madVR - high quality video renderer (GPU assisted)

[James Freeman]

Does 4:2:0 require more bandwidth? Since 4:4:4 is the best option AFAIK, 4:2:0 requiring HDMI 2.0 is kind of backwards, isn't it?

No it does not require more bandwidth.
It is kind of backwards...

HDMI 2.0 required for 4:2:0.
HDMI 1.4 does not support 4:2:0 directly, it (DVD/BR/PC etc...) will send 4:2:2 or 4:4:4 with the 4:2:0 data in it (just copies the missing date from the adjacent pixel).

https://en.wikipedia.org/wiki/HDMI#Version_comparison

But, if you are using madVR from a PC you should not care less, you should be using RGB anyway.

[Anime Viewer]

Madshi you should write a madVR benchmark tool like SVP has. Then all the users can upload their results to a public spreadsheet. It would help troubleshoot performance problems by allowing comparisions and give a general rule of what to expect with a certain gpu/machine with different settings.

I think a benchmark program would be a waste of time. It would tell speed under certian conditions, but with so many options/variables avaialbe in madVR no two people are likely to be comparing under the exact same settings. Additionally, some systems would be able to support some settings, but not others. A lot of system would likely either crash themselves or crash the benchmark program trying to run settings that are either incompatible or too high.

MadVR is mainly about getting the best image quality while not straining the system from the power/performance/longevity of system components side. A benchmark program isn't going to tell us what looks best on our systems (what we've been testing, and reporting all these features under controlled conditions for all this time).

[James Freeman]

Why do you need a benchmark program? As stated by others, settings are individual thing per user.

Just keep the rendering time under frame time.
1/"frame time" = maximum rendering time before dropped frames.

[QBhd]

MadVR is mainly about getting the best image quality while not straining the system from the power/performance/longevity of system components side.

I push my system to the max with everything I watch. It's there to be used, I don't pay for electricity and I get the best PQ possible from my system. So the above statement is not very accurate, since I am sure I am not the only one squeezing out every drop of performance.

And yes a benchmark is pretty much useless.

QB

[James Freeman]

I don't pay for electricity

Care to elaborate?... :D

The difference between 20W and 1000W system is VERY significant.

[noee]

I don't pay for electricity...
QB

Yeah, I'd like know this too. Even my solar/battery set up costs me in terms of battery/panel maintenance and replacement, nevermind losses in the DC-AC inverter. The proper batteries and controller(s) are not cheap.

What I've found with madVR, is that I can use the CPU for decoding all sources and max my GPU(s) for the PQ and this seems to give reasonable power draw with fantastic PQ.

[Anime Viewer]

I push my system to the max with everything I watch. It's there to be used, I don't pay for electricity and I get the best PQ possible from my system.

QB

If you're running with your GPU at full load with the fan(s) at full bore, and the render times on the edge of being over your vsync interval then you're not likely getting the best picture quality (not all the time). Just because you're using a performance draining setting doesn't mean you'll get the best quality. Different options give different image output. While a video of real life people/environments may look good with Jinc other content (with drawn lines) may look better with Mitchell-Netavali. Some of the more taxing settings can produce artifacts, ringing, or aliasing. The second a scene gets more taxing than what your system can normally handle (be it with lighting effects, panning scenes, fancy subtitles, etc) you'll get a dropped frame or a presentation glitch and your picture/video will not be so smooth and fluid any longer.
Running your system at full throttle is also going to wear your hardware components faster than someone who runs things under control, so you're component's/system's lifespan will be shorter.

[nevcairiel]

Does 4:2:0 require more bandwidth? Since 4:4:4 is the best option AFAIK, 4:2:0 requiring HDMI 2.0 is kind of backwards, isn't it?

Its not backwards. It doesn't require more bandwith, but previous versions of HDMI just didn't specify 4:2:0 support. HDMI 2.0 does specify it, so HDMI 2.0 is required.

However, because it doesn't need more bandwidth, NVIDIA GPUs can do this on HDMI 1.4 transmitters - assuming the display supports HDMI 2.0 and understands the signal.

Have to make a distinction between the HDMI transmitter speed, and the protocol version. A HDMI 1.4 transmitter can be upgraded via software to get HDMI 2.0 features like 4:2:0 .. but software cannot add more bandwidth.

