Getting CUDA acceleration ready for PixInsight
In PixInsight, a lot of modules are very slow. They can be sped up, and that is what we will teach you in this article. We will guide you on how to install NVIDIA’s CUDA, which makes several PixInsight modules up to 1500% faster.
How does Pixinsight benefit from GPU acceleration?
GPU (Graphics Processing Unit) acceleration is a technology which enables a computer’s CPU (Central Processing Unit) to offload certain tasks to the GPU. It can significantly improve the performance of certain types of applications, especially PixInsight which is known to be a memory (and CPU) consumer.One example of GPU acceleration is NVIDIA’s CUDA (Compute Unified Device Architecture) technology. It is a platform for programming GPUs that enables them to be used for computing tasks.
In the case of PixInsight, CUDA accelerates a lot of different modules, for example:
- StarNet++ : Star removal
- StarXterminator : Star removal
- BlurXterminator : One-click deconvolution
- NoiseXterminator : Denoise linear and non-linear images
Prerequisite
Prior to installing CUDA, we recommend updating Pixinsight to the latest version. To use NVIDIA CUDA GPU acceleration, you must have:
- CUDA-enabled GPU: Open this link, click on “CUDA- enabled GeForce and TITAN Products” and verify your GPU is in the list.
- CUDA-compatible operating system: Windows or Linux.
In this article, we will focus on the Windows operating system. If you wish to install CUDA on Linux, click here to learn how.
Speed Comparison
For the following comparisons, we used an MSI gaming laptop with the following specifications:- RAM: DDR4, 16 GBytes
- GPU: NVIDIA RTX 2060, 6GBytes
- CPU: Intel Core I7-9750H, 2.60GHz
We left the default settings for all the modules.
| Module | Without CUDA | With CUDA |
| BlurXterminator | 1min 34s | 8s |
| NoiseXterminator | 25s | 13s |
| StarXterminator | 6min 35s | 24s |
| Starnet ++ | 1min 52s | 21s |
Installation of NVIDIA CUDA on Windows 10/11
- Download the following files. Note: you must download these versions precisely, otherwise it might not work.
- NVIDIA CUDA Version 11.8 (Chose Network Installation) : https://developer.nvidia.com/cuda-11-8-0- download-archive
- NVIDIA CUDNN Version 8.7 for CUDA 11.X (You need to have a NVIDIA Developer account. It’s free, you just have to register) https://developer.nvidia.com/rdp/cudnn-download
- LibTensorFlow-GPU Version 2.7.0: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-gpu- windows-x86_64-2.7.0.zip
- ZLIB DLL : http://www.winimage.com/zLibDll/zlib123dllx64.zip
- Open the CUDA .exe installation file (setup). Deselect everything except CUDA -> Runtime -> Libraries* and CUDA -> Development -> Tools -> CUPTI
- It is important to deselect the Demo Suite in Libraries too as you will not even need it.
-
Extract the downloaded cudnn-windows-x86_64-8.7.X.XX.zip archive.
a. Copy the files of the folder ‘bin’ to the corresponding folder in ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\V11.8\bin’ - Copy the files of the folder ‘include’ to the corresponding folder in ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\V11.8\include’
- Copy the files of the folder ‘lib\x64’ to the corresponding folders in ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\V11.8\lib\x64’
- Extract the downloaded libtensorflow-gpu-windows-x86_x64-2.7.0.zip
- Copy the tensorflow.dll from ‘lib’ folder to ‘C:\Program Files\PixInsight\bin’, overwriting the existing file
- Copy the tensorflow.dll from ‘lib’ folder to ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\lib’
- Copy the files in each folder of ‘include\tensorflow’ to ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\include’
- Extract the downloaded zlib123dllx64.zip
Copy ‘zlibwapi.dll’ from the ‘dll_x64’s folder to ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\’
- Search for environment in Windows Search Barand Select Edit the system environment variables
In the 'System Properties', window click “Environment Variables” and in the new window:
- a) In ‘System Variables’ (on the bottom of the window) enter the following three new variables (click “new” button three times):
- name TF_FORCE_GPU_ALLOW_GROWTH, value= TRUE
- name CUDA_PATH, value=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
- name CUDA_PATH_V11_8, value=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8
- b) In ‘System Variables’ (on the bottom of the window), edit Path environment variable and add the following paths:
- C:\Program Files\NVIDIA GPU Toolkit\CUDA\v11.8
- C:\Program Files\NVIDIA GPU Toolkit\CUDA\v11.8\bin
- C:\Program Files\NVIDIA GPU Toolkit\CUDA\v11.8\include
- C:\Program Files\NVIDIA GPU Toolkit\CUDA\v11.8\lib
- C:\Program Files\NVIDIA GPU Toolkit\CUDA\v11.8\extras\CUPTI\lib64
How to know if it works?
If everything is working, you should see that the modules stated earlier work a lot faster than usual.
Furthermore, if you search for the application Task Manager and open it, you should also see that your GPU shoots at 100% (or around 100%) while running these specific modules. If this is not your case, you will find specific information below to troubleshoot the issue.
In case it still does not work, what should I do ?
1) Check if all above installation steps were done correctly, in the appropriate order.
2) Check if your graphics card is included in NVIDIA’s official list shared with you earlier in this article.
3) Check if you have the appropriate operating system.
4) Re-boot your hardware.
5) Re-launch PixInsight.
6) (Optional, but highly suggested) Installing the latest PixInsight update.
7) Contact us if you have done all 6 above steps, we will help you find the solution.
Conclusion
Now that CUDA is installed on your computer, you should see a huge difference. You will save a lot of time processing your images!
How much time did you save on processing with CUDA enabled ? Tell us in the comments.
Clear Skies,
the Team David Astro.
Leave a comment