AV1 encoding with Resolve and Nvidia RTX 4070

For anyone into such things, here's a bit rate and bit depth test with AV1 from an Nvidia GPU with Resolve.

While the source material lends itself to low bit rates anyway, due to lack of HF and temporal complexity and a lot of bokeh. The low end "sweet spot" is about 5Mb/s and this is 4K/DCI at 24FPS. I've also tested at lower than 5Mb/s and the quality is easily good enough for previews.

There's a bunch of files in two folders, 4K and HD. It's the same one minute timeline, just tested at different low bitrates.

The files are very small, their file size info should show up in the UI next to the file names.

Also, bear in mind that this is hardware acceleration that's faster than realtime. With an in-depth, slower software solution there's a potential to go even smaller. Plus, this is 4K/24 at a lower bit rate than what your average DVD used to be and that was only SD. Considering that this flavour of AV1 is inter-frame, just as the MPEG2 flavour was for DVD. It's absolutely mind blowing just how efficient these GOP based codecs have become.

https://www.mediafire.com/folder/xsfjqlcxo32ri/AV1+TEST


I've now added a video about the workflow. It's really long and boring but there's chapters to the main bits. There's also individual YouTube uploads for each of the bitrates that I tested.

Workflow:

10000Kb/s encode/upload

5000Kb/s

4000Kb/s

3000Kb/s

2000Kb/s

1000Kb/s

Impressive, indeed!

I did a bunch of export testing when I got my new PC a couple months ago.... AV1 is it for me going forward.
Testing was geared towards comparing codec/quality/file size.
Compared h.264/h.265/DNxHR HQ/AV1 with many different settings and compared mostly higher Kb/s>5000.
At similar Kb/s - file size, AV1 is considerably better than h.264 / a fair amount better than h.265 / similar to DNxHR HQ at Kb/s>20,000 with massive file size savings.

I'm interested to hear discussion on AV1 - ProRes is great and all - but for 'non-Hollywood needs' etc - AV1 is really good. You know for the masses like me. :=D
Lots of AV1 export settings in the Resolve deliver page.
Not a lot of info/direction on their usage.
The 'Rate Control' drop down seems to be of particular importance/investigation/discussion/testing.
In my testing, I exported many methods at high bitrate Kb/s>20,000-40,000 - adjusting each method to get same file size.
Some of them behave like the Resolve Native encoder ie. wonky kb/s vs. 'Restrict to' vs Automatic.

For my export needs, AV1 'Variable Bitrate' and 'Restrict to' 80,000 is virtually identical to the original.
40,000 Kb/s for YouTube upload etc.
Export settings:

export 2.jpg (82.48 KiB) Viewed 8129 times

What NVIDIA GPU did you use Dave ? I am thinking of updating my 1080Ti so lowest cost to get AV1 etc is of interest. The 4070 or 4060Ti with 16G is on the list.

What NVIDIA GPU did you use Dave ? I am thinking of updating my 1080Ti so lowest cost to get AV1 etc is of interest. The 4070 or 4060Ti with 16G is on the list.

Hi Ron, it’s the 4070.

I’m not sure how much better the 16GB would be on the new 4060 Ti, as there’s also an appreciable difference in the core GPU processing between it and the 4070. It all depends on what Resolve favours more on X86, GPU cores (ability) or VRAM.

BTW. Pay no attention to any of the BS YouTube videos that are bleating on about how bad the 4070 is. The only thing that’s bad about it is its price, which has been exactly the same for every new GPU for the last couple of years or so. The actual GPU itself is very impressive. It’s roughly on par with a standard 3080 but with newer tech, which I’m not sure Resolve takes account of yet.

Best bit for me is that the 4070 only maxes out at 200w. Which seems a strange thing to say when an entire M1 Max computer probably averages a quarter of that during fairly heavy use. All in all on a 13600K based machine, average power draw for average timeline use will fluctuate wildly between about 150w to 400w and will go up slightly more with seriously heavy stuff going on, obviously dependent on whatever else is in your system. However, this is actually quite efficient for an X86 machine.

I did a bunch of export testing when I got my new PC a couple months ago.... AV1 is it for me going forward.
Testing was geared towards comparing codec/quality/file size.
Compared h.264/h.265/DNxHR HQ/AV1 with many different settings and compared mostly higher Kb/s>5000.
At similar Kb/s - file size, AV1 is considerably better than h.264 / a fair amount better than h.265 / similar to DNxHR HQ at Kb/s>20,000 with massive file size savings.

