Why kernelExecTimeoutEnabled is TRUE for all GPUs?


I have just noticied that kernelExecTimeoutEnabled device property is TRUE for all the GPUs in the system, not only for those that have a monitor attached.

Definitely, some time ago kernelExecTimeoutEnabled was TRUE for the GPU with a monitor and FALSE for all others, however, after drivers updgrade or may be after CUDA upgrade this behaviour has changed unexpectedly.

What might the the reason? System config is: Win7 x64, CUDA 5.5, eight GTX 580 GPUs, 320.49 drivers.

Thanks in advance.

I believe registry keys have to be set for every GPU on which you want to disable the timeout.
And it only works on non-display GPUs.

TDR settings are not per-GPU but system-wide… also, the purpose of kernelExecTimeoutEnabled property is in identification of GPUs that have active watchdog, it is CUDA that reports whether particular GPU is subject to be TDR’d or not.

My understanding is that CUDA can turn off TDR only for GPUs running the TCC driver, while it is in force for all GPUs using the WDDM driver, and needs to be specifically disabled by the user, at the user’s risk. Relevant Microsoft documentation can be found here: