just fired up GPUGrid for the first time in a long time.
picked up

18/02/2014 8:49:13 PM | GPUGRID | Restarting task 196-GIANNI_ntl-1-4-RND2439_0 using acemdlong version 815 (cuda55) in slot 8

It's only loading the GPU to 67% or so, is that right or do I need to update something?

There is a problem using GTX 780 and GTX 780Ti with any Windows version from Vista onward at GPUgrid that causes the low utilization. It's the WDDM used in those versions. I'm a Linux guy and not so much into Windows so I'm not aware of all the details but the recommended solution for that problem is to switch to XP (which doesn't have the WDDM) or Linux.

Mind you 67% seems lower than what other Vista/7/8 users have been getting so you may have other problems in addition to the Vista/7/8 problem. For example your GPU may be overheating and downclocking to reduce temperature and protect itself. Try to keep it below 70C. You might be throttling the GPU and/or CPU that feeds the GPU. Many users allow 1 free virtual core per GPU, perhaps 1 real core (or 2 virtual cores) would be better?

There are a few other very recent threads on the matter with advice from more experienced Win users. Most of your concerns are addressed in those threads already.
____________
BOINC <<--- credit whores, pedants, alien hunters

Perhaps not in this case the current GIANNI WU's have a lower GPU load on my 660, 770 and 780Ti.
And with the hampering of Win7 (or 8) the 780Ti is twice as fast as a 660, but three times more expensive.

____________
Greetings from TJ

The GIANNI were only 55-60% on my 780Ti on Win7 vs the Santi's being 70%. For fun, I ran two WU's. Watched the times for a single WU go from 17,200 to 17,400 up to 27,300 to 27,500 seconds with a 80% utilization.

Even my 680 on XP went from a normal 95-97% utilization down to 90% utilization with the GIANNI WU's.

well that is a shame.
Currently running a Santi @ 77% which is much better.
I think I still have a copy of XP around somewhere.
I might just have to install it on a VM.
what kind of usage are people seeing with a 780 under XP?

I think I still have a copy of XP around somewhere.
I might just have to install it on a VM.

I'm currently using driver 327.23. I have to use this driver because I do Folding as well and any later driver doesn't work with Folding.

What do you mean by have the CPU set to 95% usage and dropping it to 50%?
I leave a couple of CPU cores free (2600K I have 8) and run 6 cores of Docking@home and 1 instance of GPUGrid, should I drop to 4 cores of Docking?


It seems it is only the GIANNA units that have this problem to such a degree, SANTI units run at 77/78%.

When using the Beta drivers and CPU set to 95% usage, these GIANNI WU's were only using 65 and 66% of the GPU. When I changed to 75% CPU this rose to ~68% and 70%.
Dropping to 50% CPU usage allowed the GPU utilizations to rise to 71% and 75%.
With no CPU WU's running these tasks run at 75% and 77%.
Process Hacker tells me that Virtualization is enabled for these WU's.

got you. Currently crunching a NOELIA_FXArep @ 84-88% with 2 threads free. best usage yet. Looks like it is going to take about 5.5 hours to run.

327.23 should be fine.

I just mean setting the CPU to 95% (use 7/8 CPU threads) and 50% (4/8 threads).

When you use less CPU for CPU projects there is a slight increase in GPUGrid performance. The GIANNI WU's utilize the GPU to a lesser extent to begin with, so they are more vulnerable to resource contention and thus benefit more when you free up more CPU cores/threads. Even with no CPU projects running I only got the GIANNI WU's to run at 76% GPU utilization. With a SANTI_MAR task and a GIANNI_ntl on my system the GPU Utilizations are 84% and 72% respectively - so the SANTI_MAR tasks are about 17% faster on my setup when running mixed task types.

I've got an FX-8370 with a GTX 750 in Linux.

On an idle system with only GPUgrid in boinc, it's pretty much a solid 98% for GPU utilization and 40% for PCIe(2.0) utilization. I'd imagine not having a PCIe 3.0 mainboard would be a serious bottleneck for running current GPUgrid tasks on the fastest cards.

For me, with 8 cores for boinc CPU tasks, GPU utilization probably averaged ~70%.(It jumped around a lot.)
With 7 cores, it averaged somewhere around 87%.

Disabling PCI/PCI-e CLK Status in bios got it to ~91%.

With 7 cores for boinc CPU, PCIe clk status off and an idle desktop:
If I just taskset GPUgrid(acemd...) to core 0, it's worse than 70%.
If I then taskset the other active tasks to all but core 0, it's a little better, with frequent brief moments of >95%, but still bad.
If I leave the other tasks with an affinity for all but core 0 and GPUgrid alone is unrestricted, it probably averages around 93%, jumping around a bit.
If GPUgrid is unrestricted, affinity for one other boinc task is all but core 0 and everything else is set to avoid both core 0 and core 1, GPU utilization is a pretty steady 95-97%.

With 8 cores for other boinc, the affinity hole doesn't do much.

Regardless of system load or cores given to boinc, other boinc GPU tasks I've run are almost continually 99%. They also demand very little PCIe.