I'm interested to hear discussion on AV1 - ProRes is great and all - but for 'non-Hollywood needs' etc - AV1 is really good. You know for the masses like me. :=D
Lots of AV1 export settings in the Resolve deliver page.
Not a lot of info/direction on their usage.
The 'Rate Control' drop down seems to be of particular importance/investigation/discussion/testing.
In my testing, I exported many methods at high bitrate Kb/s>20,000-40,000 - adjusting each method to get same file size.
Some of them behave like the Resolve Native encoder ie. wonky kb/s vs. 'Restrict to' vs Automatic.

For my export needs, AV1 'Variable Bitrate' and 'Restrict to' 80,000 is virtually identical to the original.
40,000 Kb/s for YouTube upload etc.
Export settings:

export 2.jpg

Yes, AV1 is going to be great for certain deliverables once it’s become mainstream.

I’ve done some stuff with complex 4K/60 gameplay captures and am able to go very low with the bit rate. In fact, to the point where I had to double check the output file size and make sure of the bit rate, as it looked too good.

Try using very slow with full two pass and reduce your bit rate. Depending on the scene complexity, it’s possible to get visually close to the original with even lower bit rates.

I did a bunch of export testing when I got my new PC a couple months ago.... AV1 is it for me going forward.
Testing was geared towards comparing codec/quality/file size.
Compared h.264/h.265/DNxHR HQ/AV1 with many different settings and compared mostly higher Kb/s>5000.
At similar Kb/s - file size, AV1 is considerably better than h.264 / a fair amount better than h.265 / similar to DNxHR HQ at Kb/s>20,000 with massive file size savings.

I'm interested to hear discussion on AV1 - ProRes is great and all - but for 'non-Hollywood needs' etc - AV1 is really good. You know for the masses like me. :=D
Lots of AV1 export settings in the Resolve deliver page.
Not a lot of info/direction on their usage.
The 'Rate Control' drop down seems to be of particular importance/investigation/discussion/testing.
In my testing, I exported many methods at high bitrate Kb/s>20,000-40,000 - adjusting each method to get same file size.
Some of them behave like the Resolve Native encoder ie. wonky kb/s vs. 'Restrict to' vs Automatic.

For my export needs, AV1 'Variable Bitrate' and 'Restrict to' 80,000 is virtually identical to the original.
40,000 Kb/s for YouTube upload etc.
Export settings:

export 2.jpg

Alright, Jack.

If you're interested, I've now added a video about the workflow. It's really long and boring but there's chapters to the main bits. There's also individual YouTube uploads for each of the bitrates that I tested.

Cheers, Dave.

Workflow:

10000Kb/s encode/upload

5000Kb/s

4000Kb/s

3000Kb/s

2000Kb/s

1000Kb/s

Cheers,
Dave.

My tests showed no significant change in quality between Av1 and h.265 on Youtube at reasonable bitrates

https://www.reddit.com/r/videography/co ... lity_vmaf/



This was with libaom-av1 as I don't have any hardware that can do av1

At inadequate bitrates AV1 is better, but that is like asking if getting stabbed in the foot hurts less than getting stabbed in the hand, they both suck.

My tests showed no significant change in quality between Av1 and h.265 on Youtube at reasonable bitrates

https://www.reddit.com/r/videography/co ... lity_vmaf/



This was with libaom-av1 as I don't have any hardware that can do av1

At inadequate bitrates AV1 is better, but that is like asking if getting stabbed in the foot hurts less than getting stabbed in the hand, they both suck.

Then you have completely missed the point.

You may as well say that there is no significant difference between MPEG2 and AV1 if you encode them both at a very high bitrate that favours MPEG2.

The whole point to using AV1 for delivery is to get the lowest possible bitrates and ultimately the smallest file sizes and therefore the lowest bit-stream while occupying the smallest footprint on storage medium, while maintaining the best possible visual quality.

You say "Inadequate bitrates", what a load of BS. It's only inadequate if you're talking about H.265. AV1 clearly beats the crap out of H.265 when using low bitrates. So H.265 is obviously inefficient by comparison.

I just took a look at your BS post on reddit and all you have proven with that is your absolute ignorance with regard a whole bunch of things.

