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 28th January 2014, 04:47   #22101  |  Link
ShadowVlican
Registered User
 
ShadowVlican's Avatar
 
Join Date: Mar 2007
Location: Toronto, CANADA
Posts: 41
Quote:
Originally Posted by mrcorbo View Post
Enabling any of the OpenCL options (except process NV12 surfaces) causes my GPU usage to spike to 99% with full clocks on my HD 6970 (win8.1/Cat 13.12) even when everything else is turned off/down as low as it will go.

Is anyone else seeing this issue with a 69XX Radeon? Up until now my GPU has handled any kind of processing that madVR could throw at it without breaking a sweat.
i've got a 6950 and prefer not to use OpenCL error diffusion either... the performance hit is massive

NNEDI3 doubling? not a chance. even 16 neurons makes everything unplayable.

i'll just stick with Jinc3 for everything for now

computer specs:
ShadowVlican is offline   Reply With Quote
Old 28th January 2014, 05:30   #22102  |  Link
MistahBonzai
Registered User
 
Join Date: Mar 2013
Posts: 101
Quote:
Originally Posted by cyberbeing View Post
madshi, could you explain how 'chroma upscaling = NNEDI3' and 'image doubling, double Chroma resolution' relate when upscaling a video exactly 2x?

640x360 -> 1280x720

GPU load 37%
chroma upscaling = NNEDI3
image doubling luma = NNEDI3
image doubling chroma = NNEDI3
opencl dither = enabled

.
.
.

[Edit]
Same thing happens with GPU load changing when 'image doubling luma = NNEDI3' is active for 640x360->1280x720 2x scale, depending on the setting for 'image upscaling'.

I was under the impression that 'image doubling luma' & 'image doubling chroma' would override 'image upscaling' & 'chroma upscaling' when the video is resized to a 2x multiple.

Am I misunderstanding something?
I have to be overlooking something here..or my rig (PCIE 3 2GB HD 7850 (latest Catalyst 13.2 driver pkg), i73770@3.4GH, Windows 7 64bit on a SSD) is a lot weaker that I had imagined.

When I set-up with those values I'm seeing a steady 29% GPU load. No image doubling (1280x720) is occurring and I see similar values to yours. If I select alt+3 (200% up-scale) while playing in MPC-HC I get 79% GPU load. If I scale the video via ffdshow with "Lanczos4Resize(1280,720)" in a simple script I still see GPU load similar to yours. That's because neither of the scaling triggers for Luma or Chroma have kicked in.

Bottom-line is I open the native video (640x360) with the values set as in your test (29% GPU load for me) and then double the video window size via alt+3 (1280*720) and my GPU load jumps to 79%. Otherwise the modifications to the image quality aren't active...

Last edited by MistahBonzai; 28th January 2014 at 05:35.
MistahBonzai is offline   Reply With Quote
Old 28th January 2014, 06:37   #22103  |  Link
pie1394
Registered User
 
Join Date: May 2009
Posts: 212
Quote:
Originally Posted by mandarinka View Post
In case anybody was wondering - I just tried the openCL dithering on the Llano APU (A8-3850, so the model with best IGP) with 1600MHz/CL11 DDR3 RAM (dual channel).

... skip ...

Using nnedi3 luma/chroma doubling for a 704x480 -> 1920x1080 upscaling but without OpenCL dithering raises rendering time "just" to 156 ms or so, BTW.
(Note: othere settings: Jinc3AR luma, lanczos3AR chroma)
- The rendering time is about 18ms on my system with the similar + all Jinc3AR setting for 720x480i60 16:9 DVD MV contents.
- It is dropped to 15ms with Bicubic75AR setting... (suggested by madshi. It is just enough when vector-adaptive deinterlacing is also counted for 60fps playback)
- It only takes 6.2ms with pure Jinc3AR setting.

Anyway this shows how powerful HD7970 is if compared to IGP/APU.

ps: My card is made by Lantic. Bought it in March, 2013 and price was about US$320 --- still quite expensive as a gaming / madVR usage card.

