I’m working with OpenCL over Nvidia devices. I have one program and two machines.
The first machine is an Intel Xeon X5650 connected to two S2050 cards (driver 270.x)
The second machine is an Intel Xeon E5506 connected to two C2050 cards and one Quadro FX 380 (all with drivers 280.x)
Well, I’m executing the program single-GPU (over one of the S2050 in the first system and over one of the C2050 in the second) and I obtaining a strange
–> In the first system, over an 40%,
–> In the second system, grows to 200%.
I’ve investigating this performance and with ATI drivers occurs this. It seems that exist a bug when we are working on multi-GPU systems. In my case, both systems are multi-GPU, the only difference is that the second is ‘heterogeneous’.
Any help is welcome :)
Thanks in advance!