# CUDA error: MPS server is not running (cudaErrorMpsServerNotReady)

- **ID:** `cuda/cuda-mps-server-unavailable`
- **Domain:** cuda
- **Category:** system_error
- **Error Code:** `cudaErrorMpsServerNotReady`
- **Verification:** ai_generated
- **Fix Rate:** 88%

## Root Cause

The CUDA Multi-Process Service (MPS) control daemon is not active, but the application attempted to connect to it via CUDA_MPS_PIPE_DIRECTORY or similar environment variables.

## Version Compatibility

| Version | Status | Introduced | Deprecated |
|---------|--------|------------|------------|
| CUDA 12.4 | active | — | — |
| NVIDIA Driver 550.54.14 | active | — | — |
| Ubuntu 22.04 | active | — | — |

## Workarounds

1. **export CUDA_MPS_PIPE_DIRECTORY=/tmp/nvidia-mps; nvidia-cuda-mps-control -d** (90% success)
   ```
   export CUDA_MPS_PIPE_DIRECTORY=/tmp/nvidia-mps; nvidia-cuda-mps-control -d
   ```
2. **unset CUDA_MPS_PIPE_DIRECTORY; unset CUDA_MPS_LOG_DIRECTORY; python your_script.py** (95% success)
   ```
   unset CUDA_MPS_PIPE_DIRECTORY; unset CUDA_MPS_LOG_DIRECTORY; python your_script.py
   ```

## Dead Ends

- **** — The daemon must be started explicitly; restarting the app alone does not launch it. (100% fail)
- **** — The environment variable only points to the daemon's socket; if the daemon is not running, no path works. (95% fail)