madVR - high quality video renderer (GPU assisted)

[MysteryX]

the new AMD fluid motion stuff is a frame interpolation that makes frames up (with all of it's artifact of course).

madVR SM has a totally different aim. it is simply trying to display a given frame rate on a given refresh rate. it is not trying to come up with new frames and it doesn't create error like interpolations like SVP or this AMD stuff. technically all of them are interpolations but only AMD and SVP try to create things that's doesn't exist.

http://www.svp-team.com/forum/viewtopic.php?id=2018

Did they rename their Smooth Video feature into Fluid Motion or is it something different?

[baii]

That link to a older feature I believe. Svp can use cpu and gpu, the Amd is probably just gpu. Anyways, their goal is different than what madvr try to do like Huynh said.

Sent from my 306SH

[MysteryX]

Each of these video processing enhancements, along with the 5K panel support, has different (and sometimes incredibly specific) system requirements:

AMD Fluid Motion Video requires a 35W or higher AMD 7x00 A-series APU or an AMD Radeon R9 295X2/R9 290X/R9 290/R9 285/R7 260X/R7 260 GPU, Cyberlink PowerDVD 14, and Blu-ray content.

http://www.ign.com/articles/2014/12/10/amd-revamps-its-approach-to-drivers-with-the-catalyst-omega

I have rarely seen a computer with a Blu-ray drive to begin with :)

[leeperry]

Leeperry I am still using the GTX770 and latest nvidia drivers and could not be happier with madVR
Right, well I just put back the 7850 and PQ is as stunning as ever with monostatic ED1 LL so I'll just send back the upgraditis monster to where it came from. 1080p60 content looks jaw-dropping.

Just want to give a shout out to all the talented programmers giving me an amazing home theater experience.
Hard to deny, madshi's excellent work on mVR deserves major kudos as PQ goes way beyond all my expectations http://forum-images.hardware.fr/images/perso/ayuluna.gif

Hopefully someday we'll be able to repay him, should he finally decide to allow payments that is :p

[MysteryX]

Hard to deny, madshi's excellent work on mVR deserves major kudos as PQ goes way beyond all my expectations http://forum-images.hardware.fr/images/perso/ayuluna.gif

Hopefully someday we'll be able to repay him, should he finally decide to allow payments that is :p
Now that I finally decided to look into AviSynth and where the rescaling algorithm came from, I have even more respect for what madshi has done.

Compared to all the work that was done around AviSynth over the years, not only did madshi recode all these features to produce better output quality, but he's also the only one to have reprogrammed all these filters to use the GPU in an optimized way, and definitely the only one that can provide good results in real-time.

This goes way beyond anything else that has been done before over the years.

(technically, someone rewrote the NNEDI3 filter to work with the GPU, but that version often has worse performance than the CPU version!)

[6233638]

NEEDI3 chroma upscaling + 970 doesn't work properly
Asus Strix 970
344.75

and there are so many bugs trying to watch MKV movies. black flashes
[...]
Using Potplayer + MADVR + LavFilters CUVIDCheck that you have "use a separate device for presentation" enabled.
I would also suggest using DXVA Copy-Back rather than CUVID.

I must admit, I am disappointed with the performance of my 970 Strix in madVR though.
It can just barely manage 16 neurons NNEDI3 chroma upscaling and luma doubling with 720p60 content.

For SD content I can now use slightly higher NNEDI3 settings than my old 570, but it really doesn't seem to be much of an improvement in performance at all. Gaming performance is massively improved though.

