Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Hardware & Software > Software players

Reply
 
Thread Tools Search this Thread Display Modes
Old 11th December 2016, 21:35   #41401  |  Link
BluesFanUK
Registered User
 
Join Date: Sep 2015
Posts: 60
Quote:
Originally Posted by Asmodian View Post
It is supposed to drop frames on open or seek, that is normal. You can reduce the number with the rendering -> general settings option "Delay playback start until render queue is full". The frames dropped on playback or seek probably aren't important though, I prefer faster response times without that setting even though it drops a lot of frames during open/seeks.
Odd. It annoyed me so much last year to the point I replaced my 270x with a 980ti. Elimated the issue immediately.

Now it's back, and I never had a delay on it before. Ah well, have to live with it now I guess.
BluesFanUK is offline   Reply With Quote
Old 11th December 2016, 21:43   #41402  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
I had a 980Ti until a few months ago, I don't think I ever saw 0 dropped frames on open or seek.
I didn't try to optimize settings to achieve that though... I usually run pretty aggressive settings.

Though it does seem to drop more during open/seeks now than it used to.
__________________
madVR options explained
Asmodian is offline   Reply With Quote
Old 11th December 2016, 21:48   #41403  |  Link
flashmozzg
Registered User
 
Join Date: May 2013
Posts: 77
Quote:
Originally Posted by Asmodian View Post
Can you use NGU-high for luma? I really do recommend leaving everything on "let madVR decide" and picking the highest luma option you can. Once you have maxed out luma you can start turning up the other settings a little, if you still have performance available.

Edit2:
The point of "let madVR decide" is to use reasonable options given the performance the choice for luma implies. Anything below NGU-veryHigh uses Bicubic60 AR for chroma doubling, which is a very reasonable option for chroma doubling.
NGU-high increases avg rendering stats slightly (to 6-7 ms). Very-high though ramps up them to 30-38ms.

Btw, I'm a little confused hear with chorma uspacling. There is two separate options for chroma upscaling - chroma upscaling tab and chroma-upscaling option in image upscaling tab. And If I chose to sue ngu med for chroma in both tabs I get:
chroma > NGU-med
luma > ...
chroma > NGU-med

seems wrong. Is it supposed to happen?
flashmozzg is offline   Reply With Quote
Old 11th December 2016, 22:04   #41404  |  Link
Neo-XP
Registered User
 
Neo-XP's Avatar
 
Join Date: Mar 2016
Location: Switzerland
Posts: 140
Quote:
Originally Posted by Asmodian View Post
The point of "let madVR decide" is to use reasonable options given the performance the choice for luma implies. Anything below NGU-veryHigh uses Bicubic60 AR for chroma doubling, which is a very reasonable option for chroma doubling.
Lanczos3 AR is better for this, without any significant impact on resources (1% more charge on my GTX 970 for FHD content), but you can't choose it anymore on new versions.
Lanczos3 AR should be the "normal/low" option and Bicubic60 AR the "low/very low" one, or remove it completely.

Seriously, I hope that madshi will gives us a little more choices for chroma doubling. There is too much hate on chroma... and I need to test more combinations
The new users can always use the "let madVR decide" option anyway.
Neo-XP is offline   Reply With Quote
Old 11th December 2016, 22:13   #41405  |  Link
BluesFanUK
Registered User
 
Join Date: Sep 2015
Posts: 60
Quote:
Originally Posted by Asmodian View Post
I had a 980Ti until a few months ago, I don't think I ever saw 0 dropped frames on open or seek.
I didn't try to optimize settings to achieve that though... I usually run pretty aggressive settings.

Though it does seem to drop more during open/seeks now than it used to.
I think it's got to be something in the recent updates, because I know for certain I got rid of all dropped frames and use settings that pushed the GPU to its limits. I'm not too conerned with power consumption, just the setting that gives me the best results.
BluesFanUK is offline   Reply With Quote
Old 11th December 2016, 22:14   #41406  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by Asmodian View Post
I am not sure what the "kill my card" is about but in their newer drivers Nvidia changed the default power management mode from "adaptive" to "optimal power". Optimal power clocks too low for madVR though so you need to change back to adaptive, after I do this I don't get any dropped frames with the newest madVR at ~85% GPU usage.

