Max avatar

Hi All,

Looking for feedback and guidance on GraXpert Denoising in PixInsight. During my most recent processing session, one image took 49’37” to go through the process. Other images took 30’+ as well. This seemed excessive from my previous processing sessions. Strength is at 0.80, “Latest” AI model, batch size 4, and GPU acceleration is active.

Am I doing something wrong? Should I check any settings to get better performance?

Thanks and clear skies!

System specs:

  • i9-13900K

  • RTX 4090 24GB

  • 128GB DDR5

  • 8TB NVME

  • Windows 11 :(

    📷 Screenshot 2025-09-23 142129.pngScreenshot 2025-09-23 142129.png

Well Written Respectful
Die Launische Diva avatar

Hi,

The image you are trying to denoise is huge. Is this a mosaic?

Tony Gondola avatar

I use Graxpert all the time and those times really don’t seem to be unusual.

Max avatar

No mosaic. I did a couple nights with my my full frame and RedCat 51 for wide-field data. Definitely under-sampled like crazy, but I wanted to see the results.

Image size would help explain the delta from the other images I’ve processed.

Die Launische Diva avatar

I typically work with 18 Megapixel images and GraXpert takes a couple of minutes. You are working with a ~231 Megapixel image so it is expected to take at least more than a couple of tens of minutes, isn’t it? 🙂

Well Written Concise
Ruediger avatar

This looks quite long to me. Just ran a test on my machine (doing many other things parallel). I’d suggest you are verifying that the GPU is really used and to what extend.

Based on my test (half width and height) your result should be 4 times as much. Something around 10 Minutes.

My system: Win10, 13th Gen Intel Core i9-13900KS, 64GB Ram DDR5, NVIDIA GForce RTX 4070Ti.

CS
RÜdiger

📷 2025-09-24_11h26_02.png2025-09-24_11h26_02.png

📷 2025-09-24_11h24_03.png2025-09-24_11h24_03.png

Russell S avatar

I think that is probably about right for cpu - so your gpu acceleration isn’t working. do a test run like and look at task manager and go onto the gpu section to see if it is being utilized (like in the image in the post above).

Max avatar

Hi Russell and Rüdiger,

Thanks for the replies. I’ve watched my hardware performance and my GPU is at 90-95% for 3D and dedicated memory. CPU is at idle and main memory isn’t being utilized too much during the process.

Next data set I process, I will turn off hardware acceleration and see how long one image takes. Also, I’m planning to use GraXpert outside of PI to see if I get similar performance.

Max

Well Written
Aloke Palsikar avatar

My experience of directly using Graxpert or through Siril has been good. Typically takes 3-5 min using the updated AI algorithm. I am currently using the Beta version of Graxpert which has Object and Stars Deconvolution option as well and this is also working fine for Denoising.

Pls check if the source stacked file is corrupted