Creation of checkpoints taking over my CPU!

Moderators: Site Moderators, FAHC Science Team

arisu
Posts: 466
Joined: Mon Feb 24, 2025 11:11 pm

Re: Creation of checkpoints taking over my CPU!

Post by arisu »

Realtime priority reduces latency, it does not increase throughput. FAH is not super latency sensitive. Unless you have competing processes that are trying to use up all your CPU at all times, you'll spend about the same number of cycles with high priority compared to running at idle priority. The realtime scheduler determines when the process gets cycles relative to when it has more work to do, not how many cycles it gets.

That is why audio and video programs often run with a high or realtime priority. They don't need a lot of cycles but when they do need them, they can't afford to wait because it needs to decode the next video frame or the next batch of audio samples in time for it to be played back, else it stutters.

I've tested it myself on Linux. When I GPU fold on SCHED_FIFO with kernel.sched_rt_runtime_us=1000000 and then do day to day tasks I get about the same TPF as when I fold on SCHED_IDLE. And those are polar opposites. SCHED_FIFO is such high priority that any other task on that core will not progress unless FAH voluntarily yields. And SCHED_IDLE is such a low priority that the folding process will only be scheduled if the kernel would have otherwise scheduled the idle task (which does nothing but put the CPU in a low-power mode and sleep until it gets preempted).
Post Reply