Edit:


How odd, though I don't use reduce ringing artifacts normally it only adds ~2.3ms on my system. I don't think the Titan X (pascal) is 4x faster than the 980Ti.



Can you use NGU-high for luma? I really do recommend leaving everything on "let madVR decide" and picking the highest luma option you can. Once you have maxed out luma you can start turning up the other settings a little, if you still have performance available.

Edit2:
The point of "let madVR decide" is to use reasonable options given the performance the choice for luma implies. Anything below NGU-veryHigh uses Bicubic60 AR for chroma doubling, which is a very reasonable option for chroma doubling.
I agree, the best strategy for new users is to maximize luma quality and let madVR decide everything else. I use this strategy on my own setup. Choosing let madVR decide saves so much performance. I am able to run NGU high on all sources with a GTX 750 Ti without any tinkering or advanced knowledge.

The average user may never get that far and give up for NGU medium or low without knowing they left performance on the table.

Last edited by Warner306; 11th December 2016 at 22:49.
Warner306 is offline   Reply With Quote
Old 11th December 2016, 22:19   #41407  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by Neo-XP View Post
Lanczos3 AR is better for this, without any significant impact on resources (1% more charge on my GTX 970 for FHD content), but you can't choose it anymore on new versions.
Lanczos3 AR should be the "normal/low" option and Bicubic60 AR the "low/very low" one, or remove it completely.

Seriously, I hope that madshi will gives us a little more choices for chroma doubling. There is too much hate on chroma... and I need to test more combinations
The new users can always use the "let madVR decide" option anyway.
Is the difference between Bicubic60 and Lanczos3 enough to justify any reduction in performance?
Warner306 is offline   Reply With Quote
Old 11th December 2016, 22:39   #41408  |  Link
Crimson Wolf
Registered User
 
Join Date: Dec 2014
Posts: 51
Quote:
Originally Posted by BluesFanUK View Post
I think it's got to be something in the recent updates, because I know for certain I got rid of all dropped frames and use settings that pushed the GPU to its limits. I'm not too conerned with power consumption, just the setting that gives me the best results.
FYI: Nvidia's recent drivers have made even "adaptive" power settings glitchy with madvr+mpc-hc with low power modes when playing certain files on my system. I can recreate the low power bug with a H.265 8-bit file. Currently, I've set MPC-HC64 to use "prefer maximum performance" because of this problem. It does not matter if you choose higher load on madvr, the driver will not turn up the power mode from optimal or adaptive. Some older drivers might have worked on "adaptive" settings, but the most recent one (376.19) on my gtx980 will only do the 650mhz gpu/800mhz mem if you do not manually change the setting.
Crimson Wolf is offline   Reply With Quote
Old 11th December 2016, 22:48   #41409  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
are you using DXVA decoding?
huhn is offline   Reply With Quote
Old 11th December 2016, 22:53   #41410  |  Link
Crimson Wolf
Registered User
 
Join Date: Dec 2014
Posts: 51
Quote:
Originally Posted by huhn View Post
are you using DXVA decoding?
LAV set to dxva-cb. I think gtx980 has a hybrd h.265 8-bit decoder, but not 10-bit.
Crimson Wolf is offline   Reply With Quote
Old 11th December 2016, 23:04   #41411  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
Quote:
Originally Posted by flashmozzg View Post
NGU-high increases avg rendering stats slightly (to 6-7 ms). Very-high though ramps up them to 30-38ms.

Btw, I'm a little confused hear with chorma uspacling. There is two separate options for chroma upscaling - chroma upscaling tab and chroma-upscaling option in image upscaling tab. And If I chose to sue ngu med for chroma in both tabs I get:
chroma > NGU-med
luma > ...
chroma > NGU-med

seems wrong. Is it supposed to happen?
Chroma is stored in standard video at half the resolution of the luma. The "chroma upscaling" page is for upscaling the chroma to the same resolution as the luma. This is reasonably important and a good algorithm here is not wasting GPU power.