[j5627429]

max my GPU(s) for the PQ

Does MadVR actually take advantage of crossfire/sli?

[noee]

Does MadVR actually take advantage of crossfire/sli?

sorry, meant different gpus in different machines....I have no experience with madVR in either x-fire or sli.

[iSunrise]

...Regarding Fury X vs. 980ti, it seems like a no brainer. When spending $650 on a video card today, you probably want it to be HDMI2.0/HDCP 2.2 compatible to ensure it is future-proof. GTX960 and 980ti are, FuryX isn't. From what I've read recently, AMD's Fury is no different in 4K output format than my 3 year old 7970.
Personally, I'm trying to decide between the 960 and 980ti. Although 980ti does not have hardware h.265 decode for when we get 4K content, if it allows significantly higher quality settings for 1080>4K without any stuttering/glitches in playback, it is a sacrifice I'd be willing to make.
We already discussed this is another thread, Fury X seems to be limited to HDMI 1.4a, which support up to 4K 30fps just fine, which may be OK for local playback at the moment, but it's not future-proof. Apart from that, Fury X has a lot more raw shading power (at 1221MHz you have 10 TFLOPs SP) compared to the 980Ti, which is a cut-down GM200 (Titan X). Therefore, ALU-limited algorithms within madVR would probably be faster on the Fury, but that depends on the algorithms and their API. madshi uses various APIs to optimize for specific cases (e.x. NNEDI3, dithering). Fury X also has alot of raw bandwidth (512GB/s without overclocking), which can also be relevant.

So it depends what you want. If you are fine with 4K and max. 30fps, Fury X should be just fine for you.

So it really depends on your use-cases for it. We would need a comparison with Fury X and a 980 Ti when we're talking madVR, since game benches can be misleading. I am not entirely sure about 980Ti's HDCP 2.2 support, since several sites claimed that only the GTX960 supports HDMI 2.0 as well as HDCP 2.2.

[aufkrawall]

In benchmarks, Fury X oddly seems to be hardcore bandwidth limited: more core clock scales hardly, but more HBM clock gives a significant boost.
There's currently also the beeping tone which annoys many people.
The card is probably also limited with NNEDI3 by the interop copyback of the catalyst driver.
Well, its VPU is superior to GM200, but there is a whole lot of drawbacks otherwise.

[QBhd]

Care to elaborate?... :D

The difference between 20W and 1000W system is VERY significant.

Yeah, I'd like know this too. Even my solar/battery set up costs me in terms of battery/panel maintenance and replacement, nevermind losses in the DC-AC inverter. The proper batteries and controller(s) are not cheap.

What I've found with madVR, is that I can use the CPU for decoding all sources and max my GPU(s) for the PQ and this seems to give reasonable power draw with fantastic PQ.


It's simple, electricity is included in my rent so I can use as much as I want.

If you're running with your GPU at full load with the fan(s) at full bore, and the render times on the edge of being over your vsync interval then you're not likely getting the best picture quality (not all the time).

Running your system at full throttle is also going to wear your hardware components faster than someone who runs things under control, so you're component's/system's lifespan will be shorter.

Well NEDDI3 has won in all content I have thrown at madVR, for both Chroma and Image... so yes I am getting the best PQ to my eyes by maxing out GPU performance (128 neurons is ALWAYS going to be better than 64 if you can push it that far)

As to longevity of components... Computers are made to be used and periodic bursts of +90% usage is not going to shorten the lifespan by any significant amount. Turning your computer off after using it will shorten it's life far quicker than pushing it to it's limits. In 15+ years of running my PC's 24/7 (while gaming and encoding and now using madVR) I have had just one component fail. And that was a very old motherboard who's old style capacitors finally blew up due to extreme old age.

QB

[littleD]

i hope we get some MadVR benchmark (using MAX NNEDI3 settings)
from guys who just purchased said Monsters hehe.

Hm.. i guess Fury X will be the better buy?Great idea!
That fits my idea too.
Threre are so many different videos that can broke our settings and play with stuttering because our GPU is slow. What about benchmark AND then Adaptive Scaling Settings based on that benchmark. Let our graphic card use best scaling algorithm depends of its power. Let avoid situations of 4k or 60fps video makes video unwatchable. If Madvr automatically change to faster scaling, user will be happy anyway. This can make Madvr more user friendly than any time.

