Very interesting reading

[Furlozza]

Bored at the moment, and can't sleep, so popped over to Collatz and looked at how theGnat and GTS250 compared with xtermiGnator and GTX285.

However the thing that amazed me was that whilst both nVidia cards were reporting on cpu usage of between 30 and 40 seconds on jobs that took between 2200 and 3200 secs to complete (run time), the ATI cards that did the same work were reporting cpu time and run time with a difference of about 2 secs on average... albeit that the total run times were <=1000 secs.

So the question then arises: do ATI cards speed through the wus, but basically hog one cpu for that time?

Also, does this apply with milkyway as well?

[kashi]

Yeh I can't sleep either, I'm too hot.

The CPU time reported for Collatz ATI is not correct. The CPU time is actually the GPU time and the Run time is the elapsed (wall clock) time. Or something close to that. It is reported this way because older BOINC versions use the CPU time as a way of working out the resource share and the true CPU time for ATI tasks is extremely low. So if that very low CPU time was used then the BOINC scheduling would get totally mucked up.

If you drill down by clicking on the Task ID and look at the actual results you will see for ATI tasks that the CPU time (at the bottom) will only be a few seconds for a recent card using the default parameters. This is because after Gipsel changed the ATI application a while ago to use a lookup table, just about all the processing is done only on the GPU, except for a second or two at the start and at the end. I use much more CPU time than the default ATI settings because I use the b-1 parameter.

So no, ATI cards on Collatz don't hog the CPU at all, they barely use it.