The GPUs ID number does not match that reported by boinc on the boinctasks program. For example, the following LOG from tthrottle does not list an ID.
nvidia: found 3 logical devices
nvidia: found 3 physical devices
nvidia: Temperature 66 °C, max Temperature 127 °C
nvidia: Temperature 58 °C, max Temperature 127 °C
nvidia: Temperature 52 °C, max Temperature 127 °C
nvidia: GeForce GTS 250, GeForce GTS 250, GeForce GTX 650 Ti
The following screen shot shows the temperature display from tthrottle
(http://stateson.net/images/tth.png)
I am guessing that the ID numbers in the above log should correlate to 2,1,0 since the message log (boinc) shows the following
2/6/2012 9:46:08 PM | | NVIDIA GPU 0: GeForce GTX 650 Ti (driver version 310.90, CUDA version 5.0, compute capability 3.0, 1024MB, 8381368MB available, 1425 GFLOPS peak)
2/6/2012 9:46:08 PM | | NVIDIA GPU 1: GeForce GTS 250 (driver version 310.90, CUDA version 5.0, compute capability 1.1, 1024MB, 974MB available, 705 GFLOPS peak)
2/6/2012 9:46:08 PM | | NVIDIA GPU 2: GeForce GTS 250 (driver version 310.90, CUDA version 5.0, compute capability 1.1, 1024MB, 950MB available, 705 GFLOPS peak)
I wish that the log in tthrottle show the same GPU ID number that BOINC shows. The reason for this is that I restrict some GPU's from crunching certain projects because some overheat. Yesterday, I had a gts250 up at 92c on primegrid. This one, with the high temperature, was in the middle of 3 PCIe slots and does not get enough air even with eVga precision 3.04 setting the fan at %100. It runs just fine on collatz, poem, and other projects that use opencl. However, primegrid easily overtaxes this board and I would have to set the gpu and memory clocks down to minimum to prevent overheating. It is easier just to exclude the gpu from processing. To do that, I need to know the device number that boinc uses.
Put this down at low priority because boinctasks does show the temperature and the device (gpu) number. However, I do not always have boinctasks available and if you have only tthrottle and boincmgr then there seems no way to properly id the gpu. Obviously, this affects only those systems with more than 1 gpu.