The chroma upscaling option in image upscaling is for scaling the chroma to the destination resolution, if that is above the luma resolution. This option is much less important, you have already doubled the chroma resolution so there isn't much fine detail to preserve.

This is why you get the two chroma entries in the OSD, the first one is reasonably important but the second one isn't. I wouldn't use NGU-med for chroma doubling unless I was already using NGU-veryHigh for luma doubling.

All the chroma scaling comparisons posted recently are for the first chroma upscaling option, it is hard to find samples which show obvious differences with the chroma options in image upscaling.
__________________
madVR options explained
Asmodian is offline   Reply With Quote
Old 11th December 2016, 23:25   #41412  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by Crimson Wolf View Post
LAV set to dxva-cb. I think gtx980 has a hybrd h.265 8-bit decoder, but not 10-bit.
yes and this hybrid decoder is making problem it is using the GPU to decode the video and is increasing the rendertimes.

try to disable HEVC hardware decoding in lavfilter a cpu is usually not really slower but doesn't stress the GPU.
huhn is offline   Reply With Quote
Old 11th December 2016, 23:32   #41413  |  Link
arrgh
Registered User
 
Join Date: Dec 2007
Posts: 125
MPC-BE; LAV-Filters; xySubFilter; madVR;

give Problems with Subtitles (idx/sub) and cropped h264 Video packed in a MKV:

I have chosen in madVR (under "zoom control") to move the subtitles to the bottom of the screen/window;
but in the case of the cropped Video the Subs remain always within the activ/Video area (I'm in full Screen, not windowed);
also in xysubfilter the "override Placement" Option does not allow to leave the active Video area;

is that a bug, or am I missing something?

Thanks
arrgh is offline   Reply With Quote
Old 11th December 2016, 23:41   #41414  |  Link
Crimson Wolf
Registered User
 
Join Date: Dec 2014
Posts: 51
Quote:
Originally Posted by huhn View Post
yes and this hybrid decoder is making problem it is using the GPU to decode the video and is increasing the rendertimes.

try to disable HEVC hardware decoding in lavfilter a cpu is usually not really slower but doesn't stress the GPU.
Render times aren't the issue here. The GPU refuses to go higher than low power mode when I play the file. When I choose "prefer maximum performance" there is no issue. If I had not monitored my gpu core speed and memory, I would have thought something was wrong.
Crimson Wolf is offline   Reply With Quote
Old 11th December 2016, 23:44   #41415  |  Link
nevcairiel
Registered Developer
 
Join Date: Mar 2010
Location: Hamburg/Germany
Posts: 10,343
Quote:
Originally Posted by Crimson Wolf View Post
Render times aren't the issue here. The GPU refuses to go higher than low power mode when I play the file. When I choose "prefer maximum performance" there is no issue. If I had not monitored my gpu core speed and memory, I would have thought something was wrong.
Try the suggestion anyway, hybrid decoders can sometimes do funny things to your GPU.
__________________
LAV Filters - open source ffmpeg based media splitter and decoders
nevcairiel is online now   Reply With Quote
Old 11th December 2016, 23:55   #41416  |  Link
Crimson Wolf
Registered User
 
Join Date: Dec 2014
Posts: 51
Quote:
Originally Posted by nevcairiel View Post
Try the suggestion anyway, hybrid decoders can sometimes do funny things to your GPU.
Hmm. Can Nvidia do something to my system with the GFE optimization updates? I swear I check every driver release with these tests and I was able to recreate this low power bug repeatedly even after reboots.

Now, I can not activate the bug even with "optimal" AND H/W decode enabled. I haven't checked if s/w decode would have worked or not but right now, everything is working...

I will investigate further. I'm very confused now.

