Restoring System

[Handruin]

I'm probably seeing a disk I/O bottleneck with my system. I don't have super-speedy drives like you guys. I processed 36 images and as you can see from the CPU graph, I wasn't even pegging both cores all the time. Using separate drives sounds key for better performance.

 

[LunarMist]

Even it was single or poorly threaded, I believe the windows scheduler would divide the time across the CPU's and you won't see one stuck at 100% with others idle...at least not for any great length of time. Maybe the context switching is too high or it simply can't scale for some other unknown reasons? I'm sad to see it doesn't max all four cores to make better use of them.

There seems to be a multi-part conversion cycle. First the RAW file is read (low CPU), then some processing occurs that causes the CPU utilization to increase, but it is not 100%. In the next phase the CPU is at least briefly 100% on all four cores. (David may not be seeing that because taskman displays the 4-core i7 as a bogus 8-core CPU.) Subsequently the output (TIFF) file is written and CPU utilization drops again. Even when the input and output files are on a RAM Disk there is still a sawtooth wave, however.

 

[Handruin]

I was noticing similar activity but I wasn't quite sure what process was happening behind the scenes. If the DPP worker was rewritten to allow user-definable number of spawned processes and some sort of queuing system, it might help a little to maximize the CPU usage a little better. I would guess the upper end of your sawtooth waves are longer on your system because you work with larger RAW images?

edit: Never mind they aren't longer on your system. I just scrolled up and looked at your screen capture. They are longer on mine which might be because I have a slower CPU.