[kalston]

Just got a ROG Swift monitor and I can't believe how demanding upscaling to 1440p is! I was originally using Jinc 3 taps / Jinc 3 taps / Error Diffusion but my GTX 670 was barely keeping up with blu-rays or even DVDs (queues looked pretty bad). I switched to Softcubic for "image upscaling" (left Jinc for Chroma) and "ordered dithering" and all is good and my queues are filled again (GPU usage at 100% when watching a blu-ray though haha). I don't know if those are the "best" settings as I didn't spend much time experimenting though :) (yet)

This is not a complaint as it actually looks fantastic and BTW, G-Sync works with madVR and it's much better than Reclock or JRiver's VideoClock since you can keep untouched bit-perfect audio (and you also don't need to worry about switching display mode :D ). G-Sync is a revolution and not just for gaming it appears :)

[e-t172]

G-Sync works with madVR

Can you explain what you mean by that? I'm interested in G-Sync but from the way it seems to be implemented it sounds like it would have no effect in madVR.

[kalston]

Well, it's just like games when you go in fullscreen exlusive mode : screen flickers briefly and the display reports that g-sync is on. And it looks perfectly smooth.

I've tried different framerates (25, 59, 60, 100...) but when I run fraps it always says 144fps (again, in fullscreen) so I'm not sure what to think now.

Even weirder, when I set the desktop refresh rate to something other than 144hz (say 100, 120) it still goes to 144hz + g-sync when I go in fullscreen exclusive mode in madVR. And when I disable g-sync... well screen still flickers briefly when entering fullscreen, display reports that g-sync is off, but refresh rate is back to... 144hz. And yes I've disabled display mode switching in madVR & JRiver. So I don't know, this is nothing to worry about for 24 or 23.976 fps IMO (and OSD didn't report the slightest dropped frame or anything during a 3h film, nor did I see anything with my eyes) but it's possible that 25 or 60 fps doesn't look "perfect" I guess. Maybe the high refresh rate just makes it a lot harder to notice. I don't have the very latest nvidia drivers (344.11 for now) as I am not rushing to upgrade my drivers nowadays. My GPU is getting old and I don't play a lot of freshly released games :)

Also, this means I can't use motion blur reduction (ULMB) in films because the display resets to 144hz and ULMB only works at 85, 100, 120hz. I don't know if it would look any good though, I'm just curious.

edit : hum there is a setting in nvidia control panel that I should probably play around with : "preferred refresh rate". Let's see...
edit 2 : ok that worked, no more 144hz when I enter fullscreen and ULMB does work for video playback now.

[dansrfe]

Just got a ROG Swift monitor and I can't believe how demanding upscaling to 1440p is! I was originally using Jinc 3 taps / Jinc 3 taps / Error Diffusion but my GTX 670 was barely keeping up with blu-rays or even DVDs (queues looked pretty bad). I switched to Softcubic for "image upscaling" (left Jinc for Chroma) and "ordered dithering" and all is good and my queues are filled again (GPU usage at 100% when watching a blu-ray though haha). I don't know if those are the "best" settings as I didn't spend much time experimenting though :) (yet)

That's strange, my EVGA GTX 670 FTW keeps up pretty well with Jinc 3 AR for all upscaling and Catmull-Rom AR LL for downscaling as well as NNEDI3 for x2 Luma.

Maybe a slight overclock is required over the GTX 670's base clock?

[kalston]

That's strange, my EVGA GTX 670 FTW keeps up pretty well with Jinc 3 AR for all upscaling and Catmull-Rom AR LL for downscaling as well as NNEDI3 for x2 Luma.

Maybe a slight overclock is required over the GTX 670's base clock?

Ok you got me curious. And here's the catch : with g-sync off, yep, flawless performance no matter the settings. G-sync on > queues don't fill up, I get a presentation queue of 1/4 sometimes even 0/4 and fraps reports a very fluctuating framerate below 144 (even though madVR doesn't report dropped frames or glitches) and it does look bad. But as soon as I switch back to windowed it looks perfect though. Oh yeah and GPU usage at 100% is also totally due to g-sync, it's a reasonable 30% or so without it. So that settles it, g-sync off for JRiver.

[mzso]

What's recommended to downscale 2560*1440 to fullHD. It seems like a YT channel I watch moved past fullhd.

[Asmodian]

edit : hum there is a setting in nvidia control panel that I should probably play around with : "preferred refresh rate". Let's see...
edit 2 : ok that worked, no more 144hz when I enter fullscreen and ULMB does work for video playback now.

