I was reading about Nividia's Fermi GPU architecture. [https://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf] It seems somewhat confusing strange because they seem to use slightly different terminology for what is here called simultaneous and interleaved multi-threading?

From my understanding of GPU architecture, Fermi and GPUs in general don't use simultaneous or interleaved multi-threading. The cores are grouped up into some number of streaming multiprocessors, but each individual core runs a single thread at a time. Threads are scheduled as warps (groups of 32 threads) executing the same code, and these threads run simultaneously, but each thread is executing on its own core without SMT

To clarify on the previous comment, I was referring to SMT at the level of a CUDA core as opposed to an SM, in the sense that it isn't SMT in the same sense as a CPU sharing the same core resources for multiple threads