hello!
Any big jobs on the horizon?

hello!
Any big jobs on the horizon?

Aren't you getting any language model work?
You have nice GPU's
I'm sitting here with a older 1080 and I think I have to delete this project (out of BOINC) until the fall as there is no work for me and I have flatlined for weeks now.

Ok thanks Greg _BE then I change the gpu with the old ones (I save the gpu for the autumn) for now I only work with the cpu.

In this latest run of WUs, it seems that only the Linux app works, the Windows one fails all of them...

In this latest run of WUs, it seems that only the Linux app works, the Windows one fails all of them...

I stand corrected, it seems that 'Turing' completes them on Windows OS too

The apps here have always has minimal driver revision levels, CUDA capabilities. Older gpu families haven't been supported in years. Posts of minimum requirements are usually announced in the thread announcing any new app or batch release.

The new ATM tasks needs to have their work generator template limits revised for Windows hosts who are all experiencing exceeded time limits errors on computing the tasks.
This is what the current tasks are showing for limits.

Task size: 1000000000.0000 GFLOPs (rsc_fpops_est)
App speed: 10000000000.0000 GFLOPs/sec (rsc_fpops_bound)

I am having no such issues on my Titan V cards on Linux which are theoretically much slower than 3000 and 4000 series cards.

[EDIT] Now believe the issue is in fact the card processing speed of 3000/4000 series cards are being incorrectly calculated. The app thinks they should be processing faster than reality.

The app is getting calculated speed of Turing class (2000) cards correct on Windows hosts.

Thanks GLIS. I did 2 fails and was starting a 3rd when I saw your post yesterday.

Glad to help, Keith.
By the way, I have a secondary GTX 1660Ti (Turing) card, so I can confirm that it works with Windows.

Steve, the admin here posted on their Discord server that he 10X'd the rsc_fpops_bound values for the new tasks.

He is seeing 4080's succesfully process the work now without getting 'time exceeded' errors.

May take a while for new tasks to outnmber older resends in circulation.

Steve, the admin here posted on their Discord server that he 10X'd the rsc_fpops_bound values for the new tasks.

He is seeing 4080's succesfully process the work now without getting 'time exceeded' errors.

May take a while for new tasks to outnmber older resends in circulation.

Thanks for the info, it actually doesn't give that error anymore.

Does checkpoints on ATM: Free Energy not work?

Installed newest GPU driver and had a Windows update so I restarted Windows (after shutting down BOINC properly) and WUs started from 0.

I don't think it works.

No, checkpoints do work. They restart from zero, but then after 2-3 minutes replay their progress to the last checkpoint.