That setting defaults to leave the refresh rate alone in the newer drivers. There have also been other minor improvements for G-sync in the newer drivers (unless you have SLI, then there are major improvements).

madVR's smooth motion works quite well at 144 Hz but using 120 Hz is optimal, of course.

You absolutely DO NOT want ULMB for low frame rate content with madVR. The frame is repeated four or five times.

When your eyes are tracking motion ULMB causes you to see five separate images instead of normal motion blur so it looks like there is motion ghosting instead of motion blurring. Ghosting is worse than blur (to most people at least) as we are used to motion blur in film. Actually when watching filmed content with a lot of motion blur you are unlikely to notice the ghosting so maybe it doesn't matter that much but with anime ULMB looks bad. I can notice what looks like very bad LCD ghosting, especially during pans, with ULMB on.

ULMB would probably be nice with some sort of motion interpolation (e.g. SVP), except for the motion interpolation artifacts. :p

(PG278Q ROG Swift, SLI Titans, Win 8.1 x64)

[e-t172]

You absolutely DO NOT want ULMB for low frame rate content with madVR. The frame is repeated four or five times.

When your eyes are tracking motion ULMB causes you to see five separate images instead of normal motion blur so it looks like there is motion ghosting instead of motion blurring. Ghosting is worse than blur (to most people at least) as we are used to motion blur in film.

Huh? I thought this was exactly how traditional film cinema projectors work: each frame is "flashed" two or three times, see Wikipedia: Shutter (http://en.wikipedia.org/wiki/Movie_projector#Shutter). So I'm somewhat confused - it seems like doing ULMB would bring the display closer to what a traditional cinema projector does, which sounds like a good thing. In fact, I was looking forward to using ULMB on 24p content precisely for that reason.

[kalston]

Huh? I thought this was exactly how traditional film cinema projectors work: each frame is "flashed" two or three times, see Wikipedia: Shutter (http://en.wikipedia.org/wiki/Movie_projector#Shutter). So I'm somewhat confused - it seems like doing ULMB would bring the display closer to what a traditional cinema projector does, which sounds like a good thing. In fact, I was looking forward to using ULMB on 24p content precisely for that reason.

I think so too and from my testing it didn't look bad at all, probably because madVR is outputting 120 and not 24fps.

In games however if the framerate isn't high enough ULMB looks horrible, as expected.

[6233638]

Huh? I thought this was exactly how traditional film cinema projectors work: each frame is "flashed" two or three times, see Wikipedia: Shutter (http://en.wikipedia.org/wiki/Movie_projector#Shutter). So I'm somewhat confused - it seems like doing ULMB would bring the display closer to what a traditional cinema projector does, which sounds like a good thing. In fact, I was looking forward to using ULMB on 24p content precisely for that reason.
This is done to reduce flicker, not because it is optimal.
Motion will not be smooth unless the refresh rate matches the framerate.

The issue with low persistence displays is that anything less than 50Hz or so flickers too much to only display the image once.
Even then, going back to 50/60Hz CRT-style flicker would not be acceptable to most people these days. That's why ULMB only works at 85/100/120Hz.

What is needed to solve this problem is high framerate video. That's why most of the LCD televisions which offer backlight scanning typically combine it with interpolation.
This allows them to use a low persistence backlight scanning mode at 120/240Hz without repeating images - though interpolation brings its own problems.

[e-t172]

This is done to reduce flicker, not because it is optimal.
Motion will not be smooth unless the refresh rate matches the framerate.

Of course. The question is whether ULMB looks better than simply outputting 24p@120Hz without ULMB. Or even compared to Smooth Motion.

[6233638]

Of course. The question is whether ULMB looks better than simply outputting 24p@120Hz without ULMB. Or even compared to Smooth Motion.Do you prefer motion blur or stuttering motion? That's all it really comes down to.

[e-t172]

It is not clear to me how UMLB would cause stuttering. Unless you are coming from the position that traditional cinema and CRT stutter.

