madVR - high quality video renderer (GPU assisted)

[clancy688]

Because every time there's a stutter, the frame drop counter increases. In my previous post I said that madVR isn't reporting "dropped frames" because I was to stupid to actually look at the frame drop counter. I only looked at the "1 dropped frame every X minutes" stat.

[Pat357]

The stutters (in all modes, FSE on, off, Aero on, off) are dropped frames... every time the image stutters, the dropped frames counter increases by one or two...
But I think I've found a way to "activate" and deactivate" those stutters. When I open a video while the player's (tested with Pot and MPC) already in fullscreen, or open a video and activate fullscreen without triggering the seek bar, no frames are dropped.
But when I move the mouse cursor to the bottom of the screen and activate the player's seek bar for a short moment, soon after light frame dropping and stuttering starts.
This seems to be also the case when the player's NOT in fullscreen. As long as the player GUI is visible (and for a short time period thereafter), there's stuttering. With MPC not in fullscreen (standard interface), I had terrible stuttering. With PotPlayer (I've set it up so that there's only the video itself visible and the seek bar pops up when needed) not in fullscreen, no stuttering. But with the seek bar visible in windowed mode -> stuttering.
So an immeadiate workaround would be operating in fullscreen and NOT using the seek bar... ^^;
Still I'm curious why the GUIs are triggering such behaviour.

This is the situation with my 23p TV.

But for my main monitor there's another thing which's strange and which I stumbled upon by chance.

With FSE disabled there are no dropped frames on my 60 Hz main monitor... but with FSE enabled there are...? I always thought FSE is the allround crash-fix for everything which's not working properly, and now it's creating stuttering which wasn't there before...?

Any idea what this means?

Seems indeed complete illogical....
What hardware are you using ? GPU ? Mobo ? CPU ?
Installed version your GPU - drivers ?
What content where you playing when this happened ?
(x264 , vc-1, mpeg2....? fps ? )
What codec & mode (softw, DXVA-native or other, CUVid, Quicksync, ....) are you using for decoding your content ?
(LAV, build-in MPC, build-in Potplayer, .......)
Are subtitles involved ? What do you use to display subs ?

Let's start with this to try to figure out what's going on with your system..

[clancy688]

What hardware are you using ? GPU ? Mobo ? CPU ?


Sapphire Radeon 6870, Asrock A790GXH/128M, Phenom II X4 945


Installed version your GPU - drivers ?


Catalyst 12.6, newest I guess (but I've disabled all video processing options in the driver).


What content where you playing when this happened ?
(x264 , vc-1, mpeg2....? fps ? )


So far everything (dropped frames). But I'm mostly testing with some Hi10P 1080p 23.976fps content because that's what I'm playing 90% of the time.


What codec are you using for decoding your content ?
(LAV, build-in MPC, build-in Potplayer, .......)
Are subtitles involved ? What do you use to display subs ?


LAVFilters (Audio + Video), Haali, madflac, PotPlayer and MPC-HC, I'm using the internal subtitle engine of both of them. And yes, subtitles were involved.


I can't make further tests right now. It's quite late over here and there's a thunderstorm going on, so I've disconnected every valuable item from the power grid / network. Main PC and TV included. Better safe than sorry, I've a history with lightning induced equipment failures... ^^;
Do you mind if I try out any suggestions you may come up with tomorrow...? ^^;

[Pat357]

Sapphire Radeon 6870, Asrock A790GXH/128M, Phenom II X4 945



Catalyst 12.6, newest I guess (but I've disabled all video processing options in the driver).



So far everything (dropped frames). But I'm mostly testing with some Hi10P 1080p 23.976fps content because that's what I'm playing 90% of the time.



LAVFilters (Audio + Video), Haali, madflac, PotPlayer and MPC-HC, I'm using the internal subtitle engine of both of them. And yes, subtitles were involved.