--
Core i5-3570K + Z77 + dual-ch DDR3-2400 + HD7970@925MHz Catalyst 13.12 (forced VSync ON, Vector-adaptive deinterlace mode) + Win7x64SP1 + MPC-BE 1.3.0.3 + LavFilter 0.60.1 (DXVA2) + madVR FSE (deband,angle-detect, all Jinc3AR for up-scale)

Last edited by pie1394; 28th January 2014 at 06:39.
pie1394 is offline   Reply With Quote
Old 28th January 2014, 06:44   #22104  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
Quote:
Originally Posted by madshi View Post
That really is odd. I rather think it's random. Try again with v0.87.4. And if the problem still occurs, please double check with v0.86.11 to make sure it's a new bug and not an old one. I'm only interested right now in bugs which were introduced in v0.87.x.
v0.87.4 fixed my queue size issues, I can use any queue sizes with any resize options. Well at least every combination I tried worked; I didn't do full factorial testing. I cannot reproduce that black screen (I also down reved my Nvidia drivers for obvious reasons, I assume that wasn't it).

With all NNEDI3 2x and 4x options at 64 neurons, Jinc3AR for chroma+luma upscaling, debanding low, smooth motion off, and no trade quality for performance options checked, my Titan (@1071MHz) hits 96% watching 352x240 29.97p at 1920x1440. I am also using a 3DLUT and gamma processing, queues at 32/24/8, FSW.

I will have to play around a lot to find the appropriate trade offs. I didn't expect that for a while with a Titan; until I saw all these wonderful NNEDI3 options.

Great work as always, amazing.

edit: I think I like Jinc3 chroma + NNEDI3 32 neuron doubling (chroma+luma) for watching 720p at 1440p, very sharp. With smooth motion on I am at 92% GPU usage and the GPU reports it is pulling 345 watts.

Last edited by Asmodian; 28th January 2014 at 07:51.
Asmodian is offline   Reply With Quote
Old 28th January 2014, 07:53   #22105  |  Link
antonyfrn
Registered User
 
Join Date: Nov 2012
Posts: 17
Quote:
Originally Posted by madshi View Post
Thanks guys. More details would be:

(1) Can be reproduced by unchecking the "use random dithering instead of OpenCL error diffusion" option in the "rendering -> trade quality for performance" section of the madVR settings.

(2) Works fine with 327.23 and older drivers. Problem occurs with anything newer than 327.23.

(3) Technical cause is that D3D9 <-> OpenCL interop doesn't seem to work properly. All APIs return success, but the OpenCL kernels seem to produce zero/black output at all times.
Well some good news ManuelG responded about this in the new driver thread and will take a look into it.
__________________
Windows 10 Pro 64Bit
i7 6700k + H110i GT @4.6Ghz
16GB Corsar Vengance DDR4
EVGA GTX 970 FTW ACX 2.0
ASUS Maximus VIII Hero
ASUS Xonar D2X
antonyfrn is offline   Reply With Quote
Old 28th January 2014, 09:09   #22106  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Humm it's hard to deny that a used HD7950 would really hit the spot in order to permanently force 1.2/2.4x 64neurons NNEDI

And to my eyes, two things improve the subjective pop effect on my 3500:1 Sammy TV: BFI, as there isn't motion blur smearing everything anymore and Error Diffusion as the dithering noise is not a just a veil of random noise thrown onto the picture(much like what a VGA connection looks like due to the increased noise level). I see the same sort of "pop effect" improvement with mVR's error diffusion as I was getting with SmoothL, too good to be true

My only problem right now is that using PotP's GUI in 2D mode flashes black frames during the FSE/FSW transitions(that wasn't the case on XP) but PotP's D3D GUI would appear to completely alleviate this problem, so no biggy.

