duration correction factor
20 May 2012 13:28:22 UTC
Topic 196340
(moderation:
What does this number mean? What is the context this factor is used in? What is it correcting? How can a single number adjust GPU and CPU tasks simultaneously? How is this factor affected by the "dangerous" GPU utilisation factor?
My computers startet with 1.0. Now they have 2.37 and 2.13. Are bigger numbers "better" than smaller ones?
Sorry for asking so many questions and thx for answering them.
Gruß von Heinrich
duration correction factor
)
Hi Heinrich,
first, it does not effect your credit.
The "dangerous" GPU utilisation factor is a helper for a well known fact: openCL-tasks are always cpu/gpu-tasks, which means, when the wu uses the cpu for some time, the gpu is in idle. On fast gpu's the usage may be less than 30%. So it makes sense to run a second one (or even 3 ones) simultanously on that gpu, which results in a gpu load factor near 95%. Dangerous if you overclock and have a poor cooling concept, an underrated power supply aso.
The correction factor is a helper for the BM. When a wu-type is loaded the first time, BM does not know how long it will last to cruch it, so a suggestion is made based on cpu-type, math speed aso. But BM keeps an eye on that and after finishing the wu he calculates a corretion factor for the next wu's. This is necessary to fill the work buffer correct.
HTH.
To answer your 4th question;
)
To answer your 4th question; in general having one DCF for your CPU and GPU fails badly. If you're lucky your CPU and GPU will have a similar DCF and things will work well. IN the typical case they'll be widely divergent and the shared DCF will swing wildly up and down resulting in the boinc client scheduler getting badly confused. If you're running an app on GPU I advise disabling the app on the CPU in the E@H project settings.