I can't make further tests right now. It's quite late over here and there's a thunderstorm going on, so I've disconnected every valuable item from the power grid / network. Main PC and TV included. Better safe than sorry, I've a history with lightning induced equipment failures... ^^;
Do you mind if I try out any suggestions you may come up with tomorrow...? ^^;

Are the stutters there with all these decoders ? Both in software and DVXA mode ?

Have you tried different flush settings in MADVR ?
Wat are your current flush settings ?
Do you have "use seperate device for ...." checked ?
Is DX11 checked ?
Maybe it's easier that you post your full Madvr settings ?

[clancy688]

Are the stutters there with all these decoders ? Both in software and DVXA mode ?


Tested only LAV Filters and DivX HD, but so far stutters are there with both decoders. I'm not using DXVA since most of my vids are Hi10P.


Have you tried different flush settings in MADVR ?


Nope, I don't have the slightest idea what those flush settings are doing, so I didn't want to meddle with them.


Wat are your current flush settings ?


From top to down:

windowed:

8

flush
flush & wait (sleep)
don't flush
don't flush


exclusive mode:

16

flush
flush & wait (sleep)
don't flush
don't flush


Do you have "use seperate device for ...." checked ?
Is DX11 checked ?


Nope.
I tried turning it on. Didn't change much for windowed mode. But in FSE mode, madVR isn't dropping frames anymore. On neither monitor.
What are those options doing?


Maybe it's easier that you post your full Madvr settings ?

Is there a plain text file with my settings somewhere? So far I only found the settings.bin which's not text...
I added it to this post (with use separate devide and DX11 options still disabled)

[Pat357]

Tested only LAV Filters and DivX HD, but so far stutters are there with both decoders. I'm not using DXVA since most of my vids are Hi10P.



Nope, I don't have the slightest idea what those flush settings are doing, so I didn't want to meddle with them.



From top to down:

windowed:

8

flush
flush & wait (sleep)
don't flush
don't flush


exclusive mode:

16

flush
flush & wait (sleep)
don't flush
don't flush



Nope.
I tried turning it on. Didn't change much for windowed mode. But in FSE mode, madVR isn't dropping frames anymore. On neither monitor.
What are those options doing?


Is there a plain text file with my settings somewhere? So far I only found the settings.bin which's not text...
I added it to this post (with use separate devide and DX11 options still disabled)

Besides the attachment needs first approval, I'm afraid I can't read this file : in this file is also your monitor info IIRC, and you can't exchange these between completely different systems.
Maybe another idea : make screen-shots instead from your MadVR settings and upload these to http://www.mediafire.com/

Another very important thing is the following :
When a frame get dropped, do you see the other cues changing ? Like the decoder cue, upload cue, render cue or backbuffer cue. (All displayed on OSD)
It's important to watch carefully and tell me in what order they change : which one changes first, which one next,..... or doesn't any from these change when a frame get dropped ?

[clancy688]

FSE mode disabled:

decoder queue 11-12 / 12
upload queue 7-8 / 8
render queue 7-8 / 8
backbuffer queue 6-7 / 8

When a frame is dropped, nothing happens for decoder queue. But upload queue and render queue change to 5-8 / 8 or 6-8 / 8 and then back to 7-8 / 8. Backbuffer queue reverts to 0-7 / 8 and then goes back to 6-7 / 8. All three change and go back to their previous values exactly the same moment.


Should I observe the bevaviour in FSE mode as well? There I don't have dropped frames anymore thanks to the separate device and DX11 options.


madVR settings - I think I'll just write what options I've enabled, that's probably the fastest way. Everything not mentioned is unchecked


Devices

Main Monitor:
Properties: PC levels, native display bitdepth 7 bit
Calibration: disable calibration
display modes: switch to matching display mode when playback starts
color & gamma: Enabled, pure power curve, 2.40

TV
Properties: PC leves, native display bitdepth 8 bit or higher
Calibration: disable calibration
display modes: switch to matching display mode when playback starts, restore original display mode when media player is closed, display modes: 1080p23, 1080p30, 1080p50, 1080p60
color & gamma: Enabled, pure power curve, 2.20


