Data-Level Parallelism in Vector, SIMD, and GPU Architectures - Computer Architecture: A Quantitative Approach - page 315

FIGURE 4.25 Conversion from terms used in this chapter to official NVIDIA/CUDA and
AMD jargon . Note that our descriptive terms “Local Memory” and “Private Memory” use the
OpenCL terminology. NVIDIA uses SIMT, single-instruction multiple-thread, rather than SIMD,
to describe a streaming multiprocessor. SIMT is preferred over SIMD because the per-thread
branching and control flow are unlike any SIMD machine.