Ubuntu 16.04下POWER9架构CUDA配置问题求助

Let’s work through this problem step by step—since you’ve already confirmed the file exists and is readable, we can focus on the most common pitfalls that trip people up with systemd service files.

1. Double-check the command you’re running #

Systemd service files aren’t executable scripts, so trying to run the file directly (like ./nvidia-persistenced.service) will throw an error. Instead, you need to use systemctl commands to interact with it. First, make sure the file is in the correct systemd directory (preferably /etc/systemd/system/), then run these commands:

# Reload systemd to detect the new service file
sudo systemctl daemon-reload
# Enable the service to start on boot
sudo systemctl enable nvidia-persistenced.service
# Start the service immediately
sudo systemctl start nvidia-persistenced.service

2. Validate the service file’s syntax #

A tiny typo in the service definition (like a missing bracket or misspelled directive) can break everything. Run this command to check for syntax errors:

sudo systemd-analyze verify /path/to/your/nvidia-persistenced.service

It will flag any issues directly—fix them if you see error messages.

3. Check file permissions and location #

Systemd requires specific permissions to read service files. Run this to confirm:

ls -l nvidia-persistenced.service

You should see permissions like -rw-r--r-- (read/write for owner, read-only for others). If permissions are too restrictive, fix them with:

sudo chmod 644 nvidia-persistenced.service

Also, make sure the file is in one of systemd’s recognized directories:

• /etc/systemd/system/ (for custom user-added services, this is the best spot)
• /lib/systemd/system/ (for package-provided services)
  If it’s elsewhere, move it to /etc/systemd/system/ or use the full absolute path in your systemctl commands.

4. Dig into error logs for specifics #

If the service fails to start, get detailed context with these commands:

# Check the current service status
sudo systemctl status nvidia-persistenced.service
# View full journal logs for the service
sudo journalctl -u nvidia-persistenced.service -xe

These logs will tell you exactly what’s wrong—whether it’s a missing nvidia-persistenced binary, incorrect user/group settings in the service file, or a dependency that hasn’t been met.

5. Confirm the nvidia-persistenced binary exists #

The service file’s ExecStart directive points to the nvidia-persistenced executable. Verify it’s installed and accessible:

which nvidia-persistenced

If this returns nothing, you may need to add CUDA’s binary directory to your PATH, or update the ExecStart line in the service file to use the full path (e.g., /usr/bin/nvidia-persistenced).

内容的提问来源于stack exchange，提问作者raggot