Optimizing app

[Hal Bregg]

Guys,

I was wondering if changing those settings would improve crunching speed.

[davidbam]

I don't know, sorry.

What I have seen though - on gtx1660ti - is that the project will give you cuda31 WU, then cuda60, then cuda70 then nvidia101. I assumed it was testing out your card before giving you the best WU

[Hal Bregg]

It seems like it.

After what you wrote I checked again mine results and noticed that nvidia_101 tasks pay over 6k per WU (approx 44min runtime) on GTX950. Had some cuda31 WU for slightly over 1k for each WU (approx 16min runtime).

[davidbam]

The other bad thing about cuda31 is that a single WU will use all available GPUs (and still pay less)

[Bryan]

In olden times you needed to check which "core" worked best with each GPU/CPU ... you can set them in "preferences" and it used to make a HUGE difference. The auto-detect may be better now.

Look at THIS thread. It tells how to run the benchmark test to determine the best "core" value.

Basically copy the executable to its own folder and then launch it with the --bench option.

[Hal Bregg]

I will look into it, Bryan.