Compilation Phases

The CUDA Compilation Trajectory

GPU Compilation

sm architecture
sm_50, sm_52 and sm_53 Maxwell support
sm_60, sm_61, and sm_62 Pascal support
sm_70 and sm_72 Volta support
sm_75 Turing support
sm_80, sm_86 and sm_87 NVIDIA Ampere GPU architecture support
sm_89 Ada support
sm_90, sm_90a Hopper support

Two-Staged Compilation with Virtual and Real Architectures
Two-Staged Compilation with Virtual and Real Architectures

compute architecture
compute_50, compute_52, and compute_53 Maxwell support
compute_60, compute_61, and compute_62 Pascal support
compute_70 and compute_72 Volta support
compute_75 Turing support
compute_80, compute_86 and compute_87 NVIDIA Ampere GPU architecture support
compute_89 Ada support
compute_90, compute_90a Hopper support

Just-in-Time Compilation of Device Code
Just-in-Time Compilation of Device Code

Using Seperate Compilation in CUDA

Reference