I was having the problem of sluggish framerates and high response time when hcc1 tasks were running on my primary display GPU, even just navigating Windows' GUI. Another user suggested using the exclude_gpu element in the cc_config.xml file to limit crunching to a different GPU. I have two GTX 460s, so I gave it a go.

Here is my cc_config.xml:

<cc_config>
<options>
	<use_all_gpus>1</use_all_gpus>
	
	<exclude_gpu>
		<url>http://www.worldcommunitygrid.org/</url>
		<device_num>0</device_num>
		<type>NVIDIA</type>
		<app>hcc1</app>
	</exclude_gpu>
</options>
</cc_config>

This solved the framerate issues, even if it did sacrifice roughly half my crunching capability. But I noticed that the fan of my primary display card --€” the excluded GPU0 --€” was higher than it should have been for a practically idle card.

I checked out the cards with NVIDIA Inspector's graph, and saw that the clocks and voltages (and temperature) of GPU0 were kept high when GPU1 was crunching --€” even when GPU0 was idling. I changed cc_config.xml to exclude device 1 instead of 0, and the idle card still heated up when the other card was crunching hcc1.

Clocks and voltages returned to normal idle levels when GPU activity was suspended.

Here's an imgur album of NVIDIA Inspector graphs comparing GPU clocks, voltages, and temps while crunching, suspended, and during normal, non-BOINC usage.

I have no idea what the problem is, but I don't like that hcc1 is consuming power on an idle card.

Windows 7 64-bit
GeForce driver version 306.97 WHQL
BOINC 7.0.36

In case it matters, the cards are not in SLI and not connected by an SLI connector.