I found out that using my Nvidia Quadro A4000 Gigapixel AI produce more blurry images, also they are offset down by ~2pixels, compare to original (bicubic scale) or processing with CPU or intel GPU.
So I made some tests using CPU, Intel GPU HD 630 and Nvidia Quadro A4000 using low and medium memory settings, but since difference between low and medium not so noticeable, will only post all 3 images with medium memory settings.
Other settings used are : AI model Standard, Suppress noise and remove blur 50, Reduce color bleed and Face refinement are OFF.
On provided examples it is clearly seen that using CPU and Intel GPU Gigapixel scale and process the image, but using Nvidia Quadro it is scale, but not process image or process is not so good (sharpening, removing blur and so on)
My PC config is CPU i5-8500, 32GB of RAM, Nvidia Quadro A4000
Gigapixel AI 5.6.1 Trial with enabled model downloading (models were downloaded before processing images).
Observations I found. When using old version of Gigapixel 4.4.5, images processes by Quadro A4000 looks much better than on new version 5.6.1. They very close to the images processed by CPU and iGPU.
So something broken on new version.