madVR - high quality video renderer (GPU assisted)

[SamuriHL]

Like I said, it can do it, but, I have to make compromises in some areas. Which is fine. It's just that if you want a more complete experience where you can crank things up, you'll want a faster card.

[Alexkral]

You're talking about MadVR's dynamic tonemapping alone without additional tools, right? I have not tested it yet but I also expected it to be faster.

[tp4tissue]

Originally Posted by SamuriHL

As a 1060 6gb owner, I can tell you that if you have any intention of using madvr's tone mapping at all, there WILL be compromises. I honestly wish I had a much faster card but at the time the stupid crypto mining idiots were buying up all our cards. LOL I don't know what current prices look like but if you're looking at tone mapping I'd bump that up to a 1070 to get some breathing room. The 1060 CAN do it...it just requires some...convincing here and there.

What compromises ?

[ryrynz]

Well it's always going to be quality that's sacrificed.. But no point in commenting really until the final build is out.
Hopefully this month?

[huhn]

Originally Posted by 422415

I have attached the madVR OSD and my projector info when playing hdr files

can you upload them somewhere else? attachment on this forum my take sometime.

[SamuriHL]

Originally Posted by Alexkral

You're talking about MadVR's dynamic tonemapping alone without additional tools, right? I have not tested it yet but I also expected it to be faster.

Yea, the dynamic tone mapping that's being tested right now is pretty hefty on resources. That may change once they get it dialed in.

Originally Posted by tp4tissue

What compromises ?

Things like changing the chroma scaler from NGU to something that uses less resources like bicubic, changing dithering to ordered, etc. I try to avoid the trade performance options as much as possible but this is all just personal taste. What I mean is you're not cranking chroma scaling to NGU high and other options while tone mapping on a 1060. It's just not capable. But if you compromise some settings, tone mapping works just fine.

[SamuriHL]

Originally Posted by ryrynz

Well it's always going to be quality that's sacrificed.. But no point in commenting really until the final build is out.
Hopefully this month?

I wouldn't bet on this month. It would be nice but they're neck deep in the dynamic tone mapping stuff right now. madshi has been cranking out test builds like they're going out of style. LOL I've not been able to keep up with all of them lately, but, his main testers have and they've been working overtime. I thought he was going to release what he had in late December but he has continued to refine the HDR code. And the result has been WAY worth it.

[Warner306]

Originally Posted by SamuriHL

As a 1060 6gb owner, I can tell you that if you have any intention of using madvr's tone mapping at all, there WILL be compromises. I honestly wish I had a much faster card but at the time the stupid crypto mining idiots were buying up all our cards. LOL I don't know what current prices look like but if you're looking at tone mapping I'd bump that up to a 1070 to get some breathing room. The 1060 CAN do it...it just requires some...convincing here and there.

I was talking to someone at AVSForums who was getting 33ms rendering times with the latest test builds with a GTX 1060 with moderate settings:

https://www.avsforum.com/forum/26-ho...l#post57600310

You must be using very different settings.

[tp4tissue]

Originally Posted by SamuriHL

Yea, the dynamic tone mapping that's being tested right now is pretty hefty on resources. That may change once they get it dialed in.



Things like changing the chroma scaler from NGU to something that uses less resources like bicubic, changing dithering to ordered, etc. I try to avoid the trade performance options as much as possible but this is all just personal taste. What I mean is you're not cranking chroma scaling to NGU high and other options while tone mapping on a 1060. It's just not capable. But if you compromise some settings, tone mapping works just fine.

There might be a kink in your setup. I bought the same gpu as u

Madvr 92.17
Nvidia 417.71
Evga 1060 FTW+, 6GB , 2.1ghz core, 9ghz vram
G3258 CPU, 4.7ghz

Running on a 4K TV, These are tested on 1080p and 4K-hdr Remuxes, 40mbit (1080p), and 80mbit+ (4K)


For 4K HDR 24fps files 1:1 mapping

NGU Sharp- Med chroma
+ Dynamic tone map
+ Highlight Recovery


For 1080p 24fps 8bit , 2.35:1, w/black bar crop,

Ngu Sharp- High chroma
Ngu Sharp- Very High Luma
Ngu Sharp (Med) 2nd chroma.