You don't even address the obvious differences between inter-frame and intra-frame codecs. You very conveniently bypass the whole issue of what filters/decoders are being used and how they are not the best options for your BS test. Plus, where was all the information about bit-depth and chroma subsampling?

Did I miss it, but where are the download files for your source files and your encodes so that others can compare and confirm what "your" results are saying?

Seriously, judging a YouTube output for anything meaningful You could upload an uncompressed YUV master and a ham sandwich and YouTube would make them look the same, and you've given no information about YouTube's filtes. That's why my examples are available for download, so people can make a proper judgement fro themselves. Plus, they can download the sources from BMD. After all, this is a Resolve forum by BMD isn't it?

You also say "But if your goal is the best quality on YT, you should not be uploading a 20mbit file" But you don't mention if that's bitrate or files size, although, the way you've expressed it, that would be file size. Regardless, you give absolutely no context. If you meant 20Mb/s, you never gave any indication of temporal complexity of the source. What frame rate are you talking about? What resolution? What codec? What encoder? What settings for the encoder? etc. etc. etc.

I've demonstrated that two files that are a half or a quarter of 20Mb/s look fantastic and visually indistingushable from the RGB uncompressed master, "even" after YouTube's compression. So your 20Mb/s may well be some random number that you plucked out of thin air.

As for your analogy. Stabbing yourself in the hand or the foot? I don't know what type of video encoding work you do, but if it involves impaling any of your appendages. Then I suggest you find yourself a new encoder

For anyone who comes across this post, take it for what it is. It's simply a demonstration of how well Resolve and the Nvidia 40 series NVENC encoder work for super fast AV1 encoding, which also look great at low bitrates.

Or. You could always read too much into what some guy with a knife fetish, who obviously doesn't have 30 years of video/audio post experience behind him and who didn't start his professional post career encoding and authoring for VCD and CDV and every popular global delivery standard for home media thereafter, who thinks that posting flawed BS on reddit means anything to anyone outside of like-minded deludeds, has to say

Seriously, judging a YouTube output for anything meaningful You could upload an uncompressed YUV master and a ham sandwich and YouTube would make them look the same, and you've given no information about YouTube's filtes.

Correct, that is the entire point. I don't care what the file looks like on your screen. I care about what it looks like on youtube. That was the entire premise of my test.

What frame rate are you talking about? What resolution? What codec? What encoder? What settings for the encoder? etc. etc. etc.

FPS and Resolution is irrelevant. The question is will AV1 fundamentally provide you better quality on YT that H.265.

Yes a 60fps file will need more bitrate than a 24fps, but this is the case for H.265 and AV1. And 4k will need more than 1440p.

I did not seek to provide the 'Best' settings for X fps and Y resolution. I did test to that effect a bit in my prior post I linked to at the beginning;

https://www.reddit.com/r/videography/co ... &context=3

Codec, I don't have access to test Nvidia vs AMD vs Intel for Av1 or H.265. There is certainly the potential some are worse than others.

Settings. I did provide those in the ffmpeg commands I used. For Av1 I did select "cpu-used 8", is the lowest quality, fastest mode. Even then it is 1 fps on my 5800h.

There might be the potential that slower will be better. But I doubt we should expect using slow for 40mbit would be able to improve the results better than fast for 84mbit. and those were only 0.8 points apart to begin with.

I've demonstrated that two files that are a half or a quarter of 20Mb/s look fantastic and visually indistingushable from the RGB uncompressed master, "even" after YouTube's compression. So your 20Mb/s may well be some random number that you plucked out of thin air.

My tests produced a VMAF of 88.66 for 20 mbit with AV1. Clearly better than the 85.09 that H.265 produced at the same bitrate.

But I would not choose either of them as 40 mbit produced results of 91.2 and 90.14.



If you are in the situation where your upload speeds are limited or have a data cap, yes 20mbit AV1 could be a great choice.

If not limited in such regard don't stab yourself in the foot by limiting your quality to an artificially low bitrate.


Is there any disadvantage to AV1, No, if you have hardware that can run it fast.

Is there any improvement in YT video quality using AV1 vs H.265, No. Both will give mediocre results.



I did yt-dl your clips to compare with the originals you linked too, but the duration is different, YT versions have title screen the others do not. And strangely the YT files are 3840x2026 and the originals are 4096x2160. So you can't overlay them or VMAF.

