Hi,
I am using A100 with Multi-Instance GPU (MIG). The user guide notes select MIG device with CLI, as follows. But I want to know if there is any other way to do, like through cuda api?

image665×165 38.3 KB

it can’t be done via cuda api. The reason is that mig only exposes once instance to CUDA (read mig user guide). By the time CUDA starts up, its too late to choose another instance.