Last edited by leeperry; 28th January 2014 at 09:18.
leeperry is offline   Reply With Quote
Old 28th January 2014, 09:34   #22107  |  Link
6233638
Registered User
 
Join Date: Apr 2009
Posts: 1,019
I may have to edit my 3DLUT now, as values near black seem to be darker than before when using error diffusion.

http://abload.de/img/error-diffusion-2uks1c.png
[urlhttp://abload.de/img/random-dither-2ygsae.png[/url]

Brightened:
http://abload.de/img/error-diffusion-3plsex.png
http://abload.de/img/random-dither-31wssj.png

As you should be able to see though, with error diffusion enabled each level is much more distinct from the last. When brightened like this, 20-25 all look roughly the same with random dither.



Hopefully editing my 3DLUT will not affect the results I'm seeing on a full black screen:


(brightened significantly)

With error diffusion enabled, black is actually solid black, which means that my display can turn off the local dimming zones rather than keeping them on at a low level.
This is a significant improvement and absolutely worth the performance hit.

Of course, if you don't have that problem, it may not matter, and I really hope that changing my 3DLUT values does not affect this result.

I agree with cyberbeing that error diffusion does have a tendency to create patterns though, which random dither avoids.
But the noise level is so much lower near black, and noise is monochrome with error diffusion, rather than being multicolored with random dither.

Last edited by 6233638; 28th January 2014 at 15:26. Reason: New results http://forum.doom9.org/showthread.php?p=1664865#post1664865
6233638 is offline   Reply With Quote
Old 28th January 2014, 10:12   #22108  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Perhaps we could have some sort of anti pattern dithering whilst still allowing for this level of black performance?

20-25 are looking fantastic there! That's an awesome result which should have some significant impact in your late night viewing.
ryrynz is offline   Reply With Quote
Old 28th January 2014, 10:17   #22109  |  Link
6233638
Registered User
 
Join Date: Apr 2009
Posts: 1,019
Quote:
Originally Posted by ryrynz View Post
Perhaps we could have some sort of anti pattern dithering whilst still allowing for this level of black performance?
I wonder if it would be possible to do something like render dithering at 2x and downsample it? (probably not?)

Quote:
Originally Posted by ryrynz View Post
20-25 are looking fantastic there! That's an awesome result which should have some significant impact in your late night viewing.
Yes, it's a big improvement. I actually suspect that the measured points in my 3DLUT will be the same, or roughly the same as before, it's just that each level between them is much better defined now.
6233638 is offline   Reply With Quote
Old 28th January 2014, 10:47   #22110  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by Farfie View Post
Everything is working for me, but image doubling 1280x720 to 2560x1440 is absolutely killer. With error diffusion on I can forget using anything above 32 neurons.

Opinion section:
Error diffusion vs going from 32 to 64 neurons?
FWIW, depending on the video, even 16 neurons can look almost identical to 64 neurons. And 16 neurons only consume a quarter of the preformance of 64 neurons. That said, I recommend not to go lower than 32 neurons for luma upscaling, because 16 neurons leaves too many artifacts in the image for my taste with *some* videos. Every higher neuron setting costs almost twice the performance, and only brings a minor improvement (but an improvement nonetheless).

Quote:
Originally Posted by mimi123 View Post
My primary play content is Bluray, then 1920*1080 and a frame rate of 23.976Hz.
My display is a JVC X3 projector and 110" screen.

At this moment I have not any discrete card, I use the intel hd3000.

I want to remove banding as much as possible and add a bit of sharpness.
I want a card that can do all this flawlessly and without artifacts and if someday Madvr can send more than 8 bits I want a compatible card.
And finally the card must to be as silent as possible.
That sounds like you don't need any of the more "expensive" algorithms. So the limiting factor might be the debanding. I guess a budget GPU should do the trick for you. E.g. look at gaming benchmarks and choose the cheapest GPU which has GDDR5 and which has a good performance per price ratio. Of course choosing a more powerful GPU never hurts madVR playback. But that's your decision, really. If you choose an AMD GPU, I would definitely choose a GCN model (7xxx), though, not an older model.