[dbcooper]

With the new version of madVR, I can't force film mode for a DVD. Weird.

[ryrynz]

Right, well I just put back the 7850 and PQ is as stunning as ever with monostatic ED1 LL so I'll just send back the upgraditis monster to where it came from. 1080p60 content looks jaw-dropping.


Can anyone actually prove that that the output from a 7850 in this case actually looks any different from a nvidia 750ti or higher?

[leeperry]

I am disappointed with the performance of my 970 Strix in madVR
In my short experience with that 750Ti, as much as it's supposed to be pretty much as fast as the HD7850 in games some options such as NNEDI3 chroma and luma simultaneously with ED1 made it drop frames much faster so these two boards aren't nearly equally as fast with mVR, OTOH its 128bit memory controller might be the bottleneck.

Can anyone actually prove that that the output from a 7850 in this case actually looks any different from a nvidia 750ti or higher?
Let's assume that my MSI board was faulty and call it a day if you like, or bite the bullet and give it a shot...many stores provide a free return policy(mine did) and you'll get to see whether your display's behavior will change.

The difference is huge on my Sammy TV, It's like that MSI board was processing a really thick dithering pattern on top of my videos that made the picture look flat and plain ugly in comparison. Even its built-in HD tuner beats it by a long shot.

mVR looks way better off those two GCN boards I mentioned earlier IME. Hell, I woulda never been able to DBT mVR's dithering algorithms with this 750Ti when OTOH differences are very obvious with those two AMD boards I tried. When going from ordered dithering to monostatic ED1 LL, movements are far cleaner and clearer with AMD due to the inexistent dither patterns, but with the 750Ti I would definitely agree that the additional load is not worth the hardly visible tiny improvement.

Again, ppl have been raving about AMD's superior PQ on HCFR since forever and I had a similar experience when going from an MSI 8600GT to a Sapphire HD2600Pro on my HC3100 pj.

Just like audio gear it's also about synergies sometimes, three 8bit dithering algorithms one on top of each other(mVR>GPU>display) can lead to unexpected results. I tried my luck and really wanted to like that 750Ti due to its fairly decent cooler and low power consumption....reason why I spent quite a lot of time calibrating it and tweaking every friggin' option to no available.

Now if madshi could get AMD to drop the copyback nonsense :)

[Asmodian]

Huh? I thought this was exactly how traditional film cinema projectors work: each frame is "flashed" two or three times, see Wikipedia: Shutter (http://en.wikipedia.org/wiki/Movie_projector#Shutter). So I'm somewhat confused - it seems like doing ULMB would bring the display closer to what a traditional cinema projector does, which sounds like a good thing. In fact, I was looking forward to using ULMB on 24p content precisely for that reason.

I am not sure to be honest, I only know I went "what is with this ghosting on my new screen" when I turned on ULMB with madVR. It did seem a bit sharper though.

Maybe there is still something different about the way LCDs display video that makes it more visible? Or maybe I was simply extra sensitive because I was looking for possible defects on a new screen.

I think so too and from my testing it didn't look bad at all, probably because madVR is outputting 120 and not 24fps.

In games however if the framerate isn't high enough ULMB looks horrible, as expected.

The effect in video is similar to the effect in games because the same thing is happening.

You do not see it in regular video because everything is already so blurry in high motion (due to shutter speed, etc.) you don't notice the ghosting. If watching something without motion blur it is identical to a very steady 24 fps ULMB in a game. If your eyes watch one part of the screen you do not see it but if you track a moving object it seems to have ghosting.

Without ULMB it is blurry enough you couldn't see the ghosting if it was there so to each their own. I would have said I would like sharper but the ghosting bothered me. :p

(aside: ULMB is amazing at 120 fps in games, you can even read really fast moving text but, as you say, at <120 fps the ghosting gets painful. Of course it is worse in games compared to madVR because the ratio of "time per frame" to "amount of motion represented by the frame" keeps changing while madVR's smooth motion keeps this ratio exactly one.)