[huhn]

i can already read the complains about to loud GPUs ...

and is not like there is a best scaling algorithm at all. and the next issue nnedi an AMD gpu can easily run in problems with 70 % GPU usage.

[QBhd]

There is no one size fits all, there is no "best" (one person's NNEDI3 32 neurons is another's Lanczos8) and there is certainly no point in benchmarks and adaptive settings.

QB

[Catclaw]

If you're running with your GPU at full load with the fan(s) at full bore, and the render times on the edge of being over your vsync interval then you're not likely getting the best picture quality (not all the time). Just because you're using a performance draining setting doesn't mean you'll get the best quality. Different options give different image output. While a video of real life people/environments may look good with Jinc other content (with drawn lines) may look better with Mitchell-Netavali. Some of the more taxing settings can produce artifacts, ringing, or aliasing. The second a scene gets more taxing than what your system can normally handle (be it with lighting effects, panning scenes, fancy subtitles, etc) you'll get a dropped frame or a presentation glitch and your picture/video will not be so smooth and fluid any longer.
Running your system at full throttle is also going to wear your hardware components faster than someone who runs things under control, so you're component's/system's lifespan will be shorter.

I was under the assumption that render times had to be below the movie interval, not the vsync interval, am i wrong?

[Asmodian]

I was under the assumption that render times had to be below the movie interval, not the vsync interval, am i wrong?

No, you are correct.

[Asmodian]

Does MadVR actually take advantage of crossfire/sli?

madVR hates crossfire/sli. It runs slower with it enabled.

[XMonarchY]

Its not backwards. It doesn't require more bandwith, but previous versions of HDMI just didn't specify 4:2:0 support. HDMI 2.0 does specify it, so HDMI 2.0 is required.

However, because it doesn't need more bandwidth, NVIDIA GPUs can do this on HDMI 1.4 transmitters - assuming the display supports HDMI 2.0 and understands the signal.

Have to make a distinction between the HDMI transmitter speed, and the protocol version. A HDMI 1.4 transmitter can be upgraded via software to get HDMI 2.0 features like 4:2:0 .. but software cannot add more bandwidth.

I was talking about quality. Wouldn't 4:2:0 look worse than 4:2:2 and surely worse than 4:4:4 ? If it looks worse, then why include it in HDMI 2.0 anyway?!

[huhn]

I was talking about quality. Wouldn't 4:2:0 look worse than 4:2:2 and surely worse than 4:4:4 ?

8 bit 420 has 12 bpp and 8 bit 422 16 bpp and 4:4:4 24 bpp so yes.

Cb Cr just have an even lower resolution with 420

but you ware talking about bandwidth didn't you?

EDIT: If it looks worse, then why include it in HDMI 2.0 anyway?!

bandwidth.

HDMI 2.0 can't even do 10 bit UHD 4:4:4 at 60 hz. it just terrible. of cause DP 1.2 can do that and it is way way older.

[vivan]

I was talking about quality. Wouldn't 4:2:0 look worse than 4:2:2 and surely worse than 4:4:4 ? If it looks worse, then why include it in HDMI 2.0 anyway?!To deliver video as it is without wasting bandwidth (of course this is only true when you're using some standalone player, not PC with madVR - which can upsample chroma and do other stuff better than TV).

[Arm3nian]

I think a benchmark program would be a waste of time. It would tell speed under certian conditions, but with so many options/variables avaialbe in madVR no two people are likely to be comparing under the exact same settings. Additionally, some systems would be able to support some settings, but not others. A lot of system would likely either crash themselves or crash the benchmark program trying to run settings that are either incompatible or too high.

The benchmark should have predetermined things to test that require the majority of gpu horsepower. There is no point in a benchmark used as a comparison tool if the user chooses their settings. The majority of users run a dx11 gpu, and the majority of that run an nvidia gpu. Plus there aren't many options that are supported on nvidia and aren't supported on amd or intel's igpu. Also, why would a stable system crash running demanding settings? That's not how processors work.

MadVR is mainly about getting the best image quality while not straining the system from the power/performance/longevity of system components side. A benchmark program isn't going to tell us what looks best on our systems (what we've been testing, and reporting all these features under controlled conditions for all this time).
Maybe for you since you're on a laptop and don't want to melt a hole through your desk or legs.

