My GTX 470 runs @ 99% when running tasks. Is there any way to lower the usage to 90% or lower. With help from skgiven I seemed to have finally solved the error problems.
XP Pro SP3 32 bit
MSI GTX 465 GE unlocked to 470. No overclock
Intel i7-860 running stock
4GB 1600MHz ram (2x2)
Thanks.

---

Why would you want that? Better desktop responsiveness, or a cooler / more silent card and lower electricity bill?

MrS

---

Scanning for our furry friends since Jan 2002

Why would you want that?

Better desktop responsiveness

yes

or a cooler / more silent card

yes

lower electricity bill?

yes

Currently there is no mechanism in BOINC to regulate GPU usage, and I don't think there will be anything like that for quite some time (until GPUs can be scheduled by the OS like CPUs).

Edit: actually.. what you can do is to set BOINC not to use the GPU at all while you're actively using the machine. This will trigger BOINC back on if you're watching a video, but you can also set exclusive apps which prohibit GPU usage.

However, you can lower your GPU voltage a bit to reduce heat/noise/power consumtion. Since you're not OC'ing there's some headroom left. The typical OC utilities should be able to do this. This way you're still running your chip at maximum software efficiency (99% utilization) while improving hardware efficiency. You can even go further and lower clocks + voltage. Only lowering clock speed doesn't help as much.

MrS

---

Scanning for our furry friends since Jan 2002

I would have thought there would be a way to achieve what I want with an app_info file but I don't have the expertise to construct one that will work.

I don't think so. You can run multiple WUs concurrently with an app_info to increase GPu utilization, and you can pass additional command line arguments to the application. However, GPU-Grid doesn't support any responsiveness arguments - as far as I know.

MrS

---

Scanning for our furry friends since Jan 2002

Some time ago I suggested that Boinc be able to have an in use profile, so that people could crunch on say 5 (or whatever the user wanted) of their i7 CPU threads when the system is in use, and a different number when it's not in use, say 7.

However, even if that was implemented it would just be inching towards having a more responsive system when it comes to GPU usage. What you would really need to develop was code that allowed less GPU to be used. While you could probably do this by GPC I guess you would want to do it by SM count, which would be OK for high end Fermi cards, but not great for low end cards and bad for Kepler; a GTX570 has 15 SM's but a GTX670 only has 7. A 6.25% loss of GPU capability vs ~16%. The likes of a GT640 only has 2. I'm not sure apps could handle turning an SM on and off either. Anyway, don't expect any of this to turn up.

---

FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

I think with the turbo mode of the GTX680 nVidia has a better tool for the job. If they could expose something like "desired performance level" to the application and/or user, on could choose whether to run the GPU full throttle or in a more efficient regime.

Currently using less GPC or SM wouldn't yield any better responsiveness, since GPUs can't multitask (yet, without special coding) - either they're crunching or they're refreshing the display.

MrS

---

Scanning for our furry friends since Jan 2002

I don't think so. You can run multiple WUs concurrently with an app_info to increase GPu utilization, and you can pass additional command line arguments to the application. However, GPU-Grid doesn't support any responsiveness arguments - as far as I know.

MrS

I use an app_info when I run Milkyway to accomplish want I was hoping the same idea would work here.

Yeah, it's just a milkyway-specific option.

MrS

---

Scanning for our furry friends since Jan 2002