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 16th October 2018, 18:27   #53281  |  Link
422415
Registered User
 
Join Date: Jun 2015
Posts: 5
Can anyone tell me what I'm doing wrong? I cannot get RGB Full 12bpc set in NVIDIA settings for 3840x2160 23hz. I have tried setting it to YCbCr422 then changing it after that but it always resets back to YCbCr422 limited. I am on Windows 1803 and have tried multiple NVIDIA drivers to no avail.
422415 is offline   Reply With Quote
Old 16th October 2018, 21:05   #53282  |  Link
mytbyte
Registered User
 
Join Date: Dec 2016
Posts: 212
@Warner306:

Here are the graphs I promised for the monitor, can't do the TV at this time:

LG_IPS235_graph

I performed new measurements because white balance drifted a bit so peak brightness is actually 175 cd/m2 (and we complained about plasma brightness). I also used MadVR 0.92.17 and it seems math is a bit different than before, making my previous comments irrelevant. Color tweaks and highlight recovery are set to disabled/none, not to interfere with BT.2390 math.

Screenshot titles are self explanatory but I'll recap:

LG_IPS235_175nits_Gamma2.2.png - this is the measured gamma of the monitor when set to 2.2 preset (no 10-point adjustment available) and measured with HCFR's internal automatic SDR patterns nits to check if gamma is tracking well even at peak brightness

LG_IPS235_175nits_PQ.png - this is the PQ tracking (MadVR set to clipping), manual HDR pattern advance

LG_IPS235_175nits_BT.2390.png - this is the BT.2390 tracking when HCFR reference settings are set to default

LG_IPS235_100nits_BT.2390_MinL=0.13_MaxL=10000.png - this is the BT.2390 tracking when MadVR is set to minimum allowed peak brightness of 100 nits and Master MinL is set to monitor's measured black level, while Master MaxL is set to theoretical maximum of 10000 nits

provided HCFR math is correct, it seems Master MinL and MaxL values affect the reference BT.2390 curve and make MadVR's math, as measured, track closer/smoother. BT.2390 tracking drifts significantly at 50 & 60% stimulus even though the measured gamma the BT.2390 curve is transposed to is "perfect"..

perhaps the measured black level value should also be added to MadVR's math?

As for your patterns, I can see very faint flashing at the right side of the screen with "Black Level 10-bit" as well as slightly lighter right side compared to left side with "Black level 8-bit/10-bit combined".

Last edited by mytbyte; 16th October 2018 at 22:11.
mytbyte is offline   Reply With Quote
Old 16th October 2018, 22:20   #53283  |  Link
Manni
Registered User
 
Join Date: Jul 2014
Posts: 942
Quote:
Originally Posted by nevcairiel View Post
Everything works with Software decoding or DXVA2/D3D11 Copy-Back (because copy-back looks like software decoding to a renderer). And between those last two, DXVA2 Copy-Back is usually preferable because its more efficient, unless you need D3D11 to access a headless GPU.
I've done some tests and was quite surprised with the results.

Playing 4K23 content (chroma NGU High and minor enhancements), I got:

DXVA2 Native: 12.5ms
DXVA2 CB: 21ms

D3D11 Native: 16ms
D3D11 CB: 22ms

While I was aware of the performance loss between D3D11 native and CB, I would have expected DXVA2 CB to do better than that. There isn't much of a performance gain between DXVA2 CB and D3D11 CB, at least here.

DXVA2 native produces the most significant gain, and the only one that would make a difference for me, as it means I could play 4K60 in the same quality. Unfortunately, not an option for me due to the lack of black bars detection.

Weirder, with 1080p23 content (NGU High chroma, NGU very high luma), all modes give around 22ms, there doesn't seem to be any performance gain going native.

So I'm not sure what the advantage is to use DXVA2 CB vs D3D11 CB if there is little to no performance gain.

Hopefully we'll get one of the native modes to support black bars detection at some point, without losing too much performance and getting to CB level.