Do you think the average person watching YouTube on their phone, tablet or HD monitor on their PC is going to tell the difference? AV1 for me gives the opportunity to store my files at smaller sizes than keeping the originals. Especially converting old tapes for archiving. The option was h265. Also the speed of NVIDIA encode is important. I have no interest in uploading to YouTube.

Is there any disadvantage to AV1, No, if you have hardware that can run it fast.

Is there any improvement in YT video quality using AV1 vs H.265, No. Both will give mediocre results.



I did yt-dl your clips to compare with the originals you linked too, but the duration is different, YT versions have title screen the others do not. And strangely the YT files are 3840x2026 and the originals are 4096x2160. So you can't overlay them or VMAF.

You really need to take a step back and have a serious rethink about what you are saying and also start using the correct expressions and terminology.

Your whole premise is about “better” quality. Nothing is better quality, regardless of codec. You’re not going to magically improve anything beyond the original.

You clearly didn’t compare anything to any originals.

You don’t seem to understand the difference between subjective and objective.

You obviously don’t understand the difference between UHD and DCI.

You don’t understand what YouTube is doing with the VAR, or does it?

You don’t understand that the downloadable files and the YouTube uploads have the exact same encoder settings. The only difference is one has a start title for identification and the other doesn’t.

It wouldn’t surprise me to know that you are not downloading the YouTube streams “properly” anyway.

You clearly think that you’ve got a method for comparison, which is very likely something that you’ve copied from something that you’ve seen someone else do. Likely in some bogus post on a platform that requires no technical standard or proof by someone with absolutely no experience in what they are talking about, such as Reddit Which is why “your testing methodology” all goes to pot when presented with anything that doesn’t fit.

Again, for anyone casually browsing the forum who comes across this post.

This post and it’s content are simply just examples of one way to use Resolve to generate a video output that is as small as it can be for a given resolution and frame rate, that maintains high visual fidelity. This example wasn’t specific to YouTube, which is why I supplied downloadable files.

This post was not comparing one delivery codec to another.

I don’t usually engage in responding to comments that are either by trolls or by misguided “know it alls”. However, I do sometimes engage when a comment is made that can confuse the viewer who may not have the experience to differentiate the truth in what’s being said. I do this to keep the point on track and clear, so confusion, deliberate or otherwise, doesn’t get in the way of the point or example being made.

As for YouTube. I’ve posted almost 2000 videos on YouTube across multiple channels, my own and for others. Out of that almost 2000, well over 1200 are visible, a few of hundred are private and about another 300 or so private videos that are very specific video tests for many different scenarios testing codecs, scaling, bit-depths, bitrates, scene/content complexity etc. Coupled with over three decades working professionally in audio and video post, within music, TV and cinema and having one of the very first free to view video streaming websites in the UK almost 20 years ago. I’d say I have a little bit more than an average practical understanding of video delivery.

For anyone new to all this stuff. Don’t get bogged down with meaningless statistics and “so called” tests in random posts on the likes of Reddit by people who clearly have no industry experience. Beyond some common sense basics, just let your eyes be the judge of the end result. This is exactly what the viewer does and they have no interest, nor should they, as to what codec, bitrate blah blah blah was used.

Remember, most online streaming videos are viewed on tiny phone screens, then tablets, then laptops, then desktop systems with the least viewed playback platform being a good TV. Do you seriously think that anyone watching on most of these devices can tell the difference between anything?

Plus, if your outputs are more for your own personal archive and for sharing with friends and family. Other than yourself, it’s so unlikely that anyone is going to appreciate your time and effort anyway or what settings you’ve used and they shouldn’t have to just to enjoy whatever the video content is.

Now that’s not to say that you shouldn’t take pride in your work and not bother about doing the best you can for your final delivery. But don’t waste time getting all bent out of shape about the minutia that know one cares about. Put that time back into what matters more, your content production.

Getting back to this forum and this post. For anyone new to video production and using Resolve, BMD have a lot of excellent teaching materials. There’s also a number of good YouTubers out there and other online resources that may be helpful.

The point of this post and the videos was simply to give people an idea of one specific use case scenario for using Resolve with a specific hardware encoder and a specific codec. While the videos themselves and my super awesome presentation method may not be for everyone. The workflow and what I’m doing to achieve the end result, maybe useful beyond the subject matter, as what I’m doing is applicable to many other uses of Resolve.