[6233638]

It is not clear to me how UMLB would cause stuttering. Unless you are coming from the position that traditional cinema and CRT stutter.Repeating the same frame multiple times on a low persistence display results in stuttering, whatever type of display it is.

Any time the framerate is a divisor of the refresh rate, rather than being equal to the refresh rate, will result in stutter/"double"-images. (really, as many images as the frame is repeated)

A common complaint from Plasma TV owners (which have something like 50% persistence) is that 30 FPS content has "double-images" when the camera pans. These issues are most obvious with video games since they either have no motion blur, or artificial motion blur which does not help mask the issue the same way that real camera blur does on something which is filmed.

You don't get this stuttering/double-images on a full persistence display like a Sample & Hold LCD or OLED, since repeating a frame just means that it is held for a longer duration.

[James Freeman]

Best motion I've ever seen was on a 72Hz monitor, 1 movie frame 2 black frames.
Resulting in a crazy flickering of 24Hz but the motion was so very smooth almost looked like HFR because the brain completed the missing data.

After about 10 minutes of getting used to the horrible 24Hz flickering I could watch the entire movie like this.
I rewatched several segments of several movies to discover the amazing fluidity of TRUE Single Flash 24Hz experience, no motion blur besides what was actually recorded on camera.
It was the best motion-blur-less experience I have seen of 24fps movies besides the small resulting artifact of epilepsy-inducing flickering. :D

In terms of motion blur this is how movies designed to be watched in the beginning of the 20th century, single flash at 24Hz.
When Double and Triple flashing appeared to mend flickering it also created a double and triple ghosting in our brain visual system.

For movies, a good balance would be 48fps together with a tolerable 48Hz single flashing.

PS, can't wait till Friday 19th for the 3rd Hobbit movie in 48fps. Hurrah.

[aufkrawall]

I must admit, I am disappointed with the performance of my 970 Strix in madVR though.
It can just barely manage 16 neurons NNEDI3 chroma upscaling and luma doubling with 720p60 content.

Try Jinc for chroma and disable ED, then 32 neurons for quadrupling 720p60 should work (at least with OC).
There's hardly any reason to use NNEDI3 for chroma.

High fps seem to be a gamebreaker with NNEDI3.

[ryrynz]

High fps seem to be a gamebreaker with NNEDI3.

I wonder if it's worth having dual neuron selection for high frame rate content. Half 16 half 32 for example.. giving some finer grain adjustments and tweaking things to the max..

Now if madshi could get AMD to drop the copyback nonsense :)

they replied that their engineering team was aware of the problem but didn't have the time to look at it.

We don't need Madshi for that. AMD have allocated more resources to their driver development team and improved the system. If enough people submit the problem on their issue reporting page (http://www.amdsurveys.com/se.ashx?s=5A1E27D216F2CD32) it'll get some attention and hopefully resolved promptly.


Can anyone else see any differences on their display with AMD hardware in comparison to Nvidia?

[aufkrawall]

Can anyone else see any differences on their display with AMD hardware in comparison to Nvidia?
Sometimes one can read that AMD was to be more "clear, sharp, whatever" than Nvidia.
I could never spot any difference between Nvidia and Intel, AMD neither. I'd assume that it's kind of modern myth or even FUD.
Unless there are measurements with RGB capture cards.

[huhn]

Sometimes one can read that AMD was to be more "clear, sharp, whatever" than Nvidia.
I could never spot any difference between Nvidia and Intel, AMD neither. I'd assume that it's kind of modern myth or even FUD.
Unless there are measurements with RGB capture cards.

the picture quality it totally different between intel nvidia and amd. but not for video rendering it is in games and they handle things different resulting in different colors, sharpness and the rest.

have a look at this: https://www.youtube.com/watch?v=0Md1Xpwq1X0

so i'm geuss most problem come from the fact he isn't using a full range RGB signal in the hole chain (he a has good reason for that "BFI")

but we should stop here because this is for sure not an madVR problem.