Processing
Decoding: Everything unchecked
Deinterlacing: automatically active, if in doubt, deactivate, everything else except disable automatic source type detection enabled


Scaling Algorithms
Chroma Upscaling: Lanczos 3 taps
Luma Upscaling: Lanczos 3 taps
Luma Downscaling: Bicubic, sharpness 50


Rendering
General settings: delay playback start until render queue is full, use a separate device, D3D11, CPU queue size 12, GPU queue size 8
Windowed mode settings: Already posted above
Exclusive mode settings: Already posted above
Trade quality for performance: Everything unchecked

[madshi]

Hey guys,

sorry for the lack of replies. Still busy with my commercial projects. Will need maybe 2-3 more weeks before I find time for madVR again.

Please do not repost your questions. I will go through the thread and read and reply (where needed) to all your comments when I find some time. There's no need to repost anything.

*However*, this weekend I played around with resampling a bit and ended up creating a new madVR test build which you can try if you like:

http://madshi.net/madVRantiRinging.rar

This is just a test version, no official build and it only contains the changed files. So copy these files over your 0.82.5 installation. There are 2 key changes in this test build:

(1) The Lanczos Luma upscaling algorithms were totally rewritten to use more math and less texture accesses. On my ATI 3850 they now run about 25-30% faster than before. It would be nice if you guys could double check if they run faster for you, too, on all the funny GPUs you have. Please also check if image quality is alright. For speed comparisons you can simply switch between e.g. Lanczos3 and Spline3. Spline3 uses the old algorithm and performs exactly the same way as the old Lanczos3 had performed. Same with Lanczos4 <-> Spline4. If the new algorithms run faster for all of you, I can probably also convert the other resamplers (Spline, Mitchell, Bicubic etc) to the new algorithm logic to get some speed improvements. The speed improvement will probably be largest with Lanczos, due to how the math works, but there might be some benefit for the other resamplers, too.

(2) I've added an optional *anti-ringing filter*!! Currently it only works for Lanczos Luma upscaling (support for Spline and Bicubic could be added in a future version). In-depth discussion about the technical aspects of this new anti-ringing filter can be found here:

http://forum.doom9.org/showthread.php?t=145358

Please note that this anti-ringing filter does *not* remove ringing that is encoded into the source. If there is ringing in the source, even with this anti-ringing filter activated, you will still see ringing, and Lanczos has a tendency to make the ringing even more obvious. However, if you have a clean source, the anti-ringing filter should mostly avoid adding new ringing to the image. The filter is not perfect, it might sometimes miss some ringing. I've tried to find a good compromise between (a) removing ringing artifacts and (b) not making anything else worse. Just give it a try and let me know about your impressions. FWIW, the anti-ringing filter does eat some performance. But it's not terribly slow, so if your GPU is halfway decent, you should be able to use it.

Here are 2 comparison images from the Monsters Inc. Blu-Ray, upscaled to double size with Lanczos8 with vs. without anti-ringing filter. I've used Lanczos8 to make the artifacts more obvious. The ringing is not as strong with Lanczos3 and Lanczos4, of course, but it's definitely there, too.

http://madshi.net/monsters2.pnghttp://madshi.net/monsters1.png

[nevcairiel]

Good job, i was wondering when we would see your anti-ringing Lanczos in madVR. :)

Now 6233638 should re-check Lanczos 3/4 in his quality benchmark ;)

[SamuriHL]

WOA! Nice job. :) Can't wait to try that out when I get a free minute. Thanks!!

[noee]

HD6570, 12.7b driver.

No difference in load according to the GPU tab of Process Explorer, for Spline3/Lanczos3, Spline4/Lanczos4, upscaling a 480p -> 1080.

When I turn on anti-ringing, occasionally I can see some vertical line scaling artifacts and load GPU load jumps up about 5%.