For 1080p 24fps 8bit, 16:9, _________ Alternatively

Ngu Sharp- High chroma __________ Ngu Sharp- Med chroma
Ngu Sharp- Very High Luma _______ Ngu Sharp- Very High Luma
Bicubic60 AR 2nd chroma _________ Ngu Sharp- Low 2nd chroma


No quality reduction settings enabled, dithering on error diffusion

[Warner306]

Did you try the any of the latest test builds at AVS Forums? The performance is slower than the current official build.

[tp4tissue]

Originally Posted by Warner306

Did you try the any of the latest test builds at AVS Forums? The performance is slower than the current official build.

Hrrrm... That might be the kink then... haha

Which one is the latest.

[Warner306]

Try this one: https://www.avsforum.com/forum/24-di...l#post57681506

You need to uncheck all of the trade quality for performance checkboxes.

[tp4tissue]

Originally Posted by Warner306

Try this one: https://www.avsforum.com/forum/24-di...l#post57681506

You need to uncheck all of the trade quality for performance checkboxes.

All trade quality performances un-checked

4K HDR performance remains the same.

What's suppose to be slower/ changed from 94.17 ??

wait wait wait, hold on.. it's 7ms higher. but still playable. with my current NGU Med setting.

Tested on Lego ninjago movie, Let me test a 16:9 movie, maybe that will push it over the edge.



Does that Overlay on the top right add to render time? It's higher render times but not astronomically so, such that Madshi wouldn't be able to wiggle a few ms out of it to get back to 94.17 performance level right ?


OKOKOK, upon further testing, NGU med is indeed unplayable.

I don't know why, but it seems when i tested 2 minutes ago, Nutcracker 4 realms, the movie (Ngu Medium) plays at 37ms 42ms peak, Maybe this particular movie have very little HDR highlighting, I can't tell

NEARLY Every other movie plays at the 42ms avg 60ms peak.

I don't understand what the difference between the playable and unplayable movies are, since the bitrates are comparable, they're all 4K hdr. ?

SDR performance unchanged.

OK, Figured it out.. (maybe) Nutcrack 4 realm measures almost entirely in the < 200nit range, I guess that's why it's playable, because it doesn't actually trigger the heavier HDR processing.


It doesn't make sense though, cuz it says HDR on the box,

Halloween (2018), also measures almost entirely below 200, yet it is not playable on ngu medium.


OK, Found another one.. Dunkirk (2017).. This one does measure about 200 range, into 300, But also playable with NGU Medium 37ms avg


Hrrm.. I think I've got it now.. The HDR meta data ? Because for Dunkirk it says HDR 323 nit, and Nutcracker , HDR 183 nit.


So, If that number is BELOW the Target Nit setting of my monitor, the movie will play 37ms, if I set the target nit to LOWER than the meta data number, then it will kick into 42ms.

So if it reports 1000nit, and I set to target 1100, then it will drop to 37ms.. (not that you'd do this).. Only that Dunkirk and Nutcracker movie are the Exception with HDR meta at values lower than my TV's target nit.



Mystery solved..

[Warner306]

Did you try the settings posted at #55125. The HDR content in the movie shouldn't make any difference, nor should the bitrate.

[tp4tissue]

Originally Posted by Warner306

Did you try the settings posted at #55125. The HDR content in the movie shouldn't make any difference, nor should the bitrate.

It absolutely does make a difference, try it for yourself

If you set the hdr target nit to above the metadata nit, the render time decreases.


That explains the anomaly i experienced with dunkirk and nutcracker, because they have meta data at nits below my target nit..



Using ngu medium, my system will not run smooth with 4K HDR files.


His laptop from post 55125 says 1060 max-q which is more/less the slowest 1060 variant.



It looks like he's outputting 10bit 24hz though, so that makes sense that he could get 33ms on ngu medium.


