Well there seems to be a silver lining to these beasts, at least for those also using the cpu to Fold on their rig(s).
According to posts on the Folding Forums these WU's seem to use a lot less cpu time, speculation is it's because the checkpoints are further apart.
From this thread
Folding Forum • View topic - New Gpu Wu's good news for bigadv Folders.
Post by
Jester;
Quote:
Now that these bigger Gpu Wu's have been released to -advmethods (here viewtopic.php?f=24&t=19390)
there seems to be little to no impact on bigadv smp frametimes here so far due to the larger, 5 minute+ frametimes,
Running a GTX470 @ 775 core alongside an i7 970 @ 4.2hz and seeing little difference in smp frametimes with and without the Gpu client running,
the new Gpu Wu's do add more stress though, even watercooled the GTX470 Gpu temps went up 3-4c so I'd heed the warning in the linked thread with
air cooled and/or overclocked Gpu's,
Compared to the smaller Gpu Wu's with frametimes of a little over a minute these big Wu's make smp + Gpu Folding a viable option again, with the smaller Wu's
bigadv smp on my 970 Cpu took a ppd hit of around 6k ppd ...... |
Response by Pande Group Menber
tjlane;
Quote:
Thanks, Jester! This makes some sense. Basically, the FAH GPU core is very good about keeping all of its calculations on the GPU, where calculations can be done quickly, and only uses the CPU to send results back to be written to memory. These new bigger units will write less frequently, since their calculations take longer, thus use less CPU power.
Thanks for the cool news!
|
-Bigadv and GPU Folding together again...