I’m running WinXP64 with the latest CUDA driver. I have one Geforce 285GTX and one 8500GT, which is driving the Windows display. In this configuration streams are terribly slow on the 285GTX(see output of simpleStreams below). If I switch the 285GTX to be the display driving card stream performance is quite alright.
Can anybody explain this correlation? Is it a bug?
[codebox][ simpleStreams ]
Device name : GeForce GTX 285
CUDA Capable SM 1.3 hardware with 30 multi-processors
scale_factor = 1.0000
array_size = 16777216
non-streamed: 48.10 (1015.83 expected)
4 streams: 8001.18 (265.64 expected with compute capability 1.1 or later)