various tradeoffs.
(DIR) Post #AdKrKorFChgQXsuxXM by lawik@fosstodon.org
2023-12-30T12:41:57Z
0 likes, 0 repeats
@lanodan @niconiconi I believe an old school (non-dirty) NIF gets all the upside of pre-empting, work stealing and so on.https://www.erlang.org/doc/tutorial/nif.html