Introduction to Cuda C Programming-Programming Interface (3.4) Computing Model

Tesla solutions running on Windows Server 2008 and later versions or Linux can be used through the NVIDIA System Management Interface (NVIDIA-SMI, as a tool for release drivers) set any device in the system to run in one of the following three computing modes:

• Default computing mode: Multiple Threads can use devices at the same time (use the Runtime API to call cudasetdevice () on this device, or use the driver API to associate the current context with the device ).
• Computing mode exclusive to the processor: create only one Cuda context across all the processor systems, and as many threads as possible for one processor.
• Computing Model with exclusive processor and thread: Only one cross-system Cuda context is created, and only one thread can be created.
• Disable computing mode: The Cuda context is not created on the device.

This means that, in particular, if device 0 is the computing mode exclusively occupied by the processor and used by another processor, or the computing mode exclusively exclusive to the processor and thread, or the computing mode is disabled. A host thread that does not explicitly call the cudasetdevice () Runtime API may be associated with a device other than 0. Cudasetvaliddevices () can be used to set a device from the priority list of a device.

The application can check the properties of the computemode device to query the computing mode of the device.