I don't have a problem with suspending BOINC myself before gaming, but in my previous post I have described somewhat effective ways for me to keep from going over 24 hours to process a WU. The best seems to be for me to every once in a while, delete the second task that hasn't yet begun processing. I do this a few times a day to ensure that I will always have a more fresh WU. 33.4% credit (in my previous example), just because of not being able to upload a WU is rather harsh. If I am trying to maximize my RAC, then it proves necessary that I keep cancelling WUs and wasting bandwidth by downloading WUs somewhat regularly and abandoning them. As it can take a while to maximize RAC, I have to do anything I can to earn all credit (Including Bonus).

I know that not every BOINC user participates in Gridcoin, but in Gridcoin payouts are determined based on the % of RAC of a team member to the whole of the team. The higher reward is too much for me to give up, if it's just a matter of me cancelling a few WUs and wasting bandwidth. If I don't waste bandwidth, then I end up getting less RAC which translates into less Gridcoin earned. The best way for me to maximize RAC is to be wasteful, and to tediously monitor + configure my BOINC client. The jobs take a very long time to complete, and even on top cards it is possible to miss the deadline for a large % of the users participating in the project just because their computer isn't available to crunch 24 hours a day.

If you want to maximize your RAC by maximizing your opportunity for GPUGrid bonuses, automatically, then you should consider setting both of your buffers (Store at least x days of work) and (Store up to an additional y days of work) ... set them both to 0.

Then, work fetch will only start asking projects for more work, when the current work is about 3 minutes away from leaving a resource idle.

Have you tried this yet?

Edit: In fact, you can figure out about how long it takes you to download a GPUGrid task, then you can set "Store at least" to that value (something like 0.02 days, for instance).

Nope, I can give this a shot though. ;)

I edited the post to make a 2nd recommendation, since it does take some time to download a GPUGrid task.

How I imagine your GPUs after OPM:

Are we finished these off in 9 days? :) Or you took them off?

I don't know if others are still seeing the high error rates, but it hasn't been a problem for my GTX 970 (Win7 64-bit) yet. It has just been running a week, so maybe the work units have been fixed. Or else it is the usual problem with overclocking that some people run into.

I have also had never-ending problems with anti-viruses on my main PC, and just don't use them on my dedicated machines where I do GPUGrid, so that may help too.
https://www.gpugrid.net/results.php?hostid=336002

But more interesting to me, and maybe it deserves another thread, is that according to a note on one of the other sites (maybe Einstein; I can't find it now), going back to an earlier driver can help productivity on the Nvidia cards. So I went from 361.43 (the first of the CUDA 8 drivers) back to 359.06 (the last of the CUDA 7.5 drivers), and picked up about a 4% improvement in times on Einstein BRP6-Beta on my GTX 750 Ti's (Win7 64-bit), and the CPU usage drops from 13% to 11%. I am trying it now on GPUGrid on both my GTX 970 and 960, and it seems to help here too, though how much is too early to say.

Whether that holds true for OpenCl work I don't know, but will find out eventually. And even earlier drivers might be better, but the last CUDA 7.5 driver is OK for me for the moment.

EDIT: The older drivers are here: http://www.nvidia.com/Download/Find.aspx

How I imagine your GPUs after OPM:

I managed to finish another one of your wimpy units, from your latest batch.

Name 1dbh-SDOERR_opm993-0-1-RND7068_1
Workunit 11632064
Created 8 Jun 2016 | 12:12:25 UTC
Sent 10 Jun 2016 | 11:20:39 UTC
Received 10 Jun 2016 | 21:44:17 UTC
Server state Over
Outcome Success
Client state Done
Exit status 0 (0x0)
Computer ID 263612
Report deadline 15 Jun 2016 | 11:20:39 UTC
Run time 36,946.29
CPU time 36,776.52
Validate state Valid
Credit 335,160.00
Application version Long runs (8-12 hours on fastest card) v8.48 (cuda65)
Stderr output

<core_client_version>7.6.22</core_client_version>
<![CDATA[
<stderr_txt>
# GPU [GeForce GTX 980 Ti] Platform [Windows] Rev [3212] VERSION [65]
# SWAN Device 0 :
# Name : GeForce GTX 980 Ti
# ECC : Disabled
# Global mem : 4095MB
# Capability : 5.2
# PCI ID : 0000:01:00.0
# Device clock : 1266MHz
# Memory clock : 3505MHz
# Memory width : 384bit
# Driver version : r358_00 : 35906
# GPU 0 : 59C
# GPU 1 : 54C
# GPU 0 : 60C
# GPU 1 : 55C
# GPU 0 : 61C
# GPU 1 : 56C
# GPU 0 : 62C
# GPU 1 : 57C
# GPU 0 : 63C
# Time per step (avg over 5000000 steps): 7.388 ms
# Approximate elapsed time for entire WU: 36942.074 s
# PERFORMANCE: 112027 Natoms 7.388 ns/day 0.000 ms/step 0.000 us/step/atom
17:41:29 (6076): called boinc_finish

</stderr_txt>
]]>


The GPU usage was 80% max. Power usage was 75% max. Memory usage was 1532 MB.

It was not much of a challenge. The GPU didn't even sweat. Let's see if you can do better next time.

147,415 atoms on this task. 21.3 hours on a GTX 970. 882,189 points. Nice to see these again.

Thanks for the post!