[ryrynz]

Tested on a Intel HD 3000, Lanczos3 giving me 10% less GPU load then Spline3.

Enabling anti-ringing causes the picture to go blocky for me.

http://img16.imageshack.us/img16/7282/l3normal.th.png (http://imageshack.us/photo/my-images/16/l3normal.png/)

http://img827.imageshack.us/img827/2040/l3antiringing.th.png (http://imageshack.us/photo/my-images/827/l3antiringing.png/)

[vivan]

Same problem with both intel HD 3000 and nVidia GF 540M GPU's
http://3.firepic.org/3/images/2012-07/30/slrwr71jz2n3.png
http://3.firepic.org/3/images/2012-07/30/jygx0tezdpfj.png

[leeperry]

Hi Madshi, thanks for the new build!

I will try the anti-ringing option on my Spline3 SuperSampling downscaling and report back, but I was wondering: what would you say about forcing the tightest NT timer resolution using NtSetTimerResolution()?

You can change it via boot.ini on XPSP3 and it helped Reclock a hell lot IME(it could settle down to a low ppm clock much faster and in a much more stable way) but on Vista/W7/W8 you need to force it via NtSetTimerResolution() apparently. That's what the small Fidelizer app does(but you have to leave it open in the systray), and many ppl claim that the improvement is very much there:

http://tech.groups.yahoo.com/group/DynoMotion/message/5270?var=1

http://www.mail-archive.com/audiophiles@lists.slimdevices.com/msg57473.html
ntdll:NtSetTimerResolution (1,1,0xcb1933)

HPET with the highest granularity sounds like a winner :cool:

[shimaflarex]

The new Lanczos algorithm is way slower for me.
With the old algorithm, rendering times were about aprox. 23ms for a 480p video stretched to 1600x900.
With the new, it is about 44ms (anti-ringing turned off).
The anti-ringing feature seems to cause major blocking in the same situation (the source is anime btw).
My GPU is a Geforce 8500GT.

[ryrynz]

No change in present time for me, eagerly awaiting the improvement to the other resizers. :D

[Mangix]

windows 8 64-bit with latest nvidia drivers and a gts450. here are the results with Lanczos4:

original:
8.35ms

new:
6.5ms

new+anti-ringing:
10.2ms

nice improvement even though I don't use lanczos(I use spline). Spline is about 6.5ms.

I also noticed unsmooth text. As in, aliased. This was with the anti-ringing option. Could be similar to the intel issue.

edit: forgot to mention. linear light was also enabled.

[madshi]

Only replies about the new test build atm:

Now 6233638 should re-check Lanczos 3/4 in his quality benchmark ;)
FWIW, for bad sources he will probably still prefer SoftCubic because that nicely masks some of the source artifacts, while Lanczos (with or without anti-ringing filter) considers such artifacts details and makes them more visible. But for clean sources my hope is that Lanczos with anti-ringing might be best now.

HD6570, 12.7b driver.

No difference in load according to the GPU tab of Process Explorer, for Spline3/Lanczos3, Spline4/Lanczos4, upscaling a 480p -> 1080.
Could you please compare rendering times in the madVR OSD in windowed mode?

When I turn on anti-ringing, occasionally I can see some vertical line scaling artifacts and load GPU load jumps up about 5%.
Can I see a screenshot, please? And a small sample would be great, thanks!

Tested on a Intel HD 3000, Lanczos3 giving me 10% less GPU load then Spline3.

Enabling anti-ringing causes the picture to go blocky for me.

http://img16.imageshack.us/img16/7282/l3normal.th.png (http://imageshack.us/photo/my-images/16/l3normal.png/)

http://img827.imageshack.us/img827/2040/l3antiringing.th.png (http://imageshack.us/photo/my-images/827/l3antiringing.png/)
I can see a big problem in the frame counter, and a rather small problem in the "f" of the "Aufkäufer". Is the frame counter mainly what you're talking about? Where does the frame counter come from? Is it hard coded into the source? Is it from a DirectShow filter? And which zoom factor did you use?

