# [Pixel Post] How to Set a Power Limit on My NVIDIA GPU to Keep Temperatures Under Control

My workstation is equipped with a 600W NVIDIA RTX PRO 6000 Blackwell. It's a beast — and like any beast, it runs hot. During some of my deep learning workflows, I noticed the GPU temperature creeping past 80°C. While technically still within spec, it felt uncomfortable for long training sessions, especially given the sustained power draw.

I wanted a safer and more power-efficient setup. Limiting the GPU's maximum wattage seemed like the most direct and effective route. Here's how I set it up on Ubuntu Linux.

---

### Step 1: Temporary Power Limit (Testing)

Before committing to any system-level changes, I tested the power limit manually:

```bash
sudo nvidia-smi -pl 250
```

This command sets the GPU power limit to 250 watts — a much safer ceiling for my cooling setup. You can substitute any value that fits your needs (within your GPU’s allowable range).

This change, however, only lasts until the next reboot.

---

### Step 2: Persisting the Limit with systemd

To apply the power limit automatically every time the system boots, I created a simple `systemd` service.

#### Create the service file:

```bash
sudo nano /etc/systemd/system/gpu-power-limit.service
```

Paste in the following:

```bash
[Unit]
Description=Set NVIDIA GPU Power Limit
After=default.target

[Service]
Type=oneshot
ExecStart=/usr/bin/nvidia-smi -pl 250

[Install]
WantedBy=default.target
```

Again, replace `250` with whatever value suits your use case.

#### Enable and start the service:

```bash
sudo systemctl daemon-reload
sudo systemctl enable gpu-power-limit
sudo systemctl start gpu-power-limit
```

This ensures the power limit is applied on every reboot.

---

### Step 3: Verification

After rebooting or starting the service, you can verify the new power limit with:

```bash
nvidia-smi
```

Look for the "Power Limit" field in the output — it should reflect the new value you set.

---

### Step 4: Reverting (if needed)

If you want to remove the limit or experiment with a different value:

```bash
sudo systemctl stop gpu-power-limit
sudo systemctl disable gpu-power-limit
```

Then either change the wattage in the service file or delete it entirely.

---

### Final Thoughts

Even though GPUs are built to handle high thermal loads, running them at full wattage all the time isn’t always necessary — especially during inference, fine-tuning, or other light workloads. Setting a reasonable power limit helped me maintain cooler temperatures, reduce fan noise, and worry less about long-term wear.

If you're working with high-wattage GPUs in tight or poorly ventilated setups, this is a simple but effective change worth making.

---

Let me know if you'd like a more technical version with power graphs or integration into your server bootstrap scripts.