Why do you need a benchmark program? As stated by others, settings are individual thing per user.

Just keep the rendering time under frame time.
1/"frame time" = maximum rendering time before dropped frames.

There is no one size fits all, there is no "best" (one person's NNEDI3 32 neurons is another's Lanczos8) and there is certainly no point in benchmarks and adaptive settings.

QB

You guys are missing the point. A benchmark would not be used to find optimal settings. It would be used to show what a certain gpu is capable of. There is a post every day with "I want to run 'x' setting, will gpu 'y' be good enough?" With a benchmark you can easily tell if the Fury x can handle doubling to 4k rather than guessing. When buying a gpu, you look at benchmarks. The benchmarks show various applications from cad to games. Not only does this show gpu performance in a certain program, but it shows the performance relative to other cards. Why not add madVR to that list. Maybe you're looking to buy a 970 to run 128 neurons, but think about buying a 980 instead to run 256. Then you purchase the 980 and find out it can't handle it, and end up wasting money since you used the same settings. There are an infinite number of uses for a madVR benchmark. The problem is implementing it, and that's up to madshi.

[Asmodian]

How would a benchmark differ from simply trying it? In your examples you don't need a benchmark tool (what would it do?), set the setting and see what the rendering time is. The complexity of the possible settings is the only issue and if you pick what "128 neuron doubling" means for all the other settings, source resolution, destination resolution, etc. it is easy to "benchmark" any GPU you own at "128 neuron doubling".

Like any GPU benchmark the results are highly dependent on the settings used and different users/sites use different settings so it is hard to compare results between users. Maybe a tool that played a stock video using a collection preset options with a particular player and at a specific resolution? It could report the average/min/max rendering times at each setting. This would be easy to do now, no special tool needed, but maybe review sites would include madVR performance if such a tool existed. However, it sounds like a lot of work, certainly more work than the benefit justifies before version 1.0 of madVR. ;)

madVR is changing fast so it isn't time for standard benchmark tools yet. :)

[QBhd]

^^

Well put. Let's put this slightly off topic to bed.

QB

[Arm3nian]

How would a benchmark differ from simply trying it? In your examples you don't need a benchmark tool (what would it do?), set the setting and see what the rendering time is. The complexity of the possible settings is the only issue and if you pick what "128 neuron doubling" means for all the other settings, source resolution, destination resolution, etc. it is easy to "benchmark" any GPU you own at "128 neuron doubling".
I can't test a gpu's performance in madVR if I don't have the gpu. It will help when buying a new card because I can see what scores others have gotten. Users can also post stock and overclocked results. I don't want it to be an e-peen tool, just something to view relative performance. Nothing like that currently exists. The best you can do is calculate how much faster a certain gpu is from another by using non madVR benchmarks, then apply that percentage to your current gpu's render clocks and get an estimated projection. Not very accurate.

Maybe a tool that played a stock video using a collection preset options with a particular player and at a specific resolution? It could report the average/min/max rendering times at each setting. This would be easy to do now, no special tool needed, but maybe review sites would include madVR performance if such a tool existed.
Yes this is precisely what I envision. If madVR is included in benchmarks carried out by popular sites it will gain more publicity and attract more users, including enthusiasts.

However, it sounds like a lot of work, certainly more work than the benefit justifies before version 1.0 of madVR. ;)

madVR is changing fast so it isn't time for standard benchmark tools yet. :)

I agree, there are lots of new things happening currently. When it is a bit more finalized then a benching tool can be of use.

[e-t172]

How would a benchmark differ from simply trying it?

Reading a number off a publicly available table (or better, a chart) is much easier than having to manually try dozens of combinations. The former takes seconds, the latter takes hours.

[huhn]

Reading a number off a publicly available table (or better, a chart) is much easier than having to manually try dozens of combinations. The former takes seconds, the latter takes hours.

and what is the point if you never really judge the picture quality of the scaler yourself?

[Arm3nian]

and what is the point if you never really judge the picture quality of the scaler yourself?

Nearly all of the settings, including needi3 and the newely implemented ones have a positive correlation between quality and GPU usage.

[e-t172]

and what is the point if you never really judge the picture quality of the scaler yourself?

