Difference between cuda core & streaming multiprocessor

What’s the difference between a cuda core & a streaming multiprocessor?

Are they interchangeable terms?

My only understanding is that a SMP manages thread execution through wraps (please correct me if that’s wrong!)


Each CUDA core is also known as a Streaming Processor or shader unit sigh

The streaming multiprocessor (SM) contains 8 streaming processors (SP). These SMs only get one instruction at time which means that the 8 SPs all execute the same instruction. This is done through a warp ( 32 threads ) where the 8 SPs spend 4 clock cycles executing a single instruction on multiple data (SIMD).

Consider the whole GPU to be a couple of SIMD units… Nvidia calls it SIMT ( Single Instruction Multiple Threads)