Same problem with both intel HD 3000 and nVidia GF 540M GPU's
http://3.firepic.org/3/images/2012-07/30/slrwr71jz2n3.png
http://3.firepic.org/3/images/2012-07/30/jygx0tezdpfj.png
Ouch, that looks nasty. Can I have a small sample, please? Which zoom factor did you use?

The new Lanczos algorithm is way slower for me.
With the old algorithm, rendering times were about aprox. 23ms for a 480p video stretched to 1600x900.
With the new, it is about 44ms (anti-ringing turned off).
The anti-ringing feature seems to cause major blocking in the same situation (the source is anime btw).
My GPU is a Geforce 8500GT.
Interesting. Seems the 8500GT is veeeery slow with math. Can I get a small sample which shows the blocking and maybe a screenshot?

edit: forgot to mention. linear light was also enabled.
Forgot to say that linear light scaling is not working for the Lanczos upscalers in that test build. Could you please re-compare performance with linear light scaling disabled?

-------

Thanks, guys. Will probably not have time to do anything about this for the next 2 weeks, but I appreciate the feedback.

[ryrynz]

I can see a big problem in the frame counter, and a rather small problem in the "f" of the "Aufkäufer". Is the frame counter mainly what you're talking about? Where does the frame counter come from? Is it hard coded into the source? Is it from a DirectShow filter? And which zoom factor did you use?


The whole image is blocky when anti-ringing is enabled, it's the same one as vivan posted about.

The frame counter is ffdshow's OSD, I had that on to make sure I grabbed the exact same frame. FWIW the blockiness happens even if ffdshow isn't present in the chain.


Can I get a small sample which shows the blocking and maybe a screenshot?


That would also be the same issue, happens with everything I've tested so far.

[mzso]

http://madshi.net/madVRantiRinging.rar

This is just a test version, no official build and it only contains the changed files. So copy these files over your 0.82.5 installation. There are 2 key changes in this test build:

(1) The Lanczos Luma upscaling algorithms were totally rewritten to use more math and less texture accesses. On my ATI 3850 they now run about 25-30% faster than before. It would be nice if you guys could double check if they run faster for you, too, on all the funny GPUs you have. Please also check if image quality is alright. For speed comparisons you can simply switch between e.g. Lanczos3 and Spline3. Spline3 uses the old algorithm and performs exactly the same way as the old Lanczos3 had performed. Same with Lanczos4 <-> Spline4. If the new algorithms run faster for all of you, I can probably also convert the other resamplers (Spline, Mitchell, Bicubic etc) to the new algorithm logic to get some speed improvements. The speed improvement will probably be largest with Lanczos, due to how the math works, but there might be some benefit for the other resamplers, too.

(2) I've added an optional *anti-ringing filter*!! Currently it only works for Lanczos Luma upscaling (support for Spline and Bicubic could be added in a future version). In-depth discussion about the technical aspects of this new anti-ringing filter can be found here:
[/IMG]

Doesn't any of these have sense for chroma? I usually don't do luma upsampling since I mostly watch 1080p. (Occasionally 720p but I didn't see anything noticable with bicubic)

BTW: What is linear light scaling good for, sure it was mentioned but I forgot...

[vivan]

Ouch, that looks nasty. Can I have a small sample, please? Which zoom factor did you use?It happens on any video (of any type, not just anime) with any scaling factor. With higher scailng factor it even more noticeable.

