I think this is a hardware/memory issue. Titan X

I’ve been trying to get into Cuda with style transfer. My previous Nvidia card worked, but was limited with only 4GB of memory. I upgraded to the Titan X, and started having issues. I was running everything under Windows 10.

It then occurred to me to run a test on the memory. I used this:

I switched to my linux boot to get it working, and the results are just spewing with errors. (I’ve attached the output, plus the Nvidia bug report.)

I went to NVIDIA helpdesk, but as soon as they heard the word “Linux” they sent me here.

I’m certain this is bad card, but I can’t afford to replace it. I would be delighted to be proven wrong.

Memory Tests:

Nvidia Bug Report

I’m sorry to say that you’re correct, might be a thermal defect on one of the memory chips.

That would be surprising considering it’s water cooled.

Any way to mark memory as bad so the card doesn’t use it?

I meant this:
https://www.technolab.de/en/solder-dictionary/smd-hmd/cold-solder-joints.php

Paging out memory is only supported on Teslas and some Quadros with ECC memory.