Being able to know which ones you can cross out from the start because they won't be fast enough.
When you know what the scaler looks like, but you don't know how fast it will be because you're using a new GPU.
Sometimes we trust other people to be the experts on figuring out what scalers are the best or even worth considering. Not everybody has dozens of hours to spend trying lots of combinations of settings across a wide variety of samples. They just want something that is most likely to look best on typical content considering their hardware.

There are tons of reasons why having easily accessible performance data for various option combinations is a very good idea. I suspect it would also help madshi know what is worth optimizing.

[Asmodian]

Reading a number off a publicly available table (or better, a chart) is much easier than having to manually try dozens of combinations. The former takes seconds, the latter takes hours.

I meant in the context of generating this theoretical chart. A benchmark tool does not create said chart, someone still has to run it on each card.

The desire behind this request is for someone to test madVR with a wide collection of video cards in a systematic way and post the results. A special tool isn't required for that and it wouldn't even make the process much easier.

and what is the point if you never really judge the picture quality of the scaler yourself?

This is to use madVR like 3DMark, simply as a way to compare the performance of two video cards, not to pick settings.

[Arm3nian]

The desire behind this request is for someone to test madVR with a wide collection of video cards in a systematic way and post the results. A special tool isn't required for that and it wouldn't even make the process much easier.


It's not practical to do it manually. It would take time to create all the profiles, run them one by one, record them, then write it in the spreadsheet. Plus you need the same source material. Countless of other factors can result in errors and is overall too much of a hassle. No one would participate. You need to automate the process with a script. Press one button and it runs the bench and uploads the results to the spreadsheet.

[James Freeman]

You guys are missing the point. A benchmark would not be used to find optimal settings. It would be used to show what a certain gpu is capable of.

How would a benchmark differ from simply trying it?
madVR is changing fast so it isn't time for standard benchmark tools yet. :)

Arm3nian,

You can always create a "madVR GPU Benchmark Thread" in which you'll define certain parameters with a certain madVR version as a base.
People could post their results with the GPU's they own there, and not contaminate madVR official thread with such nonsense.

As I already said.
Just keep the rendering time under frame time.
1/"frame time" = maximum rendering time before dropped frames.

Cheers.

[JarrettH]

Benchmark? It's either realtime or the video is choppy :p Maybe the benchmark should be for how low your rendering time can go.

[Arm3nian]

Arm3nian,

You can always create a "madVR GPU Benchmark Thread" in which you'll define certain parameters with a certain madVR version as a base.
People could post their results with the GPU's they own there, and not contaminate madVR official thread with such nonsense.
Read my previous on why this wouldn't work.

As I already said.
Just keep the rendering time under frame time.
1/"frame time" = maximum rendering time before dropped frames.

Three pages and you still have no idea what we're talking about. If anyone should stop the nonsense, it's you.

Benchmark? It's either realtime or the video is choppy :p Maybe the benchmark should be for how low your rendering time can go.
Benchmark results show the most demanding setting you can run real time.

[ryrynz]

Might be an idea showing GPU usage on the OSD when GPU-Z is active.

[Nevilne]

Potplayer has GPU usage built-in fyi

[wanezhiling]

Potplayer has GPU usage built-in fyi

Yes.
Not support Intel GPU though

[madshi]

Now that we have a 10-bit output, I tried using CalMAN with madTPG to create a LUT for my display, and I'm running into a few issues:

1. When double-clicking to exit FSE mode, the screen just turns black rather than returning to windowed mode.
2. madTPG doesn't work very well if you have a single display setup.

It's very awkward to switch between CalMAN to set up the reads and then quickly switch over to madTPG and make it fullscreen on the same monitor.
Ideally what would happen is that madTPG would run hidden in the background and then enter an always-on-top state or FSE mode as soon as it is being told to display patterns.
And once that series of patterns is completed, return to being hidden in the background.

Ideally there would also be a way to pause for 5 seconds at the initial switch to FSE mode, but I assume that's something which would have to be fixed on CalMAN's side of things.
Hi there,

the latest Calman release now has some improvements when running Calman on the same PC as madTPG. The changes have been inspired by your "complaint"... :D So it would be awesome if you could try the new build and maybe provide feedback here:

http://www.spectracal.com/forum/viewtopic.php?f=94&t=5729

Stacey Spears has actively asked for feedback about this. Would be great if you could tell him/us if it works better now, or if there's still something that could be improved.