240p -> 768p
http://4.firepic.org/4/images/2012-07/30/070mg4p4cr4n.png
http://4.firepic.org/4/images/2012-07/30/jm48scksndvj.png

418p -> 768p
http://4.firepic.org/4/images/2012-07/30/72a6q02qm8j2.png
http://4.firepic.org/4/images/2012-07/30/2l5ifhsrbpae.png

720p -> 768p
http://4.firepic.org/4/images/2012-07/30/1sr5e16r19zs.png
http://4.firepic.org/4/images/2012-07/30/ocqx4h0xqf69.png

I don't think sample can help, but anyway - http://www.mediafire.com/?5sm8qxhgx208gj4

I can see a big problem in the frame counter, and a rather small problem in the "f" of the "Aufkäufer"There's same problem (aliasing?) everywhere. E.g. 4x zoom of fragment of his image:
http://3.firepic.org/3/images/2012-07/30/o9ef8p1h7hc7.png
http://3.firepic.org/3/images/2012-07/30/euyz8opgfkat.png
I'm not sure - maybe there is something with your display, because it's very noticeable...

And that's what I get with your image (2x downscaled and than upscaled again with madVR):
http://2.firepic.org/2/images/2012-07/30/5nup9zcambix.png
http://2.firepic.org/2/images/2012-07/30/th2mkobmpyaf.png
I don't see such "effect" on your comparison, so it's clearly a bug (maybe some special settings are needed, but mine are all default).

[ryrynz]

Tested on my HD 4000 and my Radeon 5850 as well. Blockiness abound when anti-ringing is enabled.

maybe there is something with your display, because it's very noticeable...


He has been out of madVR development for awhile... ;)

[madshi]

Just to confirm: The new Lanczos algorithm *without* anti-ringing filter works fine, and the blocking only occurs if you activate the anti-ringing filter, correct? Look here:

http://madshi.net/Tantei Gakuen Q.png

This is how your anime source looks on my PC when upscaling with anti-ringing enabled. It seems that the problem is not a logic fault in the anti-ringing filter. Instead it seems that the pixel shader code the HLSL compiler outputs totally fails to work correctly with your GPU/OS, while it works in my situation. This is confusing, don't know how that can happen. HLSL is supposed to create pixel shader code which works on all OSs and GPUs. So the big question for me now is how I can reproduce the problem, because it doesn't happen on my PC. Will check my HTPC later...

Those of you who get this blocking issue, which OS and GPU are you using? And those who DON'T get this blocking issue, which OS and GPU are you using? We'll need to find out whether the blocking issue depends on the OS or the GPU, or the combination of both.

[ryrynz]

Those of you who get this blocking issue, which OS and GPU are you using? And those who DON'T get this blocking issue, which OS and GPU are you using? We'll need to find out whether the blocking issue depends on the OS or the GPU, or the combination of both.

Having this issue across three PCs all running Windows 7 Ultimate x64 using HD 3000, HD 4000 and Radeon HD5850 graphics.

And yes without anti-ringing, everything is fine.

[madshi]

Ok, so maybe it depends on the OS? I've developed and tested this on XPSP3.

[ryrynz]

Sounds logical. Time to upgrade Madshi :) unfortunately I don't have an XP system to test on.

[vivan]

Just to confirm: The new Lanczos algorithm *without* anti-ringing filter works fine, and the blocking only occurs if you activate the anti-ringing filter, correct?Yes.

Look here:It's awesome!

Those of you who get this blocking issue, which OS and GPU are you using?I have 2 PC, both are W7 x64. GPU's are Intel HD 3000 and nVidia 540M on notebook, ati 5870 on desktop.

UPD: I don't have XP here too...

About speed. I'm using windowed mode, 1366x768 is screen resoultion. GPU is Intel HD 3000 integrated into i5-2410M CPU.
Scaling settings are default (no linear light, lanczos4 for luma, softcubic100 for chroma).
With v0.82.5 I get about 16.6 ms rendering time (on the edge of 1/60), so 60 fps video has only few drops.
With test version I get about 18.1 ms rendering time, so 60 fps video lags...

[cyberbeing]

Anti-ringing causes the picture to block on my NVIDIA GT 440 DDR5 (@ 864Mhz Core | 1728Mhz RAM) as well.

Win7 SP1 x64
304.48 Driver

Test build Lanczos is also ~33-50% slower (higher GPU load) compared to 0.82.5 Lanczos, which isn't good.

