We have observed large quality degradation when using ADAPTIVE deinterlace, i did some regression testing :
470 - not affected
510 - not affected
515 - not affected
525 - not affected
535 - affected
545 - affected
HW used: RTX 2060 Super, RTX 2070, Quadro RTX 4000, RTX 3050
There is image which will show very bad results on 535 drivers.
Bob and Weave deinterlace is not affected
Is this still a problem with 551?
Problem still present. HW deinterlace was broken more than a year ago (since 530 series) and still not fixed.
It looks like it is fixed in 550.54.14 drivers, can anyone else confirm this?
@Thunderm I can confirm it is fixed in 550.54.14
nvidia-deinterlace-problem-535.154.05-550.54.14.zip (1.2 MB)
(left: 535.154.05, right: 550.54.14)
There is other issue on 550.76 drivers, when using different viewport resolution (1920x1080 → 1280x720) and format yuv420p(tv, bt470bg, top first) with adaptive deinterlace, on latest ffmpeg it will not even start, do you need more info, or is this known issue?
Can you share exact steps to reproduce this issue and any relevant files required to reproduce? Thanks.
Used command (BROKEN since 535 drivers):
ffmpeg -drop_second_field 1 -deint adaptive -resize 1280x720 -c:v h264_cuvid -i in.mp4 -c:v h264_nvenc -b:v 1M out.mp4
Command without using different ulTargetWidth and ulTargetHeight (CORRECT):
ffmpeg -drop_second_field 1 -deint adaptive -c:v h264_cuvid -i in.mp4 -c:v h264_nvenc -b:v 1M out.mp4
ffmpeg version is current stable, but it is observed on version 3-7
Input:
Output (broken):
Detail image of broken output:
Your file has warning about 5 reference frames used in CABAC if I ffplay it, while it signals only 4 are in use. That is bad.
Stream could be somehow not perfectly valid, but anyway drivers 525 and lower will decode and deinterlace this sample without any issues.
I could check another stream 1080i50, but i think wrong REF frames is not reason for this issue. Should i check it on other stream?
5 reference frames is a different preset and profile theoretically.
Also not even allowed on blu-ray. Please test another.
Maybe 5 reference frames are even more rare for legacy interlaced content.
It is same with another correct sample as broken is deinterlace part, not decoding part.
Write if you need more tests…
This BUG is still present in current 565 drivers.
It is visible only when height is lower than in original video, eg resizing 1080 → 720/540 etc…
Hello.
I can confirm that using 565.77 this problem still exists.
Any ideas?
Let me know if you need any more info.
Still not fixed in 570.86.16…
Adaptive deinterlace + resize is still broken in 580.95.05
How to simulate:
ffmpeg-8.0.0 -deint adaptive -resize 1280x720 -c:v h264_cuvid -i in.ts -c:v h264_nvenc -b:v 1M -an -y out.mp4
I have the same problem here.
Windows Server 2025.
Also tested on Windows 11 25H2, same result.
nVidia RTX 2080 TI Ryzen 9 5950X, 128GB DDR4 4000MHz.
Click on the image to see full resolution and compare them.
Only driver was changed, all other settings player, video etc remain the same.
Studio DRIVER 528.49:
Studio DRIVER 596.36:
Look at deinterlace (inverse telecine) on fonts and other objects.
New driver doesn’t deinterlace it correctly, it just adds blocky cubes, which is a massive degrade in image quality output, like 60% worse picture quality at 50fps.
Image you’re looking at is 4:2:2 MPEG-4 40Mbit/s.
Also whole EBU has problems with eurovision song contest, creating a progressive scan output, and it has visible doubling blocks instead of pure deinterlaced, progressive like image quality that 528.49 driver achieves.
They’re also using nvidia solutions for video and blocky output is clearly visible.
I am working for media company and we have seen with new drivers output quality is super degraded using new drivers.
We need this to be fixed soon in enterprise segment (studio drivers), as for now we’re forced to use OLD drivers that are vulnerable to exploits.
When can we expect a fix that really fixes this ?
I dont think this is same issue, can you provide short sample to test (eg 30s)? As I did some internal changes in our transcoding SW and we have switched to 585 drivers as 59X were unstable and had alotof new stability problems.
Your issue is probably the same as mine, players using the EVR renderer in the R590 series drivers can’t deinterlace properly. I reported this here five months ago, but it’s still not fixed. I almost thought there weren’t any NVIDIA staff members monitoring this forum anymore.
