Hi, I’ve been running neural net simulations in CUDA on Tesla K20 card (with Quadro K600 for video out), and I just bought Titan X card, to speed up the simulations. To be specific, I’m not planning to run a single simulation on these two cards. I need the second card so I can run other simulations in parallel.
Are there any potential issues with running both Tesla and Geforce cards? What driver should I use?
In general, you can mix a Tesla and a GeForce GPU if you plan to use them independently. As an example, P2P transactions may not work between a Tesla K20 and a Titan X.
However, whether or not a particular workstation will boot properly with a particular complement of cards is not something that can be answered in a general case. You will have to try it.
If you use the latest CUDA toolkit (7.5) it should provide a driver supporting both a Titan X and Tesla K20.
Ok, I just plugged in the Titan X card, and both cards seem to be working fine with my CUDA code.
In terms of drivers, I’m a little confused:
The driver I have now is 354.35. It was downloaded from Nvidia site when I only had Tesla/Quadro cards installed.
When I checked the site now, it offered me 354.13 version if I choose Tesla drivers, and 358.91 if I choose Geforce drivers.
If I run CUDA 7.5 installer (I already have CUDA 7.5 installed), it says it will replace my current driver with version 353.90
What should I do?
If your system is working the way you want, why not leave the driver alone? 354.35 should be fine.
Hi, I just saw your post … after I posted mine to the same group titled “Swapping a TitanX with K20 under Ubuntu 14.04”. Our questions are very similar.