Tesla K20 and GeForce in one machine possible?

Hi everyone!

We are planning to buy a Tesla K20 powered workstation. Since we are currently doing GPGPU on GeForce (OpenGL and shaders) and we want to use the Tesla as an extension to do additional OpenCL and/or CUDA we’d like to have a GeForce GTX 680 and a Tesla K20 in one machine. Last friday some guy from PNY called me and told me that it is not possible to have a Tesla and a GeForce working in one machine because of driver incompatibilities and that i would have to use a Quadro. On the other hand i have some people who see no problem in selling me an according Tesla+GF workstation… so is this PNY guy just doing marketing (i think i dont need a K5000 - for that price i can buy 3-4 GF680s)?

So my questions are:

(1) Is it possible to do Tesla+GF in one machine using windows, OpenCL/CUDA?

(2) Has anyone actually done this successfully or has anyone failed (and why)?

(3) Any restrictions on hardware components or CUDA/OpenCL usage?

I’d appreciate if someone could help me out with this since i don’t want to buy an expensive system for nothing…

Hello,

At my working place we have a k20 with a 590 gtx. The machine is stable and the operating system is fedora linux. No matter wich OS you use you should be sure the power supply is enough and that the case is large enough to accommodate the cards. I think that both Windows and linux are ok for doing CUDA.

I am typing this response from a machine that has a K20c and a GTX680, Windows 7, and I use Cuda, Visual Studio 2010, and NSight. I have had this configuration for about a month – no problems. Make sure you have a large power supply and a large case.

Good to know it works ;)
The machine will most likely have a ~1200 W power supply with enough connectors - that should do it (and if not it’ll go back…).

Thanks for your responses - happy computing ;)

Hi all,

This is the only forum i found about Tesla and Geforce. I have two workstations with Tesla K20c + Gforce GT 610 each. But my requirement is to black list Gforce hardware and only use Tesla ones. I am unable to achieve it. Any help will be appreciated.

BR!

I’ve got a similar (Linux) system. Which driver(s) do I install? If I search for Tesla or GeForce, I get two different *.run files. Do I install both, or just the most recent one, which happens to be for the GeForce? Thanks!

If you want to use CUDA, I would just install CUDA. The driver that comes with it should work for both the Tesla and the GeForce card.

In any event, you can only install one driver at a time. You cannot “install both”. The second install will wipe out the first install.

I’m making an assumption here that your GeForce card is a Fermi or newer card. If not, you will need to use a CUDA version older than 7.0 (i.e. 6.5 or older).

Thanks for the clarification, and yes my GeForce card is a Maxwell device. The confusing part is that the 352.41 Linux 64-bit driver doesn’t list any Tesla devices under its supported products tab (http://www.nvidia.com/download/driverResults.aspx/90279/en-us).

The driver the comes with CUDA 7.5 should support both your Tesla and your GeForce card.