Event Sampling in Tesla GPU

Question

I am trying to work with the CUPTI library in CUDA to read some performance counters like instructions executed, memory access etc. When executing the sample provided with CUPTI installation (/usr/local/cuda-5.0/extras/CUPTI/sample/event_sampling) it says 'Event sampling not supported for Tesla family devices". (I am using a GTS 250). Does that mean I won't be able to read the counters during application execution or is it something else?

Thanks

Answer 1

You might want to read the cupti documentation. The event sampling API is separate and distinct from the metric API. The cupti metric API can be used to read counters associated with the metrics identified as supported on the particular compute capability you are running on. For example, the supported cc 1.x counters are identified here.

The error message you received indicates that the event sampling functionality (API) is not supported on devices of compute capability less than 2.0 Tesla as used here was an NVIDIA codename for a certain family of GPUs that predated the Fermi (cc 2.0/2.1) generation.