SDK 3.0 and 3.1 Limitations

Multi-GPU applications: While TotalView supports debugging multi-GPU applications, debugging multi-GPU applications is not officially supported by the CUDA debugging environment when using the SDK 3.0 and 3.1 drivers, so you may experience problems. To test debugging multi-GPU applications using these drivers, you must set the following environment variable to 1:

CUDA_DEBUGGER_MULTIGPU_ENABLE

Debugging multi-GPU applications using the SDK 3.2 driver and TotalView 8.9.1 is officially supported.

Host memory: Host memory allocated with cudaMallocHost() is not visible in TotalView when using the SDK 3.0 or 3.1 drivers.

Multi-threaded programs:

Multi-threaded applications may not work. In particular, there are known problems with programs that contain multiple host threads (pthreads) that wish to control the same GPU device. More specifically, each thread creates a context on the same device and then wants to start launching work on it (first come, first serve). NVIDIA is actively working to solve these multi-threaded problems.

For some multi-threaded programs on Fermi, the CUDA debugging environment kills the debugger with an internal error (SIGSEGV). NVIDIA is working to resolve this problem.

Textures: Debugging applications using textures is not supported.

Running X11: X11 cannot be running on the GPU that is used for debugging because the debugger effectively makes the GPU look hung to the X server, resulting in a deadlock or crash. Two possible debugging setups exist:

Remotely accessing a single GPU (using VNC, ssh, etc.)

Using two GPUs, where X11 is running on only one

Note: The CUDA driver automatically excludes the device used by X11 from being picked by the application being debugged. This can change the behavior of the application.