I guess it's good to know that I can go back to DXVA2 native if at some point I need the performance as I can always shift my picture with the mechanical lens shift instead of doing it electronically with MadVR. Black bars detection is the only thing I would miss I think going native.

Thanks again for clarifying things for me and for prompting this re-evaluation.
__________________
Win11 Pro x64 b23H2
Ryzen 5950X@4.5Ghz 32Gb@3600 Zotac 3090 24Gb 551.33
madVR/LAV/jRiver/MyMovies/CMC
Denon X8500HA>HD Fury VRRoom>TCL 55C805K

Last edited by Manni; 16th October 2018 at 22:27.
Manni is offline   Reply With Quote
Old 16th October 2018, 22:21   #53284  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quick question to 1050 Ti users: Is the 1050 Ti fast enough to do 4Kp60 HDR tone mapping with all the bells and whistles (measurements + highlight recovery + "compromise" disabled)?
madshi is offline   Reply With Quote
Old 16th October 2018, 22:24   #53285  |  Link
Manni
Registered User
 
Join Date: Jul 2014
Posts: 942
Quote:
Originally Posted by madshi View Post
Quick question to 1050 Ti users: Is the 1050 Ti fast enough to do 4Kp60 HDR tone mapping with all the bells and whistles (measurements + highlight recovery + "compromise" disabled)?
The 1080ti isn't able to do 4K60 HDR with all the bells and whistles (see my post just above yours), unless you use DXVA2 native (in which case you lose black bars detection), so I doubt the 1050ti can. Config in my sig.
__________________
Win11 Pro x64 b23H2
Ryzen 5950X@4.5Ghz 32Gb@3600 Zotac 3090 24Gb 551.33
madVR/LAV/jRiver/MyMovies/CMC
Denon X8500HA>HD Fury VRRoom>TCL 55C805K

Last edited by Manni; 16th October 2018 at 22:37.
Manni is offline   Reply With Quote
Old 16th October 2018, 22:47   #53286  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
When talking about bells and whistles, I meant in terms of HDR tone mapping. I don't consider NGU Chroma upscaling to be crucial. So, using e.g. D3D11 native decoding, with default chroma upscaling (Bicubic), and no other fancy options activated, can the 1050 Ti do 4Kp60 HDR tone mapping with measurement + highlight recovery?
madshi is offline   Reply With Quote
Old 16th October 2018, 23:04   #53287  |  Link
Ver Greeneyes
Registered User
 
Join Date: May 2012
Posts: 447
Isn't DXVA2 Native also limited in what further processing madVR can do? Is it "just" chroma upscaling and black bars detection?
Ver Greeneyes is offline   Reply With Quote
Old 16th October 2018, 23:21   #53288  |  Link
Manni
Registered User
 
Join Date: Jul 2014
Posts: 942
Quote:
Originally Posted by madshi View Post
When talking about bells and whistles, I meant in terms of HDR tone mapping. I don't consider NGU Chroma upscaling to be crucial. So, using e.g. D3D11 native decoding, with default chroma upscaling (Bicubic), and no other fancy options activated, can the 1050 Ti do 4Kp60 HDR tone mapping with measurement + highlight recovery?
I agree NGU chroma isn't necessary, but isn't black bar detection essential for many? You lose that with native.

Otherwise yes with bicubic and native D3D11, I would expect the 1050ti to handle 4K60, but that means home cinema use is excluded, unless you find a way to support black bar detection with D3D11 native.

By the way I've just run a comparison in power use, and D3D11 native is far more efficient than copy back. In only requires around 50% CPU and 65% GPU for 4K60p with NGU chroma medium at 4K60p, while copy back requires 90% GPU and 100% CPU. This means a lot of unnnecessary heat! [EDIT: these measurements are wrong, I had Teamviewer running in the background! See this post for actual measurements)].

