GPUGrid: Happy crunchmas!

[RSS]

GPU queue Unsent In Progress
Short 0 129
Long 508 3,065 ... GPU queue Unsent In Progress
Short 497 264
Long 507 3,063

More...

[zombie67]

There we go. No down time!

[FourOh]

Just got back in town from a week with family on the west coast... of course immediately after leaving one of my machines crunching GPUGrid went down and another stopped requesting GPUGrid work. Even with all other projects suspended and no work in queue, BOINC manager 7.4.27 would not request GPU work. So, I updated to 7.4.36 which seems to have solved the scheduling issue. Back to full steam

[FourOh]

This latest batch of GPUGrid long run work units involve smaller molecules (moderator's explanation, not mine) than usual, which results in lower GPU load. My GTX 750 Tis have gone from around 90% down to 70-80% load. Unfortunately for me, I can't run 2 concurrently and still make the 24 hour bonus, but those of you with faster graphics cards might benefit from doing so.

The reduction in GPU usage has decreased my daily production by about 10%. Not a huge deal, but it has added a few days on my run to 100M.

[FourOh]

OK after thinking about this and realizing my logical error (that running 2 work units concurrently would result in 2x the run time) I decided to give it a try. After all, if I'm getting 12.5 hour run times at 70-80% GPU usage I shouldn't have any problem getting sub-24 hour times at 95% GPU load when running 2 units. Initial results are a success... running 2 concurrent Long Run units on the 750 Ti improves daily production by at least 15% with a low risk of missing the 24 hour bonus.

Note: I bumped the clocks on this 750 Ti to 1400Mhz with a 12mv voltage increase over stock. I'll reduce to factory OC to confirm the units still complete in under 24 hours, but it looks like that shouldn't be a problem.