Hi - this is a strange question. I’m wondering if anyone can recommend a workaround.
I wrote a CUDA kernel that runs in several minutes. When I run the code on my windows 10 laptop with a 970M processor, the code runs and the display is responsive. We also have a Windows 10 desktop machine with a Titan. The same code runs quite a bit faster on the desktop, however, the display is unresponsive during the kernel run.
On both machines I had to set the TDRDelay to a large number to stop windows 10 from resetting the driver during my kernel run.
I’m struggling to understand why the same code causes a display freeze on the more capable platform while the display works on the less capable platform. Does anyone know the root cause or any workaround that I can implement? it drives the users crazy when the display stops responding as they think the program has crashed.
i could buy a second video card but I’m wondering if anybody knows root cause.
Your laptop likely is not using using the CPU’s IGP, not the 970M, for display, meaning the 970M will run without interfering with the display. You can use the CUDA deviceQuery tool to see if the GPU is used for display or not (look for the runtime limit).
Yes, you can buy a second GPU for your desktop… a simple cheap one is fine. Something like a $75 GTX 750Ti is more than enough (I’m using one now for display, at 4K). It saves a lot of hassles.
Thanks for your reply! I think this is exactly what is going on.
I opened up the NVIDIA Control Panel and confirmed that some programs are using the integrated processor on the laptop. It also explains something strange that I noticed when running on the laptop: when my kernel was running, my Visual Studio window would freeze up. In the Control Panel settings, Visual Studio is set to use the High-Performance NVIDIA processor. So it makes sense now given your explanation.
looks like I’m buying a second video card.