If you ask for a specific model, well I don't know your exact budget. Given your needs, I guess a 7570 or 7670 should probably be fast enough. A 7750 or 7770 would be awesome, probably too fast for your needs, but with some performance headroom left for future algorithms. Of course you can also go with NVidia. If you need a silent GPU, I'd suggest that you read some reviews, so you pick a good/silent model from a specific manufacturer. In any case, pick one with GDDR5, and make sure the price/performance ratio is good.

Quote:
Originally Posted by cyberbeing View Post
I know you've said the NNEDI3 shifts pixels to the left, but looking at some screenshots it appears the image is shifted both up & left (i.e. diagonally)? Is that expected?
Quote:
Originally Posted by flashmozzg View Post
Maybe it's because NNEDI scales only one side (X or Y), so when it scales X it shifts left, and when Y - up.
^ flashmozzg is correct. Although it gets more interesting when you quadruple. In that case the image shifts up/left when doing the first (doubling) pass. Then when doing the 2nd (quadrupling) pass, madVR uses a modified NNEDI3 pass which moves the pixels right/bottom instead of left/up. But the move back right/bottom is smaller in size than the move left/up. So basically both doubling and quadrupling move the image left/up, but doubling moves it more than quadrupling.

Quote:
Originally Posted by Werewolfy View Post
I tested it with a 750*576 video on a 1080p TV screen and it does seem that when I use NNEDI3, Jinc upscales in one direction and Catmull-Rom downscales in the other direction simultaneously because rendering times go up when I select Jinc 8 for upscaling for example and go up too when I select Lanczos 8 for downscaling.
I suppose you're only doubling the luma channel and not the chroma channels? In that case Jinc might be used to upscale the chroma channels directly from 720x576 to 1920x1080. Luma can't be upscaled by Jinc for Y because it would be a downscale. So luma X is upscaled by using Lanczos (using the same tap number you specified for Jinc), and luma Y is downscaled using Catmull-Rom. Lanczos always replaces Jinc if Jinc is selected but can't be used.

Quote:
Originally Posted by cyberbeing View Post
And to make sure I'm no longer confused, what is the result of the following?

640x360 4:2:0 -> 1920x1080
Chroma Upscaling = NNEDI3
NNEDI3 double Luma = Enabled
NNEDI3 double Chroma = Enabled
NNEDI3 quadruple Luma = Enabled
NNEDI3 quadruple Chroma = Disabled
  • Conversion from 4:2:0 YcbCr (320x180) to 4:4:4 RGB (640x360) with 'chroma upscaling' setting NNEDI3
  • Conversion from 640x360 4:4:4 RGB -> 640x360 4:4:4 YCbCr.
  • Y & CbCr channels doubled to 1280x720 4:4:4 YCbCr with NNEDI3
  • Y channel only doubled to 2560x1440

Is it then:
  • CbCr channels upscaled 1280x720->1920x1080 with 'image upscaling' setting
  • Y channel downscaled 2560x1440->1920x1080 with 'image downscaling' setting
  • Conversion from 4:4:4 YCbCr to RGB

Or is it:
  • CbCr channels upscaled 1280x720->2560x1440 with 'image upscaling' setting
  • Conversion from 4:4:4 YCbCr to RGB
  • 2560x1440 RGB -> 1920x1080 RGB with 'image downscaling' setting
It is the first ("CbCr channels upscaled 1280x720->1920x1080 with 'image upscaling' setting").

Quote:
Originally Posted by dansrfe View Post
It might be nice to have the ability to import/export profiles so that we can potentially share and cross-check them.
You can already import/export the whole settings - but not profiles, only. I'm not sure if exporting/importing is necessary. Anyway, even if you guys want it, it's probably not coming soon.

