MadVR Image Doubling 720p
madVR 720p Image Doubling using NNEDI3 Luma Only
Below you can see the settings I used with madVR for these screenshots. The point of this comparison is to see how much madVR's new Image Doubling setting improves video quality with the help of NNEDI3. I included the average render time per frame so you can have some type of idea on how demanding NNEDI3 resolution doubling can be. Personally I usually keep LUMA resolution doubling around 64 neurons.
Artifact Removal - reduce banding artifacts: Checked default debanding strength: high strength during fade in/out: high scaling algorithms chroma upscaling: Lanczos 3 taps with active anti-ringing filter image doubling: Testing image upscaling: Lanczos 3 taps with active anti-ringing filter image downscaling: Lanczos 3 taps with active anti-ringing filter smooth motion: off dithering - Error Diffusion option 2 use colored noise: UNchecked change dither for every frame: Checked
I'm running a GTX 970 with an Intel i7-4790k and 16GB RAM with the latest geforce drivers 347.25.
Image Doubling Disabled
- ~3.90ms render time per frame with GTX 970, all types of madvr image doubling disabled. I used Lanczos 3 Tap with anti-ringing filter for Chroma Upscaling, Image Upscaling and Image Downscaling, and did not enable any type of image doubling for this screenshot.
Always Double Luma 16 neurons
- ~13.90ms render time per frame with GTX 970, Luma resolution doubling set to always with 16 neurons. I used the same settings as above, but this time I set Luma resolution doubling to always use 16 NNEDI3 neurons. I didn't notice a huge difference in terms of video quality, but you will notice that the average render time per frame jumped up by almost 4 times. Any type of resolution doubling for 720p video will be expensive in terms of GPU usage.
Always Double Luma 32 neurons
- ~15.60ms render time per frame with GTX 970, Luma resolution doubling set to always with 32 neurons. If you are going to enable LUMA resolution doubling, you might as well set it to at least 32 neurons since the performance hit is only a little more than with 16 neurons and I noticed the video was a bit smoother and richer. If you own a GTX 960, GTX 970, or GTX 980 you should be able to use at least 32 neurons if not more.
Always Double Luma 64 neurons
- ~17.80ms render time per frame with GTX 970, Luma resolution doubling set to always with 64 neurons. Render time per frame continues to go up as you increase the amount of neurons for NNEDI3 to use. It looks like there is about a 2ms bump in render times for each step.
Always Double Luma 128 neurons
- ~24.30ms render time per frame with GTX 970, Luma resolution doubling set to always with 128 neurons. This is about as high as I go in terms of using NNEDI3 to always double LUMA resolution. With a GTX 970 I still have about 15ms of render time to spare, this allows me to enable Error Diffusion Dithering Option 1 or 2 as well as Smooth Motion. I like to keep about 5ms - 10ms buffer just to make sure I am not dropping frames when playing smaller 720p or 480p video. Not all 720p video has the same resolution either so don't be surprised if some 720p videos play much quicker than others.
Always Double Luma 256 neurons
- ~40.10ms render time per frame with GTX 970, Luma resolution doubling set to always with 256 neurons. Unless you are using a GTX 970 or GTX 980 I do not suggest using 256 neurons for LUMA resolution doubling.
Other MadVR Pages