[aufkrawall]

the picture quality it totally different between intel nvidia and amd. but not for video rendering it is in games and they handle things different resulting in different colors, sharpness and the rest.

I thought we were talking about non-gaming usage. :)
It would be a total mess if colors looked different in general with each IHV, e.g. how could you be certain about pre-calibrated displays anymore? Color standards would be useless, unless you calibrated each display again with different GPUs.

Short sidenote regarding gaming, after which I will stop:
NV has somewhat better texture filtering than AMD (often negligible, however not always) and more flexibility regarding anti aliasing.
Otherwise, results should be equal (with both operating at the same brightness levels), unless driver or game developers screwed something up with IHV specific optimization.

Upcoming NV driver should have better options to select proper range levels, btw.

[leeperry]

We don't need Madshi for that. AMD have allocated more resources to their driver development team and improved the system. If enough people submit the problem on their issue reporting page (http://www.amdsurveys.com/se.ashx?s=5A1E27D216F2CD32) it'll get some attention and hopefully resolved promptly.
madshi has been in contact with a AMD dev that might care and get things done, that's all....but sure great idea, if we all send the same message whining about interop lag maybe that'd help, at least we wouldn't be stuck with the 13.12 set forever.

the picture quality it totally different between intel nvidia and amd. but not for video rendering it is in games and they handle things different resulting in different colors, sharpness and the rest.
breaking news, mVR is a game to graphic drivers.

how could you be certain about pre-calibrated displays anymore? Color standards would be useless, unless you calibrated each display again with different GPUs.
colorimetry will slightly change even if you only roll the refresh rate on a given graphic card so yeah precalibrated displays are a wild dream at best.

so i'm geuss most problem come from the fact he isn't using a full range RGB signal in the hole chain (he a has good reason for that "BFI")
Of course I am, "dispcal -R" gave the same figures on both boards and both were D65/gamut mapped.

BFI has nothing do whatsoever with color calibration, If I were you I'd spend some time polishing my english writing skills instead of making a lot of false assumptions based on an obvious lack of experimentation.

[huhn]

BFI has nothing do whatsoever with color calibration, If I were you I'd spend some time polishing my english writing skills instead of making a lot of false assumptions based on an obvious lack of experimentation.

so you are not using BFI anymore but PC mode with Samsung to avoid subsampling? if so i'm sorry.

[nevcairiel]

breaking news, mVR is a game to graphic drivers.

Not really. The differences in games come from actual 3D scene rendering, which madVR doesn't use or need. It may not behave like a "typical" video renderer, but its still rather far away from being a game.

[leeperry]

If enough people submit the problem on their issue reporting page (http://www.amdsurveys.com/se.ashx?s=5A1E27D216F2CD32) it'll get some attention and hopefully resolved promptly.
OK done, hopefully more ppl in this thread can send the message: http://thumbnails111.imagebam.com/37250/da235a372493454.jpg (http://www.imagebam.com/image/da235a372493454) http://thumbnails109.imagebam.com/37250/7b8eab372493455.jpg (http://www.imagebam.com/image/7b8eab372493455)
The issue with the madVR D3D video renderer and Catalyst drivers was clearly explained by its developer "Mathias Rauen" at http://developer.amd.com/community/blog/2014/11/04/what-do-you-need/

"(madVR is) using D3D9 OpenCL interop for video processing (upscaling etc), and while this feature generally works fine, the current AMD implementation is rather slow. The reason for that is that every image that is interop’ed is transfered from GPU RAM -> System RAM -> GPU RAM. Technically, this should not be necessary, but that’s how the current AMD drivers have implemented OpenCL interop, and this seriously slows down my video renderer. This problem does not occur with Intel and NVidia implementations. I’ve written a small benchmark tool (with full source code) which demonstrates this problem:

http://madshi.net/OpenClSpeedTest.zip

The benchmark results show that when using OpenCL, interop dramatically slows down performance, while with DirectCompute, the performance impact of D3D11 D3D9 interop is very small. E.g. using OpenCL, without interop a simple 1080p image copy runs with about 2000fps on my AMD 7750. Using interop the image copy slows down to 230fps. That is a simple copy without any actual processing. Furthermore this is on a fairly recent PCIe 3.0 PC. Users have reported much slower interop results on older PCs. I’ve seen image copy speeds reported of less than 24fps, which means on some PCs even a simple image copy isn’t fast enough, anymore, for real-time Blu-Ray video playback, when using D3D9 OpenCL interop."

