Tried a few of the MB benchmarks on Lunatics site and they all appear to be very similar. Been using MB_bench_208 and it works nicely, however I'm wonder if I can target a specific GPU. One of my crunchers has two 750ti's and one 560, I'd like to test the 560 but it's Device 2.

Tried using Fred's performance app and for some reason it errors out on the stock app. I was able to add AP V7 so I think I'm doing it right.

Tried a few of the MB benchmarks on Lunatics site and they all appear to be very similar. Been using MB_bench_208 and it works nicely, however I'm wonder if I can target a specific GPU. One of my crunchers has two 750ti's and one 560, I'd like to test the 560 but it's Device 2.
...

I'll provide what I see from a quick check of the source code. Consider it a starting point, if someone who has actually done what you're attempting contradicts me, believe them.

The x41zc CUDA apps have command line parameters which can be used to control which device is used, either "-usedcudadev 1" or "-device 0" without quotes chooses the first device.

So if you put something like

Lunatics_x41zc_win32_cuda50.exe -usecudadev 1
Lunatics_x41zc_win32_cuda50.exe -usecudadev 2
Lunatics_x41zc_win32_cuda50.exe -usecudadev 3

in the BenchCfg.txt file section for changes from the default arguments, test WUs would sequentially run on each of the devices. I suggest doing that because the 560 may not be considered device 2, your task details are showing it as CUDA Device 1.

                                                                  Joe

...the 560 may not be considered device 2, your task details are showing it as CUDA Device 1.

and I'll add for clarity that the device numbers allocated may vary by drivers/OS and slot, so the only guaranteed fixed location is the PCI[e] Bus & Slot number, which aren't used by Boinc.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Tried a few of the MB benchmarks on Lunatics site and they all appear to be very similar. Been using MB_bench_208 and it works nicely, however I'm wonder if I can target a specific GPU. One of my crunchers has two 750ti's and one 560, I'd like to test the 560 but it's Device 2.
...

I'll provide what I see from a quick check of the source code. Consider it a starting point, if someone who has actually done what you're attempting contradicts me, believe them.

The x41zc CUDA apps have command line parameters which can be used to control which device is used, either "-usedcudadev 1" or "-device 0" without quotes chooses the first device.

So if you put something like

Lunatics_x41zc_win32_cuda50.exe -usecudadev 1
Lunatics_x41zc_win32_cuda50.exe -usecudadev 2
Lunatics_x41zc_win32_cuda50.exe -usecudadev 3

in the BenchCfg.txt file section for changes from the default arguments, test WUs would sequentially run on each of the devices. I suggest doing that because the 560 may not be considered device 2, your task details are showing it as CUDA Device 1.

                                                                  Joe

That did the trick.. It's just a matter of finding out which device is which.. In the computer and according to windows the 750's are dev 0 and 1, and the 560 being in the 3rd slot is Dev 2 which makes sense.. However MB_bench see's things differently.. I tested on my 4 GPU box and for some reason this is how it worked out..

Dev 0 = Dev 3
Dev 1 = Dev 2
Dev 2 = Dev 4
Dev 3 = Dev 1

According to Boinc manager the Device ID is the in the right order 0 1 2 3, but the Stderr output shows pcibusID as 4 2 1 3.. No idea why MB_bench see's 3 2 4 1. I'll just go about my testing and let the guru's figure it out:)


Thanks for the fix:)