The difference is similar between DXVA2 native and copyback, but I seem to remember that there was some banding issues with DXVA2 native, so it looks like D3D11 native would be the way to go, provided there is black bar detection support.
__________________
Win11 Pro x64 b23H2
Ryzen 5950X@4.5Ghz 32Gb@3600 Zotac 3090 24Gb 551.33
madVR/LAV/jRiver/MyMovies/CMC
Denon X8500HA>HD Fury VRRoom>TCL 55C805K

Last edited by Manni; 17th October 2018 at 11:04.
Manni is offline   Reply With Quote
Old 16th October 2018, 23:51   #53289  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by mytbyte View Post
@Warner306:

Here are the graphs I promised for the monitor, can't do the TV at this time:

LG_IPS235_graph

I performed new measurements because white balance drifted a bit so peak brightness is actually 175 cd/m2 (and we complained about plasma brightness). I also used MadVR 0.92.17 and it seems math is a bit different than before, making my previous comments irrelevant. Color tweaks and highlight recovery are set to disabled/none, not to interfere with BT.2390 math.
Based on those graphs, I think the gamma response is perfect up to the limit of the display brightness (clipping at 175 nits).

I don't know how you will ever judge the performance of tone mapping, though, for brighter stimulus. The BT.2390 in HCFR wouldn't account for the gamut mapping in madVR. I don't know if there would be clipping or if the lack of anything but white would prevent out-of-gamut pixels. This might change the luminance of many pixels with a 10,000 nit stimulus down to 100 nits if any pixels were out-of-gamut. Have you tried tone mapping something like 500 nits down to 175 target nits in madVR?

Were you trying to find a target nits that perfectly tracked BT.2390 all the way to 100% output? The only commonality I see is that the brightness does seem to get to white too slow or too fast at the top. Maybe that is something that is lost in translation with SDR? And maybe it is close enough?

Quote:
Originally Posted by mytbyte View Post
perhaps the measured black level value should also be added to MadVR's math?
Don't know if that would help or not. The display is supposed to account for this.

Quote:
Originally Posted by mytbyte View Post
As for your patterns, I can see very faint flashing at the right side of the screen with "Black Level 10-bit" as well as slightly lighter right side compared to left side with "Black level 8-bit/10-bit combined".
That part don't make no sense. Only the 8-bit pattern should display correctly, but you shouldn't fail the black clipping test if you are clipping correctly to 175 nits. The gradient should go from right-to-left until Bar 16.

Last edited by Warner306; 17th October 2018 at 00:14.
Warner306 is offline   Reply With Quote
Old 16th October 2018, 23:58   #53290  |  Link
Warner306
Registered User
 
Join Date: Dec 2014
Posts: 1,127
Quote:
Originally Posted by madshi View Post
Quick question to 1050 Ti users: Is the 1050 Ti fast enough to do 4Kp60 HDR tone mapping with all the bells and whistles (measurements + highlight recovery + "compromise" disabled)?
At 1080p, a GTX 1050 Ti can't do tone mapping for 60 fps content with highlight recovery enabled. highlight recovery is the killer, as it pushes rendering times way over 16ms. This is with scale chroma separately enabled.

Last edited by Warner306; 17th October 2018 at 00:01.
Warner306 is offline   Reply With Quote
Old 17th October 2018, 00:30   #53291  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
@Manni

can you please check power state before you come to a conclusion rendertime are not that reliable.
huhn is offline   Reply With Quote
Old 17th October 2018, 00:36   #53292  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
Quote:
Originally Posted by 422415 View Post
Can anyone tell me what I'm doing wrong? I cannot get RGB Full 12bpc set in NVIDIA settings for 3840x2160 23hz. I have tried setting it to YCbCr422 then changing it after that but it always resets back to YCbCr422 limited. I am on Windows 1803 and have tried multiple NVIDIA drivers to no avail.
I can set RGB Full 12 bpc in 1809 and 416.34. I simply change the refresh rate first, setting it to RGB Full 8 bpc 340x2160p23, and then after I am in 23 Hz I can set 12 bpc, this lives through reboots and similar without issue. However, this is not with a custom 23 Hz mode, with a custom mode I simply cannot use >8 bit at all.
__________________
madVR options explained
Asmodian is offline   Reply With Quote
Old 17th October 2018, 00:49   #53293  |  Link
Manni
Registered User
 