Lanczos3 0.82.5: 15% GPU Load
Lanczos3 Test: 21% GPU Load

Lanczos4 0.82.5: 18% GPU Load
Lanczos4 Test: 24% GPU Load

Lanczos8 0.82.5: 24% GPU Load
Lanczos8 Test: 35% GPU Load

720x480 -> 1600x900

[sneaker_ger]

I, too, get the blocking with activated anti-ringing filter:
Win 7 x64
HD 5850
Catalyst 12.6

[ryrynz]

Looks like it may be worth keeping a legacy Lanczos option.

[cyberbeing]

Lanczos 3 Anti-ringing filter bug (http://img823.imageshack.us/img823/69/l3antiring.png)

Lanczos 3 Normal (http://img688.imageshack.us/img688/7282/l3normal.png)

I re-encoded the '4-Star Chart.mp4' test pattern to 960x540, and played it back @1600x900 on my monitor.

[madshi]

cyberbeing, XP or win7? Anybody still on XP to double check?

[madshi]

NVIDIA GT 440 DDR5 (@ 864Mhz Core | 1728Mhz RAM)

Win7 SP1 x64
304.48 Driver

Test build Lanczos is also 33-50% slower (higher GPU load) compared to 0.82.5 Lanczos, which isn't good.
Is that with anti-ringing or without? I would have expected a 440 to be faster with the new Lanczos algorithm. What do the madVR OSD average render times say, in windowed mode?

[cyberbeing]

That's without the anti-ringing.

8ms with the Test Build using Lanczos3 720x480->1600x900

6ms with 0.82.5 using Lanczos3 720x480->1600x900

cyberbeing, XP or win7?
Those images were taken on Win7 x64.

[Hprd]

I get the blocking too. Win 7 SP1 64 bit Geforce 275 w/ 301.42 drivers. Doesn't matter if I have madvr decode the video, or ffdshow. I even reset madvr to defaults with the .bat file after I updated to the test build.
Is there anybody here where this anti-ringing thing actually works (aside from Madshi of course, lol)? :(

[ryrynz]

The appropriate question is, is there anyone here besides Madshi that's still running a 10 year old OS? :D

leeperry where are you!? :)

[chros]

Anti-ringing causes the picture to block on my NVIDIA GT 440 DDR5 (@ 864Mhz Core | 1728Mhz RAM) as well.

Win7 SP1 x64
304.48 Driver

Test build Lanczos is also ~33-50% slower (higher GPU load) compared to 0.82.5 Lanczos, which isn't good.


The new Lanczos algorithm is way slower for me.
With the old algorithm, rendering times were about aprox. 23ms for a 480p video stretched to 1600x900.
With the new, it is about 44ms (anti-ringing turned off).
The anti-ringing feature seems to cause major blocking in the same situation (the source is anime btw).
My GPU is a Geforce 8500GT.

Same here for me too (for specs see my signature with desktop config):

MadVR 8.2.5 Lanczos4:
Formula1 720p 50fps GPU: 56-58% FPU: 40-44% MCU: 17-18%
Awake 720p 23.976fps GPU: 28-29% FPU: 15-20% MCU: 10%
Planetes 720p 23.796fps GPU: 27-28% FPU: - MCU: 8-9%

MadVR 8.2.5 Lanczos4_new:
Formula1 720p 50fps GPU: 78-80% FPU: 40-44% MCU: 17-18%
Awake 720p 23.976fps GPU: 47-48% FPU: 15-20% MCU: 10%
Planetes 720p 23.796fps GPU: 45-46% FPU: - MCU: 8-9%

MadVR 8.2.5 Lanczos4_new_anti-ringing:
Awake 720p 23.976fps GPU: 83-87% FPU: 15-20% MCU: 10%

[vivan]

I've made request on one tracker, so here is comment from there (http://rutracker.org/forum/viewtopic.php?p=54430915#54430915).

