Swapping TitanX with K20 under Ubuntu 14.04


I have a lenovo workstation running Ubuntu 14.04, a Quadro K620 for video,
a TitanX for machine learning and dev driver 352.41.

The machine is locking up when I am using tensorflow.

So today I pulled out the TitanX and put a K20 in its place. The machine is not locking
up when I run tensorflow.

That’s when it occurred to me … the latest version of the dev driver for a K20 under Ubuntu
and Cuda 7.0 is 346.96. The driver for the TitanX (as I mentioned previously) is 352.41.

Am I challenging fate by pulling out the boards and swapping them without thinking about
the drivers under Ubuntu?

I periodically have 1 of these 3 configurations in my Linux box.

K620 with TitanX
K620 with K20
K620 with TitanX and K20

How do you decide whether to grab the Tesla or GeForce device driver? I’ve never been certain
about this. Does the unified device driver concept extend to Linux?


So let me see if I get this: Titan X with Titan X driver is causing you problems, but K20 with Titan X driver works fine? That’s strange…

The unified driver concept extends to linux. That is why a GeForce driver can work on a Tesla and vice versa.