cuda driver api vs. runtime api

can cuda driver api fully be covered by cuda runtime api? basically for upper level, like cuBLAS, they intends to call cuda runtime api instead of cuda driver api?

I’m not sure what you mean by fully covered. The Runtime API has everything you need to start coding a program.

The Driver API offer much finer granularity, meaning there are some low-level functionality in the Driver API not available in the Runtime. Most programs can get very good performance with the Runtime API.