Originally Posted by T-Bone
On that note, when I was doing 1080p to 2160p upconvert with Blu-ray material, I found it very difficult to tell the difference between super XBR and the ngu algorithms.
With moving video, I was hard-pressed to tell the difference between the two. Super XBR gives me between 10 and 13 millisecond render times. The ngu algorithms give me about twenty one millisecond render times but it makes my fan on the card turn on.
So lately I've been using just the super XBR. Seems like it saves a lot on performance. And it is a little quieter. If I had a passive card and the fan didn't come on I probably wouldn't mind as much.
I used super-xbr for a year and thought it looked pretty good. But it does ring quite a bit compared to NGU. Not everyone will notice, but I did.
Chroma upscaling beyond Bicubic60 + AR or Lanczos3 + AR is hard to notice without the right test image. One of my displays is 4:2:2 and the other is 4:4:4, but it doesn't seem to make that much of a difference.
Considering players like Kodi have Lanczos3 without an anti-ringing filter as its best scaler, madVR doesn't have a ton of competition on the image scaling front. Most players have to cater to low-powered hardware. Anything beyond Lanczos3 + AR in madVR looks good by comparison. And other useful shaders in madVR are often ignored.
You get what you pay for; at least, on the hardware front. A good example is the Nvidia Shield. It is perhaps the best streaming box because it is the only one that will pass HD audio, but it is known for poor image quality. Even if you set the display to RGB Full, it will pass through every 10-bit, 4K source as Y'CbCr 4:2:0. It can't do colorspace conversions from BT.709 to BT.2020. And it appears to use Bilinear scaling, which would explain the "fuzzy" image users complain about. One user in this forum even complained that 4K UHD sources won't play smoothly, adding a soap opera effect. Considering the Shield is $250 or so where I come from, this isn't completely surprising. You get what you pay for...