Mine is 60 hz, I don't know if my tv supports 24hz on pc input.

but with 60hz output 8bit, with NGU medium, that pushes my render time to 42ms average , unplayable.


I guess this can be rescued with custom res.

We gotta be careful comparing render times, because the minor differences in setting make a huge difference in render time.

[SamuriHL]

At least you now understand the issue. It's very much dependent on content and how much madvr has to do for tone mapping. Some things will play fine with ngu medium. Others will not. It's right at the edge so any little hiccup pushes it into the dropped frame rendering time zone. I tend to use measurement files which helps performance. But it goes back to personal preference and prioritizing certain things to maximize the performance and quality. With this card, there are trade-offs. With a slightly faster card, you'd have a bit more rendering time to play with. Believe me, I've spent months playing with those builds and optimizing it for my C8. I didn't happen upon my settings by accident.

[tp4tissue]

Originally Posted by SamuriHL

At least you now understand the issue. It's very much dependent on content and how much madvr has to do for tone mapping. Some things will play fine with ngu medium. Others will not. It's right at the edge so any little hiccup pushes it into the dropped frame rendering time zone. I tend to use measurement files which helps performance. But it goes back to personal preference and prioritizing certain things to maximize the performance and quality. With this card, there are trade-offs. With a slightly faster card, you'd have a bit more rendering time to play with. Believe me, I've spent months playing with those builds and optimizing it for my C8. I didn't happen upon my settings by accident.

That guy with the Maxq 1060, the slowest 1060 variant

Something must be unique in his setup.

How is his render time 10ms faster than the fastest desktop variant. Unless I'm CPU bound.. which idk, could be.

I get 42ms average. I just tested on 23.98hz screen refresh as well, seems that makes no difference to render time.

His Laptop is MAGIC.

[Warner306]

Maybe he actually has a different GPU in there? Seems strange. Most people using projectors are constantly tone mapping, so the amount of tone mapping involved wouldn’t account for the difference.

[SamuriHL]

I'm running mine at 23, as well, and no it doesn't matter. I can tell you I have the slowest CPU on my HTPC. I'm still rocking a 3770k for now. I don't think that matters too much to this equation. The reason I've chosen the settings I have is because I still wanted NGU for Chroma. Call me stubborn.

NGU AA low
Ordered Dithering
Scale chroma separately

Those are my compromise settings. At the very moment I have dynamic hdr turned off. So my tone map HDR using pixel shader settings:

min target nits: 700
color tweaks: balanced
highlight recovery strength: very high
output video in HDR checked

The apply dynamic clipping and apply target nits selection are currently unchecked. Which isn't an issue for me cause as I said I tend to use measurement files to get a little performance back. With a movie like BvS I'm rocking avg rendering of about 31ms.

[tp4tissue]

Originally Posted by SamuriHL

I'm running mine at 23, as well, and no it doesn't matter. I can tell you I have the slowest CPU on my HTPC. I'm still rocking a 3770k for now. I don't think that matters too much to this equation. The reason I've chosen the settings I have is because I still wanted NGU for Chroma. Call me stubborn.

NGU AA low
Ordered Dithering
Scale chroma separately

Those are my compromise settings. At the very moment I have dynamic hdr turned off. So my tone map HDR using pixel shader settings:

min target nits: 700
color tweaks: balanced
highlight recovery strength: very high
output video in HDR checked

The apply dynamic clipping and apply target nits selection are currently unchecked. Which isn't an issue for me cause as I said I tend to use measurement files to get a little performance back. With a movie like BvS I'm rocking avg rendering of about 31ms.

Was the guy in the post using a measurement file as well ?

Because I don't see how it's possible that with a 500mhz less on core-speed and ~50 watts lower difference, he could render 10ms faster ..

Something is very odd here.

But it really could be cpu, at least for me, because I'm using a g3258 @ 4.7ghz on this htpc, It doesn't peak only ~50-75% cpu utilization, but it does have lower cache i believe.