One of the worst side effects of writing performance analyzing code is realizing just how much your OS moves your process around between cores. :-(
Näytä tämä ketju
-
I get it. As computers get more parallel sometimes you pay perf overhead (moving cores) for lower latency (earlier wake up) but it still feel wonky to see.
Näytä tämä ketju
But it doesn't have to be this way! In my homebrew engine, I only spawn as many threads as there are cores, and the entire game is just a job queue. By settings thread affinity masks, I have complete control over perf overhead.
-
-
Moreover, I can dynamically increase or decrease the # of active cores, so that means the game can load balance by hibernating cores instead of sleeping a short time until there's more work. Main menu sits at 2 cores or so, while in the game uses everything you got.
Kiitos. Käytämme tätä aikajanasi parantamiseen. KumoaKumoa
-
-
-
Well, yeah. I bring this up from my previous life in the HFT world, where we reserved processors before boot so the kernel couldn’t schedule on them at all. I was just whining.
-
Here is a screenshot where moments prior I had changed the core count from 3 to 2, and it's reflected in the task manager. I've never really seen what HFT tech looks like. Any interesting articles about it?pic.twitter.com/79GqPRwQNi
Keskustelun loppu
Uusi keskustelu -
Lataaminen näyttää kestävän hetken.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.