EnvPool is a C++ parallel execution engine that provides vectorized environment execution for reinforcement learning agents. It uses pybind11 for Python bindings and implements a lock-free thread pool for maximum throughput, achieving approximately 1M FPS with Atari games and 3M FPS with MuJoCo on DGX-A100. The project supports Gymnasium and dm_env interfaces in both synchronous and asynchronous modes, covering Atari, MuJoCo, classic control, Box2D, and ViZDoom environments.