Different programs on Tesla GPUs?

Hi All,

Is it possible to send different programs to each GPU on a Tesla?
Or is it an SPMD model across the 4 GPUs as well, running a single program on different data?


In a Tesla S1070 (which has 4 separate GPUs in a 1U enclosure), each GPU is completely independent, so you can run (by default must run) different program instances on each one. The Telsa C1060 only has a single GPU, and can’t run more than a single program at a time.