User with Windows XP SP3 (and GeForce 8600 GT GPU) experiences the same problem:
http://t.imgbox.com/abtttS85.jpg (http://imgbox.com/abtttS85) http://t.imgbox.com/adfEjl2q.jpg (http://imgbox.com/adfEjl2q)

[ryrynz]

Guess that's that theory out the window. Could be something up with Madshi's system.

[madshi]

Weird. I guess I should be able to reproduce the problem on my HTPC then. I'll try later...

[sexus]

its true just applied and tested once i activate antiringing i get hell a ton of blockiness , hope this can be fixed id like the ringing reduction, this working and lanczos 3 would be the new top dog in highest quality filter im a lanczos 3 user its the best choice except for the luma downscaling wich is set to bicubic sharpness 50 as recommended by ranpha the others are set to lanczos3

p.s> what is scale in linear light and what good does it do, should i use it?

[ryrynz]

what is scale in linear light and what good does it do, should i use it?

Search will help you out with questions like these. :)

It scales luma in linear light and affects the contrast of the picture.
It's one of those options that is there for people who generally know if they need it. Your best leaving it unticked.

[MokrySedeS]

For now I could only try with my integrated Radeon HD 3100.
Obviously it's too slow to be usable with madVR @ Lanczos (massive framedrops) but for what it's worth the anti-ringing filter seems to work fine.
I'm on Windows 7 Professional x64 SP1 & Catalyst drivers v12.3
I'll try with GTS 450 when I get it back at the end of the week.

anti-ringing off vs on
http://thumbnails58.imagebam.com/20378/a7e952203775868.jpg (http://www.imagebam.com/image/a7e952203775868) http://thumbnails39.imagebam.com/20378/750fce203776012.jpg (http://www.imagebam.com/image/750fce203776012)

[mzso]

Weird I something got really screwed here... Now everything even with going back to the latest version and settings.

Edit:
Okay I needed a restart... Happened a couple times, but not lately... (Even the LAV cuda decodr was too slow, stupid NVIDIA)

Edit2:
OK. Now I tested it properly the blockiness happens here too... win7 AMD64, GTS 450

[mzso]

Could someone enlighten me what does Scale in Linear Light do? Does it effect output?

[sexus]

thanks ill leave it alone then ryrynz

[clancy688]

Not related to the anti-ringing filter discussion:

Is there a *quick* way to enable/disable fullscreen exclusive mode?

Starting a video -> right click -> filters -> madVR -> Edit Settings -> Rendering -> General Settings

That's half a world cruise... When using the main monitor, I don't need FSE mode, when using the TV, I do.
So... is there something like a shortkey? ^^;

[madshi]

I was able to reproduce the blockiness of the anti-ringing algorithm on my HTPC. This build should (hopefully) fix it:

http://madshi.net/madVRantiRinging2.rar

I've also noticed that on my HTPC my NVidia 9400 mainboard is quite a bit slower with the new Lanczos algorithm compared to the old one. It seems that the "sin" (sine) math instruction executes very slow on this GPU, and I need to call it 12 times per pixel when using Lanczos 3. The "sin" calls alone slow down the new Lanczos algorithm so much that it's slower than the old solution although the old solution needed more texture/memory reads. However, my ATI 3850 is faster when using the new algorithm. So now I'm thinking ATI seems to do "sin" much faster than NVidia. So maybe for ATI users the new algorithm is faster, and for NVidia users the old. Can you confirm? Does anybody have a Fermi or Kepler card to double check performance?

[SamuriHL]

I have a 680 but I have limited time to look at it right now. I've been busy on another project and have neglected my HTPC's sadly. If I can get some time later I can try and test it.

[noee]

HD6570.

Yes, fixed here, no more artifacts with anti-ringing enabled. Looks good too on the source I'm testing with (TinTin movie).

Re: performance, I still don't see a difference in render (ms) on Spline vs. Lanczos. Could the CPU make a difference? I run my cpu (Athlon II, using K10stat) at 700Mhz (severely undervolted) when playing SD material.