Folding@home News 4th - 11th August

thanks for the news

thanks for the news! the 290x and 7870 did good

Thanks for the news. I've had some major results this week from removing the automatically upgraded 365 NVIDIA drivers for my Linux 970 and going down to 364.19. This made an immediate difference but then I used Coolbits to up the 970 to 1400 Boost. Going to be adding a Kraken G10 to the 970 tomorrow with a H100 to get the temps down.

How did you manage a 2m bump in production Biffa?

he upgraded to the new NVidia gpus

look how much a 64 core system costs and the PPD it use to do, and now a single gpu can do that and more using 1/5th of the power, standford screwed everybody who ran server set ups.

MikeTimbers said:

Thanks for the news. I've had some major results this week from removing the automatically upgraded 365 NVIDIA drivers for my Linux 970 and going down to 364.19. This made an immediate difference but then I used Coolbits to up the 970 to 1400 Boost. Going to be adding a Kraken G10 to the 970 tomorrow with a H100 to get the temps down.

Click to expand...

Good catch! automatic updates and folding always a recipe for disaster I find

colinmcr said:

How did you manage a 2m bump in production Biffa?

Click to expand...

oz is right, switched 970's for 1070's and my watercooled 980 for a watercooled 1080

ozaudio said:

he upgraded to the new NVidia gpus

look how much a 64 core system costs and the PPD it use to do, and now a single gpu can do that and more using 1/5th of the power, standford screwed everybody who ran server set ups.

Click to expand...

Not really, I used to get 1million PPD out of a single quad Xeon box with 64cores on the bigadv WU's. At the time I had two of those and a quad Opteron box for roughly between 2 and 2.7 million PPD depending on which project you got.

At peak I had those three and the quad Titan box, theoretical performance would have been 3.5 million maybe if you got all the perfect projects at the same time. Reality was I recorded 3.12 million at the time.



A 1080 gets about the same as a 64core Opteron box did. But the Xeons blew the Opterons out of the water. The cost of the motherboard was the main thing, the cpu's weren't too bad as they were mainly ES versions from the bay. But it was the fun and challenge of building the rig, and the contributing to the team and the mainly the feeling you were helping with the science that was the thing for me. Still the same really

I haven't sat down and compared the costs but I don't think its that much difference if you take overall costs in total vs output.

Yes the switch from bigadv CPU Wu's was a PITA and an annoying decision from Pande but hey ho.. life goes on

a guy on curcoin told me that he gets super long gpu wu's that get really high bonus points. does anyone know how to get them? they are suppose to take 12 hours on a 980 so they would be great for my 24/7 rig.

Jumper118, I know that some of the units I get really push up the ppd numbers even on a 970 where I can get as high as >400K whereas some are as low as 200K. I haven't tried checking logs to identify the pattern but your post makes me think it would be worthwhile. I'll have a look this weekend. It may not be possible however to force the selection to only the good kind.

well mine get 300k normally so 400k would be a big jump.

Hmm..

OK, so every project has 100 frames

p11416 takes 5 hours per frame on a 1070, maybe a bit longer on a 980

On a 1070 that's 8.3 hours could be closer to 10 on a 980 maybe?

Doesn't seem to give much more points though compared to others.

Looking in my Work Unit history viewer on HFM.NET I can see some ZETA_DEV cores that took longer.

p10473 took 12:11 minutes
p10472 took 10:30 minutes

They were on a GTX970 think they were beta projects. That would take 10-12 hours to do, but problem is they only gave out 70K PPD