Join Date: Jul 2014
Posts: 942
Quote:
Originally Posted by huhn View Post
@Manni

can you please check power state before you come to a conclusion rendertime are not that reliable.
I’m not sure I understand your request, but if you mean in the nVidia control panel I always set the power mode to adaptive.
__________________
Win11 Pro x64 b23H2
Ryzen 5950X@4.5Ghz 32Gb@3600 Zotac 3090 24Gb 551.33
madVR/LAV/jRiver/MyMovies/CMC
Denon X8500HA>HD Fury VRRoom>TCL 55C805K
Manni is offline   Reply With Quote
Old 17th October 2018, 00:56   #53294  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
no im' talking about the mhz the GPU runs at.
for example when my GPU has an easy task like native resolution the rendertime can look pretty high because the GPU is clocking down. while with 1080p source at UHD my rendertimes can look comparable low even through this task is much higher harder.

copyback can't cost you 9 ms that would mean the copy is so slow you can't use it for 120 FPS content because the copyback is using 100 % of the GPU. according to your sig you are running a 1080 ti so just no :-) how do i even play anything on my 1060 if a 1080 ti nearly dies to copyback.
huhn is offline   Reply With Quote
Old 17th October 2018, 01:12   #53295  |  Link
Manni
Registered User
 
Join Date: Jul 2014
Posts: 942
Quote:
Originally Posted by huhn View Post
no im' talking about the mhz the GPU runs at.
for example when my GPU has an easy task like native resolution the rendertime can look pretty high because the GPU is clocking down. while with 1080p source at UHD my rendertimes can look comparable low even through this task is much higher harder.

copyback can't cost you 9 ms that would mean the copy is so slow you can't use it for 120 FPS content because the copyback is using 100 % of the GPU. according to your sig you are running a 1080 ti so just no :-) how do i even play anything on my 1060 if a 1080 ti nearly dies to copyback.
I don’t understand a word of what you are saying. 22ms isn’t “dying”, it’s about half of what it has to do at 23p, and that’s with a lot of processing (HDR tonemapping with everything enabled). Copyback costs 6ms in D3D11 and 9ms in DXVA2, at least in that mode, and it does push GPU and CPU to 100%, even at native res, but again that’s with very hungry pixel shader processsing. I’m only reporting what I’m seeing. And yes, I have a 1080ti, otherwise I wouldn’t put it in my sig.
__________________
Win11 Pro x64 b23H2
Ryzen 5950X@4.5Ghz 32Gb@3600 Zotac 3090 24Gb 551.33
madVR/LAV/jRiver/MyMovies/CMC
Denon X8500HA>HD Fury VRRoom>TCL 55C805K

Last edited by Manni; 17th October 2018 at 01:14.
Manni is offline   Reply With Quote
Old 17th October 2018, 01:15   #53296  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
you are sure your CPU is pushed to 100 %?
huhn is offline   Reply With Quote
Old 17th October 2018, 01:18   #53297  |  Link
Manni
Registered User
 
Join Date: Jul 2014
Posts: 942
Quote:
Originally Posted by huhn View Post
you are sure your CPU is pushed to 100 %?
According to the task manager performance tab, yes. At 4.18ghz (overclocked).
__________________
Win11 Pro x64 b23H2
Ryzen 5950X@4.5Ghz 32Gb@3600 Zotac 3090 24Gb 551.33
madVR/LAV/jRiver/MyMovies/CMC
Denon X8500HA>HD Fury VRRoom>TCL 55C805K
Manni is offline   Reply With Quote
Old 17th October 2018, 01:27   #53298  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
that's more than odd. there is nothing that should be able to push your CPU to full load when hardware decoding is used.
can you run GPU-Z go to "graphics Card" and press on the question mark to start a test. after this the PCIe information can change and that number is interesting.