Performance gets a lot worse over a PCI-E 2.0 connection instead of 3.0 and so far madVR works best using the 13.12 drivers that are one year old now.

[aufkrawall]

Not really. The differences in games come from actual 3D scene rendering, which madVR doesn't use or need. It may not behave like a "typical" video renderer, but its still rather far away from being a game.
I wouldn't think the driver necessarily distinguishes between a game or any other app with a different purpose.
With DirectX, it should typically just perform what it is told by the API.
In case of madVR, there are some realtime rendering similarities compared to games, but that's probably it already.

[ryrynz]

In case of madVR, there are some realtime rendering similarities compared to games, but that's probably it already.

To settle it, if anyone has a recent AMD card and an Nvidia card they can compare, using RGB output using the same levels on the same display please describe any visual differences noticed using ordered & error diffusion dithering between the two (untick 'change dither for every frame') and take PNG screenshots of a few different frames for comparison, thanks.

[MSL_DK]

To settle it, if anyone has a recent AMD card and an Nvidia card they can compare, using RGB output using the same levels on the same display please describe any visual differences noticed using ordered & error diffusion dithering between the two (untick 'change dither for every frame') and take PNG screenshots of a few different frames for comparison, thanks.

I replaced my Asus STRIX-GTX750TI-OC-2GD5 with a Sapphire RADEON R9 270X Dual-X OC due to the same problem as leeperry, however, I would not describe it as problematic as him. I still have my Asus STRIX-GTX750TI-OC-2GD5 and would like to make this test when I have time during the week. Is there a particular scene I have to take screenshots? I am, moreover, great happy with my Sapphire RADEON R9 270X Dual-X OC, it hits 23,976 absolutely perfect! It's more than what I could say about my Asus STRIX-GTX750TI-OC-2GD5 :(

[Qaq]

IDK about Nvidia, but for AMD you have to dig very deep into registry to disable all the image processing. Untill that any comparisons make no sence IMO.

[QBhd]

IDK about Nvidia, but for AMD you have to dig very deep into registry to disable all the image processing. Untill that any comparisons make no sence IMO.

Please elaborate since I have been using AMD/ATI for over 15 years and I have never had to "dig deep" into the registry to disable image processing... Unless I have been missing something for all these years...

QB

[Qaq]

I have never had to "dig deep" into the registry to disable image processing...And how did you know you have it actually disabled?
Not all image processing options are listed in the Catalyst.
Not all image processing options disabled in the Catalyst in fact disabled in the registry (or in Profile.xml).
Thats what I'm talking about.

[leeperry]

To settle it, if anyone has a recent AMD card and an Nvidia card they can compare, using RGB output using the same levels on the same display please describe any visual differences noticed using ordered & error diffusion dithering between the two (untick 'change dither for every frame') and take PNG screenshots of a few different frames for comparison, thanks.
It's all happening between mVR and the display AFAIK so I'm not sure what you are expecting exactly with PNG screenshots of mVR's output. We would need something like a ditherometer as it would appear :D

Long story short if you are genuinely impressed by mVR's PQ on your 750Ti then live happily ever after(same goes with ppl who think that all DAC's sound the same, one less thing to worry about), if we've poisoned you to compare both offerings then I'd say get a 270 from a seller with a free return policy and see how that fares for you? The difference I'm seeing is far beyond placebo with monostatic ED1 LL, really. Yesterday I watched Lucy (http://www.imdb.com/title/tt2872732/) in D65/2.4/EBU gamut/24.000Hz, PQ was X_X

All I know is that I would fully understand ppl who think that mVR isn't worth the hassle compared to EVR if they are using a 750Ti(or maybe nvidia in general but that's their only recent board I tried), let alone think that Error Diffusion is a sheer waste of resources.

