Hi all,
I'm trying to run a GTX750ti and a GTX260 in the one machine but I'm having GPUGRID computation errors.
I'm running NVidia driver version 'NVIDIA-Linux-x86_64-340.93' as I believe this is the latest driver that will recognize my 260 card while still detecting the 750ti. Later versions of the driver reported the 260 card as 'supported under 340.xx legacy drivers' and the card as ignored.

I also run SETI@home and LHC@home on this machine

software versions:

cuda_7.0.28_linux
boinc 7.4.23
Kernel 3.19.0-15-generic #15-Ubuntu SMP x86_64

Boinc event log:

Starting BOINC client version 7.4.23 for x86_64-pc-linux-gnu
Libraries: libcurl/7.38.0 OpenSSL/1.0.1f zlib/1.2.8 libidn/1.28 librtmp/2.3
CUDA: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 340.93, CUDA version 6.5, compute capability 5.0, 2048MB, 1744MB available, 1472 GFLOPS peak)
CUDA: NVIDIA GPU 1: GeForce GTX 260 (driver version 340.93, CUDA version 6.5, compute capability 1.3, 896MB, 845MB available, 805 GFLOPS peak)
OpenCL: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 340.93, device version OpenCL 1.1 CUDA, 2048MB, 1744MB available, 1472 GFLOPS peak)
OpenCL: NVIDIA GPU 1: GeForce GTX 260 (driver version 340.93, device version OpenCL 1.0 CUDA, 896MB, 845MB available, 805 GFLOPS peak)

cc_config:

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
</options>

</cc_config>


GPUGRID failed task (fails immediately after starting task):

process exited with code 212 (0xd4, -44)
https://www.gpugrid.net/result.php?resultid=14695331

Questions:
1/Can I run two different driver versions in parallel, the older driver running the 260 while the newer driver the 750ti?
https://a248.e.akamai.net/f/248/10/10/us.download.nvidia.com/XFree86/Linux-x86/331.20/README/faq.html seems to suggest I can.

2/Could I use a cc_config file to restrict GPUGRID work to the 750ti?

The GTX 260 is no longer supported by GPUgrid.
It's great that you're looking to contribute, but you'd be much more effective if you retire that part and replace it with a current model.

With rapid innovation, the older generations are being rapidly outclassed by the new. Letting your 260 run would be allowing you to waste your electricity and your time.

For example, if you compare the GTX 260 to this generation at the same quality tier, the GTX 960, you'll see that:

The newer card has a 75% higher clock speed
Three times the memory clock speed
About four times the floating point operations per second (This matters a lot)
Double the video RAM.
Five times the shading units
And so on.

AND, the newer card uses 30% less electricity to do all of that.

I'm trying to run a GTX750ti and a GTX260 in the one machine but I'm having GPUGRID computation errors.

It's no wonder as the GTX260 (the whole 2xx series) is deprecated by GPUGrid.

I'm running NVidia driver version 'NVIDIA-Linux-x86_64-340.93' as I believe this is the latest driver that will recognize my 260 card while still detecting the 750ti. Later versions of the driver reported the 260 card as 'supported under 340.xx legacy drivers' and the card as ignored.

To avoid errors with the GTX750Ti you should use the latest driver available.
You should retire that old GTX260, its energy efficiency too low to use it for crunching.

Questions:
1/Can I run two different driver versions in parallel, the older driver running the 260 while the newer driver the 750ti?
https://a248.e.akamai.net/f/248/10/10/us.download.nvidia.com/XFree86/Linux-x86/331.20/README/faq.html seems to suggest I can.

It says "You can build and install multiple kernel modules" to "minimize software overhead when driving many GPUs in a single system", but it doesn't say they could be different versions. I'm not a Linux expert, so I can't disprove it for sure.

2/Could I use a cc_config file to restrict GPUGRID work to the 750ti?

Yes, you can do it. See the BOINC wiki.
<exclude_gpu>
Don't use the given GPU for the given project.
If <device_num> is not specified, exclude all GPUs of the given type.
<type> is required if your computer has more than one type of GPU; otherwise it can be omitted.
<app> specifies the short name of an application (i.e. the <name> element within the <app> element in client_state.xml). If specified, only tasks for that app are excluded.
You may include multiple <exclude_gpu> elements.
If you change GPU exclusions, you must restart the BOINC client for these changes to take effect.

<exclude_gpu>
   <url>project_URL</url>
   [<device_num>N</device_num>]
   [<type>NVIDIA|ATI|intel_gpu</type>]
   [<app>appname</app>]
</exclude_gpu>

Wow. Thanks for the quality replies!

I'm trying to run a GTX750ti and a GTX260 in the one machine but I'm having GPUGRID computation errors.

It's no wonder as the GTX260 (the whole 2xx series) is deprecated by GPUGrid.

I should clarify: The GPUGRID tasks that failed were running on the 750ti and not the 260.
I would be happy to run SETI@home tasks on the 260 while the 750ti was stable running GPUGRID but I don't know how to achieve this. I suspect there must be a way to get each card bound to a different driver version.

I suspect there must be a way to get each card bound to a different driver version.

It doesn't seem to be an easy thing to do. You would have to have the same module twice in your system, with two different sets of binaries. For each module, its binaries link to one-another using version-independent library names, which may cause name clashes. You would have to have two sets of binaries in your initramfs image for the same module. Lastly, you would need to load the same-named module twice, modprobe seems to support this, but not necessarily so the nvidia module, as it may assume it is only loaded once, acquiring global resources, etc.

I'm not saying this is impossible, but I'm not sure I would go down this route. Perhaps it's worth trying at least, I wish you good luck if you do! :)

---