You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
opencode uses ~30-50% CPU per process and slows LLM responses to a crawl #679
Again, only happens on my Ubuntu machine. Claude Sonnet 4 with API credentials, didn't use to happen as late as 0.1.172 and definitely was a problem by 0.1.176. I can pinpoint it down further in a few hours if necessary.
Immediately on the compose screen it ramps to around 20-30% CPU usage per process, but I've also seen it as high as 70-90% PP. LLM responses slow down with increased CPU usage too. That part could be throttling on the OS side, but I don't see a reason this machine should have much of a problem running opencode.
changed the title [-]opencode uses ~30-50% cpu per and slows llm responses to a crawl[/-][+]opencode uses ~30-50% CPU per process and slows LLM responses to a crawl[/+]on Jul 4, 2025
Looking at it more it's a bit more interesting - it seems like it "idles" at around 25-35%, and after a few seconds (10 the first time, ~15 the second) it ramps up to around 90-100 per process. It then stays there for a few seconds, before eventually dropping back down to idle in around 2-4s.
Timings don't seem consistent, and I only "time"-d 2 cycles (but I witnessed it ~4 times back to back so it probably is a loop of some sort).
Activity
[-]opencode uses ~30-50% cpu per and slows llm responses to a crawl[/-][+]opencode uses ~30-50% CPU per process and slows LLM responses to a crawl[/+]thdxr commentedon Jul 4, 2025
can you check if thi sis happening in in 0.1.190?
bczsalba commentedon Jul 4, 2025
Yep, seems roughly the same:
(~2s after startup)

Looking at it more it's a bit more interesting - it seems like it "idles" at around 25-35%, and after a few seconds (10 the first time, ~15 the second) it ramps up to around 90-100 per process. It then stays there for a few seconds, before eventually dropping back down to idle in around 2-4s.
Timings don't seem consistent, and I only "time"-d 2 cycles (but I witnessed it ~4 times back to back so it probably is a loop of some sort).
Let me know what else I can do to help!
bczsalba commentedon Jul 8, 2025
@thdxr seems fixed as of 0.2.5, no clue why (again, lmao). cheers!
bczsalba commentedon Jul 8, 2025
Actually nevermind, still broken. Not sure why it wasn't that one time.
bczsalba commentedon Jul 15, 2025
okay it seems fully gone now, i'll make a new issue if it ever comes back since it should be unrelated by then. cheers!
feat: implement --with/--without selective install flags (anomalyco#679)
fix(iris): Leads returned 400 on every board that had leads