the taskmanager GPU load is unreliable just an extreme example for you: https://abload.de/img/taskmanageriei29.png
huhn is offline   Reply With Quote
Old 17th October 2018, 01:55   #53299  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
Quote:
Originally Posted by madshi View Post
When talking about bells and whistles, I meant in terms of HDR tone mapping. I don't consider NGU Chroma upscaling to be crucial. So, using e.g. D3D11 native decoding, with default chroma upscaling (Bicubic), and no other fancy options activated, can the 1050 Ti do 4Kp60 HDR tone mapping with measurement + highlight recovery?
Yes!

I tested with DXVA2 cb and both with and without the trade quality for performance option "compromise on tone & gamut mapping accuracy". This is on a 1050 Ti, Windows 10 1803, Nvidia driver 416.34 (adaptive), and madVR v0.92.17 (reset to defaults, D3D11 fullscreen windowed 10 bit, tone map HDR using pixel shaders with SDR output), MPC-HC 1.8.3, LAV Video 0.73.1. Using the "Samsung Travel With my Pet HDR UHD.ts" demo, 10 bit HDR HEVC 3840x2160p60.

With compromise: 6.8 ms rendering, <0.1 ms present (no dropped frames)
With compromise and measure each frame's peak luminance: 8.8 ms (no dropped frames)
With no compromise: ~13 ms, with spikes up to ~17 ms (a few dropped frames)
With compromise and 3DLUT: ~9.5 ms, with spikes up to ~12 ms (no dropped frames)
With no compromise and 3DLUT: ~18 ms, with spikes up to ~23 ms (a lot of dropped frames)

Without compromising quality performance is very dependent on content. A 3DLUT also takes significantly more power. No compromise looks a lot better too, though as a way to get a 1050 Ti to do HDR 60 fps processing while using copy back the compromise is great. Most of my HDR is 24 fps anyway.

Edit: I redid the tests with D3D11 native decoding. Performance is even better, no compromise runs perfectly as long as I don't use a 3DLUT at the same time.

With compromise: 5.2 ms rendering (no dropped frames)
With no compromise: ~10ms, with spikes up to ~12 ms (no dropped frames)!
With compromise and 3DLUT: ~7.5 ms, with spikes up to ~10 ms (no dropped frames)
With no compromise and 3DLUT: ~12.5 ms, with spikes up to ~16 ms (a few dropped frames).

Even more tests, D3D11 cb:
With compromise: 6.8 ms
With compromise and measure each frame's peak luminance: 8.8 ms (no dropped frames)
With no compromise: ~13 ms, with spikes up to 17 ms (a few dropped frames)

Edit2: OK, testing the actual request this time (I hope).
D3D11 Native but with DXVA chroma upscaling disabled in trade quality for performance options:
No compromise: ~10ms, with spikes up to ~12 ms (no dropped frames)!
__________________
madVR options explained

Last edited by Asmodian; 17th October 2018 at 02:55.
Asmodian is offline   Reply With Quote
Old 17th October 2018, 01:57   #53300  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by Manni View Post
I don’t understand a word of what you are saying. 22ms isn’t “dying”
He's saying clock speed is dynamic and ideally you need to log gpu speed and load for a good minute or so and calculate the average from that. Simpler tasks can show higher render times since the GPU can clock a lot lower.

Madshi, any chance we could get an average rendering statistic in the OSD? Ideally being able to see clock speeds on the OSD would be useful too.

D3D11 is quite a bit more efficient on my 1060.

Last edited by ryrynz; 17th October 2018 at 07:04.
ryrynz 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 13:03.


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