[XRyche]

Hi there,



Subtle and Polite LOL

[madshi]

Had actually meant to send it as a PM. Ooops... :)

[ikarad]

;)

Source: H.264 8bit 4:2:0 http://forum.doom9.org/showthread.php?p=1727413#post1727413

a. SuperRes off
b. SuperRes on (passes: 2 strength: 0.75 softness: 0.10)
c. SuperRes on (passes: 2 strength: 1.00 softness: 0.00)

\ a b c
1) chroma upscaling: Catmull-RomAR
2) chroma upscaling: Bicubic75AR
3) chroma upscaling: SoftCubic50AR
4) chroma upscaling: Lanczos3AR
5) chroma upscaling: JincAR
http://s5d2.turboimagehost.com/t/23452965_test1-1.png (http://www.turboimagehost.com/p/23452965/test1-1.png.html)

\ a b c
1) chroma upscaling: Bilateral
2) chroma upscaling: super-xbr100AR
3) chroma upscaling: super-xbr150AR
4) chroma upscaling: Nedi
5) chroma upscaling: NNEDI32
http://s5d2.turboimagehost.com/t/23452966_test1-2.png (http://www.turboimagehost.com/p/23452966/test1-2.png.html)

\ a b c
1) chroma upscaling: Catmull-RomAR
2) chroma upscaling: JincAR
3) chroma upscaling: Bilateral
4) chroma upscaling: super-xbr150AR
5) chroma upscaling: NNEDI32
http://s5d2.turboimagehost.com/t/23452967_test2.png (http://www.turboimagehost.com/p/23452967/test2.png.html)

It seems like no need Chroma SuperRes.:rolleyes:

What are the diffrences because i don't see anything?

[ryrynz]

What are the diffrences because i don't see anything?

Haha. You need to look closer.. or chop the picture up and compare in an image viewer, this is why I asked for individual images, placing them together like that doesn't make comparing them very easy.

BTW feel free to edit quotes BTW, you didn't need to quote his entire post.

[flashmozzg]

Better to use something like http://screenshotcomparison.com/

[omarank]

I don’t think super-xbr can replace Jinc as the best medium speed chroma upscaler, because Jinc is more natural looking. I prefer suber-xbr chroma (sharpness 100) over Jinc AR chroma in two instances, when the source is rather soft or the image is being downscaled by a large factor using CR AR LL.

Samples/screenshots, please?

FWIW, v0.88.13+ is better quality than v0.88.12.

Yes, super-xbr has improved a lot in the latest version. It may replace Jinc now, but still in some cases I find Jinc a tad better due to a more natural look. Please open this image (http://www.mediafire.com/view/wewuu3et8pdg8xt/sample.JPG) and toggle between Jinc and super-xbr. Avatar movie can be a good sample too.

I will post my feedback on SuperRes by the weekend.

[trip_let]

Better to use something like http://screenshotcomparison.com/

Yes, e.g.
http://screenshotcomparison.com/comparison/133465

ikarad, look at the right edge of the pink band for a difference. It's something.

[toniash]

Anyone knows how to solve this:
"error: input is not a PCH file: 'cl_kernel.h.pch'
fatal error: 'cl_kernel.h.pch' does not appear to be a precompiled header file" ??
I get it when trying to use s-xbr on XP SP3 and Nvidia 352.86

[Dlget]

Can anyone suggest best setting for gtx 960 OC ,i5 2500k,8 GB RAM.
I'm new to this things

[digby]

I'm getting lockups after updating to the latest version of madvr. it was either madvr or strongene hevc decoder i recently installed. disabled strongene and lockups still occur. playing hevc with lav, ffdshow raw, & svp on intel gpu.

[THX-UltraII]

I have some confusion about the fact that there is now sharpening and Super res in 3 different places. If I am using one for upscaling and one for native res do they cross each other? For instance if I tick sharpening under Image Enhancements does it also affect chroma upscaling and/or upscaling refinement? Which order should I be using them in for which settings?

Also. Since there is now a sharpening option can we have a luma denoiser (with sizer and strength) in future builds?

I m wondering the exact same thing. It s been a long time since I upgraded madVR. All of a sudden there a litterally 1000+ combination possible and a lot of settings seems to overlap each other like you say. Really don t have a clue where and how to start.....