After a kernel fails due to “unspecified launch failure”, the only way I am able to use the device again is after a system reboot. After the ULF occurs, the device hangs on any CUDA calls. After a ULF, I try and run the “bandwidth” test from the SDK and the console window states:
Device 0: Tesla C2050
The system I am using is a Windows 7 Professional 64-bit Virtual Machine running on Parallels Extreme Workstation. I connect to the machine using Windows Remote Desktop Connetion. The board is a Tesla C2050. The installed driver on the VM is the last exclusive TCC driver for the C2050. I attempted to install later drivers (which could be TCC-enabled) using RDP, but after a reboot the driver failed to install correctly in Windows. I have continued to use the old TCC driver since I was able to run, but the reboot after a ULF is a pain.