Quote:
Originally Posted by mandarinka View Post
I guess using the same physical memory for CPU/GPU just doesn't help at all with the DX9/OpenCL interoperation overhead.
Yes, it seems so, thanks for testing. I'm not sure exactly where the overhead comes from. But everything should be happening on the GPU, so it makes sense that unified memory doesn't help.

Quote:
Originally Posted by turbojet View Post
A few issues on 87.4 that I don't think have been reported:

1. With nvidia 250 and 650 and error diffusion on 87.4 the screen goes blank when paused and takes a few seconds to come back. Is this intentional or a bug?
Can anybody reproduce this? Doesn't seem to occur here.

Quote:
Originally Posted by turbojet View Post
2. On a 60mhz display when ivtcing with smoothmotion on there is a lot of good frames removed and blended frames left. Almost as if smoothmotion is running before ivtc.
No, IVTC is definitely done first. It must be because it's done on the CPU and I don't copyback, so I couldn't accidently swap the order of these two operations.

My best guess would be that your source video is not telecined film, but already contains those blended frames. Forced film mode only works correctly with properly telecined content. For field-blended or frame-blended sources you need to use video mode deinterlacing.

Quote:
Originally Posted by turbojet View Post
3. Probably known but playing video in film mode is not good.
It's very bad, yes that's known and expected.

Quote:
Originally Posted by kasper93 View Post
Apparently AMD made great improvements in this matter. On my old HD5870 I have rendering time ~35ms with 720p23. Of course with disabled smooth motion and not upscaled. After upscaling to full hd I have ~93ms, smooth motion (to 60fps) adds ~74ms to the process So OCL performance on my gpu is dreadful, but I bet it's better than NV

I know this post doesn't help you in any way, but I just wanted to show you how performance is scaled when it comes to older generations
Yeah, it seems that the GCN generation (7xxx) helps a lot with OpenCL performance.

Quote:
Originally Posted by cyberbeing View Post
I did a bit of testing on this:

madVR 'opencl error diffusion dither':
  • Flat colors without noise, but sometimes creates a uniform checkerboard pattern.
  • Sharper hard edges.
  • Source banding is more visible.

madVR 'random dither':
  • Noisy flat colors, without any patterns.
  • Hard edges softened by noise.
  • Source banding is less visible.

To get similar visible smoothness on gradients containing banding, I need to use higher debanding settings with 'opencl error diffusion dither' than I do with 'random dither'.
Yes, I'm seeing the same thing. Although the checkerboard pattern should be with color values which are very near to each other, so it should be rather faint.

I agree that random dithering hides source banding better. Actually the original f3kdb debanding algorithm has a feature which does something very similar to madVR's random dithering, just to hide source banding. At some point I'm probably going to add an "add grain" or similar algorithm which you could then combine with error diffusion.

So would it be fair to say that error diffusion is more true to the source compared to random dithering?

Quote:
Originally Posted by mrcorbo View Post
Enabling any of the OpenCL options (except process NV12 surfaces) causes my GPU usage to spike to 99% with full clocks on my HD 6970 (win8.1/Cat 13.12) even when everything else is turned off/down as low as it will go.

Is anyone else seeing this issue with a 69XX Radeon? Up until now my GPU has handled any kind of processing that madVR could throw at it without breaking a sweat.
My impression is that AMD did a lot of changes in the GCN generation (7xxx) to improve compute performance - which is really what OpenCL is all about. So I suppose 5xxx and 6xxx generations are expected to underperform a bit with OpenCL.

Quote:
Originally Posted by ShadowVlican View Post
i've got a 6950 and prefer not to use OpenCL error diffusion either... the performance hit is massive

NNEDI3 doubling? not a chance. even 16 neurons makes everything unplayable.
Ouch, that hurts! Sounds like my 7770 performs better than your 6950 with this! Not really sure, though...

Quote:
Originally Posted by MistahBonzai View Post
I have to be overlooking something here..or my rig (PCIE 3 2GB HD 7850 (latest Catalyst 13.2 driver pkg), i73770@3.4GH, Windows 7 64bit on a SSD) is a lot weaker that I had imagined.

