SERL is a meticulously implemented code library that encompasses an efficient off-policy deep reinforcement learning method, methodologies for calculating rewards and resetting environments, a high-quality robot controller widely adopted in the industry, and a set of challenging example tasks. It provides the community with resources detailing its design choices and presenting experimental results. Remarkably, we have found that our implementation can achieve highly efficient learning, requiring only 25 to 50 minutes of training to obtain strategies for tasks such as PCB assembly, cable routing, and object relocation. These strategies have achieved near-perfect or perfect success rates, demonstrating robustness even under disturbances and emerging recovery and corrective behaviors. We hope that these promising results and our high-quality open-source implementation will provide the robotics community with a tool to further promote the development of robot reinforcement learning.