Please elaborate since I have been using AMD/ATI for over 15 years and I have never had to "dig deep" into the registry to disable image processing
All those bs options are only applied on EVR/VMR and so on AFAIK.

[Shiandow]

It's all happening between mVR and the display AFAIK so I'm not sure what you are expecting exactly with PNG screenshots of mVR's output. We would need something like a ditherometer as it would appear :D


There is a small, but non-zero chance that the difference is caused by a difference in floating point accuracy. With PNG screenshots you should be able to rule that out.

[aufkrawall]

All those bs options are only applied on EVR/VMR and so on AFAIK.
I suppose they are applied on NV12 DXVA frames only.

[TheElix]

And how did you know you have it actually disabled?
Not all image processing options are listed in the Catalyst.
Not all image processing options disabled in the Catalyst in fact disabled in the registry (or in Profile.xml).
Thats what I'm talking about.Could you be more specific, please?

As to comparison of PQ between video adapters, I can say there is definitely a difference in VGA output of Intel GMA4000 and AMD HD7990 - and the difference is immediately noticeable on desktop. The image is sharper. But that only means AMD has better DAC. We were talking about digital output and on-board processing. I believe only an A/B comparison on the same display with different sources could tell. Assuming a double-blind method is used.

[Qaq]

Could you be more specific, please?... We were talking about digital output and on-board processing.Just check AMD registry settings, its all there.
And its not only related to PQ, but to perfomance as well.

[baii]

When using madvr, there is only a few ways that the AMD "image processing" can get into effect iirc.

1. interlaced material, and you let madvr call for GPU deinterlace
2. put ffdshow raw video filter into the chain and check certain options, which i forgot what exactly ~~

[MSL_DK]

Just check AMD registry settings, its all there.
And its not only related to PQ, but to perfomance as well.

Have you read features (http://forum.doom9.org/showthread.php?t=146228)

[clsid]

Just check AMD registry settings, its all there.
And its not only related to PQ, but to perfomance as well.
Can you give a list of registry key/value names? That should be helpful for everyone interested in tweaking.

I know that for quality it can help to disable Fleshtone_NA and DynamicContrast_NA.

[DigitalLF]

I need to change my 280x i use in my HTPC.. that computer is used for well movie/series/web/mail no gaming. but PQ is priority 1 on that computer and the 2th priority is "heat vs performance"

What card would be best for MadVR? NNEDI3 128 neurons seems to demand a lot of the poor card :P

[Qaq]

Can you give a list of registry key/value names? That should be helpful for everyone interested in tweaking.

I know that for quality it can help to disable Fleshtone_NA and DynamicContrast_NA.Well, in fact NA means "not available" (or "not used") so you have to keep NA=1 if you want to disable those. Better set ENABLE=0 instead. :)

I wish I could give a enemies list, but I'm not really an expert there.
There is a whole banch of settings and its about a year since I installed, set and tweaked 13.12 AMD driver version. That was a time AMD users tried to measure OpenCL perfomance to see if they are ready for NNEDI. I found its worth first to tweak the driver settings to a death to get as much GPU perfomance as possible. DXVAChecker crashed me a whole system down on 13.12, so I have nothing to do but go to the registry (Catalyst > Info) and google/guru3d for keys and values.

I wasn't much surprised that some features that were looking like disabled in Catalyst were still enabled in the registry. I disabled those and some others as well, which I could understand. Today I have cheked the registy and some unwanted driver features look like enabled again, thanks to AMD guys that use to fool tweakers all the time...:mad:

Do all those settings affect picture and perfomance? I don't know. My point: kill them all and let god sort them out. I'm not a gamer, I only want untouched picture and max perfomance. I wish more people wanted the same, so we could diskuss it to a death somethere.