My GPU is GTX Titan Black. OS is Windows 7.
I do following steps:
- I run pre-compiled CUDA kernel in Matlab. The task is very time consuming, it takes most of GPU resources, GPU’s fan blows fast, MSI Afterburner shows 80% of energy consumption, etc.
- I run some OpenGL application. GPU fan go silent, power consumption drops, first task slows.
- I close that OpenGL application(2). But GPU computing power doesn’t restore for task 1.
- I can repeat steps 2-3 many times, and every time task 1 will slow more and more.
It seems to me that only reboot is able to fix this?
(Offtop) And i don’t understand this pricing for Tesla GPUs, with this and other bugs (it is hard for me to describe all of them). I’m going to buy DGX-1 or something like it, but it is so expensive and hard to get, and, while i evaluate CUDA on GTX Titan, it looks like i can fall in so many pits rendering so expensive device being unusable. Too risky. I got in a trap once before, it was a GeForce 3xx series GPU not capable to perform fp64 computations at all.
And default 83C temp limit for GTX Titan Black doesn’t seem to be good: one of my GPUs died during first night. I had to lower temperature limit using MSI Afterburner application.