For anyone interested in such things. I will start posting more on this forum with regard tests etc. that have some bearing on using Resolve. Again, you don’t have to like me or the content but the workflows and methodology in the videos will be useful for others using Resolve. Plus, I generally also post streaming examples and quite often downloadable examples. Which I tend to find more useful than graphs

Do you think the average person watching YouTube on their phone, tablet or HD monitor on their PC is going to tell the difference?

Wrong target.

Cater to the videophile watching from a $10,000 ISF calibrated projector on a 110" screen.

If it looks good for him, it will look good for everyone.

(But the reverse may not always be true.)

Do you think the average person watching YouTube on their phone, tablet or HD monitor on their PC is going to tell the difference? AV1 for me gives the opportunity to store my files at smaller sizes than keeping the originals. Especially converting old tapes for archiving. The option was h265. Also the speed of NVIDIA encode is important. I have no interest in uploading to YouTube.

Well said, Ron.

Aside from YouTube. I watch the vast majority of Prime, Disney and Netflix on an 11” iPad screen. Granted, it’s a very nice screen but I wouldn’t be able tell certain things apart on that size screen. Even my master files mostly get viewed on my laptop and iPad screen.

If you do go for a 40 series GPU for your Windows machine, I suspect you’ll be very happy with AV1. H.265 has been awesome for archive but when those drives start filling up, and they will and do. Getting more videos on the same drive using AV1 at the same visual quality as H.265, will make a load of sense.

I suppose the obvious advantage for such high quality video files at these small sizes, is something that’s maybe appreciated more by those who’ve been doing this for a while. Going from single field capture to MPEG1 mastering through to laying back to a DV deck and the various beta formats, through to the excellent MPEG2 codec that Canopus had for Edius and all the other acquisition formats and mastering formats up till now. AV1 is simply the evolution of where we are up to and it is very good.

Just as a side point. Although I’ve been on Mac properly now for almost a year and a half. AV1 and that new Windows system I put together, has got me thinking about moving back for Resolve. Although the Windows machine is technically a lot more powerful than the M1 Max for Resolve. I’ll have to run some tests over the next few weeks to see how certain things in my typical workflows compare. That said, I’ve already noticed some subtle and not so subtle differences between them that’s given me a bit more appreciation for just how well Resolve works on Apple Silicon (ARM).

Hope your well.

Cheers,
Dave.

Do you think the average person watching YouTube on their phone, tablet or HD monitor on their PC is going to tell the difference?

Wrong target.

Cater to the videophile watching from a $10,000 ISF calibrated projector on a 110" screen.

If it looks good for him, it will look good for everyone.

(But the reverse may not always be true.)

And the same “videophile” wouldn’t be able to tell the difference between any codec and its associated optimal bitrate in a blind A/B visual test, regardless of file size.

The simple fact here is that AV1 is the most efficient common delivery codec available at the moment, regardless of its use, either by videophiles or those delivering/watching YouTube content.

Also, something looking good to a videophile doesn’t necessarily mean it will look good to everyone else. In my experience, philes of any nature quite often miss the point of the content as they tend to get distracted with technical dissection rather than engaging with the content. Something I’ve had to level at myself over the years.

Plus, many people mess with the tone controls on their TVs to suit their own personal taste, which is usually far from a technical appreciation. So I doubt a “technically perfect” picture that would please a videophile, will be appreciated by your average viewer due to their own personal taste for crushed, blown and over saturated images. BTW, I wouldn’t have thought most people viewing this thread would fall under the category “average viewer”, or at least I’d hope not.

Do you think the average person watching YouTube on their phone, tablet or HD monitor on their PC is going to tell the difference?

Wrong target.

Cater to the videophile watching from a $10,000 ISF calibrated projector on a 110" screen.

If it looks good for him, it will look good for everyone.

(But the reverse may not always be true.)

Yes technically, however people who watch my projects are family and groups who want a DVD. It has taken a while to get some Bluray and just now a few USB drives. I am sure most of them watch on old TV's as video is not their interest. The content is what they want. Most do not even see exposure problems or noise. They want to see faces and content. Period. Yes I have a few audiophiles who don't seem to hear the noise on their vinyl discs but hate mp3 ! Content is what matters. If your video quality is perfect and the content ( including audio ) is rubbish then your video is rubbish.

I think this thread has got off topic a little. It is clear that AV1 is able to generate small file sizes with minimal quality loss. Winner.