EDIT: one thing that has changed is, I updated LAV to the latest nightly.
Crimson Wolf is offline   Reply With Quote
Old 12th December 2016, 00:24   #41417  |  Link
Patrik G
Registered User
 
Join Date: Jan 2015
Location: Sweden
Posts: 58
Quote:
Originally Posted by Warner306 View Post
I agree, the best strategy for new users is to maximize luma quality

you get an artificially created sharpness boost with Luma Doubling why do you want to use it?
its easier to bump up the sharpness on your display instead and you get the same effect with moire patterns all over the place.

run the shaprness and oversacan file from the avshd709 disc and you will see how shitty luma doubling looks.
but you decide if you want to wreck the image or not

or you have the sharpness and overscan test here
http://privat.bahnhof.se/wb192876/5-...20Overscan.mp4

for best result use Chroma Upscaling Only or visit specsavers optics for new glasses if its not sharp enough

Last edited by Patrik G; 12th December 2016 at 00:33.
Patrik G is offline   Reply With Quote
Old 12th December 2016, 00:27   #41418  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by Patrik G View Post
you get an artificially created sharpness boost with Luma Doubling why do you want to use it?
its easier to bump up the sharpness on your display instead and you get the same effect with moirie patterns all over the place.

run the shaprness and oversacan file from the avshd709 disc and you will see how shitty luma doubling looks.
but you decide if you want to wreck the image or not
That is condemnation of madVR, alright.

To each their own. I like the sharpness and know that ordinary image upscaling cannot reproduce sharp enough images, let alone edges that are too sharp. I'd rather border on too sharp than not sharp enough.

Edit: I didn't know we were talking about 200% supersampling. Of course, that is less than ideal.

Last edited by Warner306; 12th December 2016 at 18:22.
Warner306 is offline   Reply With Quote
Old 12th December 2016, 00:56   #41419  |  Link
Crimson Wolf
Registered User
 
Join Date: Dec 2014
Posts: 51
Quote:
Originally Posted by Crimson Wolf View Post
Hmm. Can Nvidia do something to my system with the GFE optimization updates? I swear I check every driver release with these tests and I was able to recreate this low power bug repeatedly even after reboots.

Now, I can not activate the bug even with "optimal" AND H/W decode enabled. I haven't checked if s/w decode would have worked or not but right now, everything is working...

I will investigate further. I'm very confused now.

EDIT: one thing that has changed is, I updated LAV to the latest nightly.
I'm sorry, but I can not recreate this bug anymore... I've tried several previous versions of LAV filter and madvr 0.91.3 as well too. If I encounter this bug again, I'll make sure to note the versions as well as winver. There was a cumulative update on win10 recently and I'm not sure if that has had any effect on this bug. I am positive that I was not the only person with this low power bug in mpc-hc with madvr, but as of now, my system is working as it should. I've disabled HEVC h/w decoding as suggested.
Crimson Wolf is offline   Reply With Quote
Old 12th December 2016, 01:35   #41420  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
Quote:
Originally Posted by Patrik G View Post
you get an artificially created sharpness boost with Luma Doubling why do you want to use it?
its easier to bump up the sharpness on your display instead and you get the same effect with moire patterns all over the place.

run the shaprness and oversacan file from the avshd709 disc and you will see how shitty luma doubling looks.
but you decide if you want to wreck the image or not

or you have the sharpness and overscan test here
http://privat.bahnhof.se/wb192876/5-...20Overscan.mp4

for best result use Chroma Upscaling Only or visit specsavers optics for new glasses if its not sharp enough
This is a good example of why 200% supersampling isn't necessarily a good idea, and is probably why madshi resisted adding a 200% supersampling option for so long. However, NGU-veryHigh looks much better than Cubic, Jinc, etc. with that video viewed at 2560x1440 or 3840x2160.

Luma doubling is not the issue in this example, messing with an image that is already at the display resolution is the problem. If you want a single pixel black and white checkerboard to look the same then don't use 200% super sampling.
__________________
madVR options explained
Asmodian is offline   Reply With Quote
Reply

Tags
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 17:40.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.