Recent content by MrSpiggott

Sorry about that. I was confidently banging on about an untested theory, which is never a good idea. It scales in all directions so I was expecting the interference patterns to show marked differences between a scaled and unscaled image even when it's not animated (luminance cycling). @Shifty...

I've remembered a web site I saw a few years ago that generates full screen test patterns at whatever resolution you've set your output to. The last picture in the contrast and brightness section is zone plate (high frequency circles), which should nicely test your scaling. It is an .ru site...

I forgot that consumer oleds do that, but for the purposes of diagnosing @Cyan scaling problem, it's better to turn it off.

That's interesting. Why do you have black borders? I think the reason you can't see a difference between the two images and resolutions is because you never actually see an unscaled image. Probably something on your PC is modifying the output somehow. When outputting the 3840 image at 3840...

In that case you're only scaling vertically, and the image data doesn't change in that axis. Therefore no artifacts.

Somethings wrong, you shouldn't see any blocks or stripes if you're outputting the 3840 image at 3840 resolution. You are looking at it borderless full screen? Do you have overscan enabled? That'll discard the outside edges of the image and cause a slightly scale to 3840. It's a legacy...

No, I work in broadcast so use hardware test pattern generators. Many years ago when interfaces were still analogue, there used to be plenty of freeware PC monitor testing software that generated patterns, but I'm having difficulty finding any now. Shifty's .png image example is fixed...

Downscaling to is much trickier than upscaling where you have extra destination resolution to help in the process, particularly when using AI upscaling. In this case, 4096 lines of generated video have to be squeezed onto 3840 physical lines. You are going to lose clarity on high frequency data...

Your 4K TV's native resolution is 3840 x 2160 which is four 1920 x 1080 screens worth of pixels. (The first 4K TV I used required four 1080 1.5Gb BNC inputs to get an image onto the screen). Anything other than this resolution will require scaling by the TV and I doubt much care or thought has...

They had no reason to be excellent quality, just good enough. The analogue output stage of an arcade PCB was far simpler, and much cheaper, than broadcast spec' equipment's equivalent; so why hook up an expensive monitor. They just had to be durable. Hidden in the blur and we all used much...

Memory at the time, both working and storage, was a real cost bottleneck. So think of it like a form of compression. You draw the image as if it's 480 and then throw half the lines away. You then let the natural blurriness of a composite signal trick your mind into filling in the gaps like it...

Halves the frame buffer size. Actually, halves the memory required is probably a better description.

Although not an official standard, 240p (or numbers thereabouts depending on region) did exist. It was where the gun was forced to write to field 1 again, after fly back, rather than alternating between field 1 and field 2. Exactly the same number of fields were being written a second, but the...

The Tensor cores are there predominantly for non gaming tasks such as machine learning. They are fitted as standard, for which you're paying a premium, so may as well be used for upscaling even if it's only for a fraction of a frame. What's the argument for there inclusion on a console?

Because they’re vital for high performance machine learning applications outside gaming. I seem to remember their adoption in game upscaling was along the lines of Nv thinking, “we’ve got these cores, what can we use them for in a gaming system?”