Installing multiple adapters for CUDA and OpenCL -- Quadro FX 570 and GeForce GTX 1050 Ti

Is it possible to have these two adapters installed simultaneously, with the old Quadro FX 570 handling the actual display and the newer GeForce GTX 1050 Ti performing only CUDA and OpenCL calculations?

Or do I need two GTX 1050 Ti boards in order to have one of them dedicated to CUDA / OpenCL?

The Quadro FX cards are too old to be used with current drivers or current CUDA installations.

You would have to limit yourself to an older (e.g. R340) driver and CUDA 6.5, which would not work at all with GTX 1050 Ti.

Hey! Thanks for the reply!

So if I yanked out the old Quadro FX 570 and added a 2nd GeForce GTX 1050 Ti then could I use one for the actual display and the other one as a “compute exclusive” device?

Yes, but the platform (windows or linux) may still have an impact on your final experience. In my opinion this is fairly easy to set up in linux, but in Windows, a WDDM display driver stack will still be built on both GPUs, even though you want to “reserve” one for compute, and at least with respect to CUDA, this will have significant implications for behavior, such as WDDM, especially WDDM timeout.

I ordered the 2nd GeForce GTX 1050 Ti. :-)

Thanks! Yes, one step at a time! :-)

My plan is to get started with these less expensive GPU boards in Windows 7 on an older machine (starting with my existing code and tools) – and then migrate to Ubuntu and higher performing equipment later.

What is the best development environment in Ubuntu? Intel Parallel Studio XE? – I have an old version of that package called Intel Composer XE 2011 on Windows.

Update… I am off to a decent start!

I have the two GeForce GTX 1050 Ti boards installed in the oldish Dell server under Windows 7.

And I have built two sample OpenCL projects in Visual Studio 2013 – one is plain OpenCL ‘C’ and the other is the oclDotProduct C++ sample from NVIDIA.

The first sample project queries the device capabilities using the OpenCL ‘C’ code from Chapter 3 of “Heterogeneous Computing with OpenCL 2.0” by Kaeli, Mistry, Schaa, and Zhang.

The second sample project runs NVIDIA’s oclDotProduct kernel – but I haven’t really examined the code yet. I did find a small bug in the NVIDIA OpenCL 4.2 SDK.