We wrote this glossary to solve a problem we ran into working with GPUs here at Modal : the documentation is fragmented, making it difficult to connect concepts at different levels of the stack, like Streaming Multiprocessor Architecture , Compute Capability , and nvcc compiler flags .