The engine for nvidia is in the works.

But, in the meantime, you can try the native CPU engine. It is as fast as the underlying ATLAS. Almost no overhead.