NVIDIA Tesla P40: Windows 10 Setup and Cooling Adapter

This project documents my NVIDIA Tesla P40 24 GB setup for experimenting with AI. Because the P40 is passively cooled, I designed and printed a fan adapter to provide the airflow it needs in a desktop PC. The project includes build photos, printing details, cooling information, and the Windows 10 configuration I used.

Warning

This is a record of one working configuration, not a universal installation guide. Back up the registry before making changes and monitor the GPU temperature carefully during initial testing.

Project contents

• Windows 10 driver and registry notes
• Preview and printing details for the P40 cooling adapter
• Printing and installation details
• Photos of the printed and assembled cooling solution

Cooling solution

The Tesla P40 uses a passive heatsink intended for server airflow. This adapter mounts a Dell PowerEdge R710 fan to the card and directs air through the heatsink.

Fan -> printed adapter -> P40 heatsink -> case exhaust

Required hardware

• NVIDIA Tesla P40 24 GB
• Printed P40 cooling adapter
• Dell PowerEdge R710 fan, part number 0RK385
• 2 × M3 × 0.5 × 9 mm screws for mounting the adapter to the P40
• 2 × fan-mounting screws; exact size unknown

Printing settings

Note

The model file is not currently included in this repository. The image above is a preview of the model used for my printed adapter.

Setting	Value
Material	PLA
Layer height	0.1 mm
Orientation	P40 mounting face touching the print bed
Supports	None
Bed adhesion	Brim

The STL bounding dimensions are about 110 × 42.5 × 75.5 mm. STL files do not store units, so import it as millimetres.

Assembly and airflow

1. Mount the fan with two screws in diagonally opposite holes.
2. Mount the printed adapter to the P40 using two M3 × 0.5 × 9 mm screws.
3. Confirm that the fan pushes air into the adapter and through the card.
4. Install the card. The fan draws more power than a standard fan header can provide in my setup, so I use a pump-capable header. Check the current ratings of both the fan and the header before connecting them.
5. Check temperatures with nvidia-smi before applying a sustained GPU load.

Fan control

I currently control the fan using Libre Hardware Monitor and a custom Python script. The setup is working on my system, but the script is not yet fully stable, so I am not ready to publish it.

Test system

Component	Model
Motherboard	Gigabyte Z690 Gaming X DDR4
CPU	Intel Core i3-12100
Memory	32 GB Patriot Viper DDR4-3467
GPU	NVIDIA Tesla P40 24 GB
Storage	Samsung 970 EVO 500 GB
Power supply	Corsair RM750x
Operating system	Windows 10 Pro

Windows 10 setup

This configuration used the NVIDIA GRID 511.65 driver. Of the several drivers I tested, this was the only one that worked. Newer drivers and different hardware combinations may behave differently.

Registry configuration

1. Open Registry Editor.

2. Export a registry backup before making changes.

3. Navigate to:

   Computer\HKEY_LOCAL_MACHINE\SYSTEM\ControlSet001\Control\Class\{4d36e968-e325-11ce-bfc1-08002be10318}
   

4. Check the numbered subkeys such as 0000 and 0001. Find the one whose DriverDesc value identifies the Tesla P40. This is especially important if another GPU was previously installed or if multiple GPUs are currently installed.

5. In that subkey, create or update the following value:

   Name	Type	Value
   EnableMsHybrid	DWORD (32-bit)	1

This removed the unwanted virtual display in my configuration.

An AdapterType value of 1 was also present in the working P40 registry subkey as a DWORD (32-bit) value. After testing several different configurations, I cannot confirm whether I set this value manually or whether it is required.

Checking the GPU

Run:

nvidia-smi

If nvidia-smi was not added to PATH, call it using its full path in PowerShell:

& "C:\Program Files\NVIDIA Corporation\NVSMI\nvidia-smi.exe"

Example output from the working system:

The screenshot shows the Tesla P40 detected in WDDM mode with driver 511.65 and CUDA 11.6. At the time of capture, the card was idle at 34°C and drawing 11 W.

Use the reported temperature to verify that the fan and duct provide sufficient airflow, especially under load.

Further reading

The Level1Techs forum thread Gaming, on my Tesla, more likely than you think contains extensive community discussion about using Tesla datacenter GPUs in desktop systems.

Known limitations

• The instructions were tested only with the system and driver listed above.
• The exact fan-mounting screw size is not documented.
• Registry locations and required values may differ with another driver or GPU configuration.
• The cooling adapter increases the overall length of the card; check case clearance before installation.