Comment by anukin

Have you observed any decrease in performance at context higher than 200k?

That used to be one of the major issues why clearing or compacting context when the context nears 150 to 170k is usually preferred.