Page 1 of 1
is proj 18432 single or double precision?
Posted: Tue Dec 21, 2021 11:08 pm
by Knish
I tried overclocking a 750ti that's been folding great as stock for over a year. I know FAHBench is outdated, but tests with double precision seemed ok yet single precision would fail even with no overclock. After a few good units were completed with the overclock, I just got a crash with 18432 so was just curious if single/double has any relevance? (the same WU started after reboot, so in about 3 hrs I'll hit the same point again without overclock this time so I still get to turn this wu in)
Re: is proj 18432 single or double precision?
Posted: Tue Dec 21, 2021 11:54 pm
by JimboPalmer
The vast majority of floating point math in F@H is single precision, as I understand it some of the checkpoint math is double precision.
Re: is proj 18432 single or double precision?
Posted: Wed Dec 22, 2021 10:10 am
by toTOW
Double precision performance is always limited by vendor, so it doesn't push the GPU very hard. It's not a good way to check stability.
All current FAH projects use mixed precision : the majority of computation is done in single precision and some specific operations are done using double precision.