NVIDIA's SMs are most comparable to the 'CUs' on AMD GPUs or cores on Apple GPUs, generally speaking. The "cores" are subsets of the SM that perform individual

NVIDIA's SMs are most comparable to the 'CUs' on AMD GPUs or cores on Apple GPUs, generally speaking. The "cores" are subsets of the SM that perform individual operations, IIRC. See this diagram from an nvidia blog post: https://developer-blogs.nvidia.com/wp-content/uploads/2021/guc/raD52-V3yZtQ3WzOE0Cvzvt8icgGHKXPpN2PS_5MMyZLJrVxgMtLN4r2S2kp5jYI9zrA2e0Y8vAfpZia669pbIog2U9ZKdJmQ8oSBjof6gc4IrhmorT2Rr-YopMlOf1aoU3tbn5Q.png https://developer-blogs.nvidia.com/wp-content/uploads/2021/g... ( https://developer.nvidia.com/blog/nvidia-ampere-architecture-in-depth/ https://developer.nvidia.com/blog/nvidia-ampere-architecture... )