When I set-up with those values I'm seeing a steady 29% GPU load. No image doubling (1280x720) is occurring and I see similar values to yours. If I select alt+3 (200% up-scale) while playing in MPC-HC I get 79% GPU load. If I scale the video via ffdshow with "Lanczos4Resize(1280,720)" in a simple script I still see GPU load similar to yours. That's because neither of the scaling triggers for Luma or Chroma have kicked in.

Bottom-line is I open the native video (640x360) with the values set as in your test (29% GPU load for me) and then double the video window size via alt+3 (1280*720) and my GPU load jumps to 79%. Otherwise the modifications to the image quality aren't active...
I'd suggest to use NNEDI3 only for luma doubling, and only with 32 neurons, if your GPU can't handle more. I would not recommend using 16 neurons for luma doubling, though.

Quote:
Originally Posted by Asmodian View Post
v0.87.4 fixed my queue size issues, I can use any queue sizes with any resize options. Well at least every combination I tried worked; I didn't do full factorial testing. I cannot reproduce that black screen (I also down reved my Nvidia drivers for obvious reasons, I assume that wasn't it).
Great!

Quote:
Originally Posted by Asmodian View Post
With all NNEDI3 2x and 4x options at 64 neurons, Jinc3AR for chroma+luma upscaling, debanding low, smooth motion off, and no trade quality for performance options checked, my Titan (@1071MHz) hits 96% watching 352x240 29.97p at 1920x1440. I am also using a 3DLUT and gamma processing, queues at 32/24/8, FSW.
Ouch! I'd suggest to use NNEDI3 only for luma upscaling. IMHO using it for the chroma channels is somewhat wasted. For quadrupling you might get away with 32 or even 16 neurons. For doubling I wouldn't go lower than 32, though.

Quote:
Originally Posted by antonyfrn View Post
Well some good news ManuelG responded about this in the new driver thread and will take a look into it.
Great!! Please keep us posted, if there's any progress...

Quote:
Originally Posted by leeperry View Post
And to my eyes, two things improve the subjective pop effect on my 3500:1 Sammy TV: BFI, as there isn't motion blur smearing everything anymore and Error Diffusion as the dithering noise is not a just a veil of random noise thrown onto the picture(much like what a VGA connection looks like due to the increased noise level). I see the same sort of "pop effect" improvement with mVR's error diffusion as I was getting with SmoothL, too good to be true
Happy to hear that!

Which Sammy TV is that? Can it do 4:4:4? I might be looking into getting a new monitor for my development PC. I'm tempted with misusing a TV for that. But 40" does sound a bit large. 30" would be ideal...

Quote:
Originally Posted by 6233638 View Post
I may have to edit my 3DLUT now, as values near black seem to be darker than before when using error diffusion.
Hmmmm... That almost looks like a different gamma curve. But it's not, is it? Do you think error diffusion is more exact here? Or how to do explain the brightness differences? It looks a bit weird to me. Ok, random dithering has higher spikes up and down compared to error diffusion. And the spikes down vanish in BTB. So that might push subjective brightness up. Maybe that's the explanation?

Quote:
Originally Posted by ryrynz View Post
Perhaps we could have some sort of anti pattern dithering whilst still allowing for this level of black performance?
How much does the checkerboard pattern distract in real live viewing? With real movie content every frame should be (at least ever so slightly) different compared to the next, which should totally change the whole error diffusion pattern. It's different with test patterns where each frame might be 100% identical to the previous.

Quote:
Originally Posted by 6233638 View Post
I wonder if it would be possible to do something like render dithering at 2x and downsample it? (probably not?)
After downsampling you end up with more than 8bit. So you'd need another dither/error diffusion pass. Or if you downsample by rounding down to 8bit, the whole supersampling processing would probably have no benefit at all.

Quote:
Originally Posted by 6233638 View Post
Yes, it's a big improvement. I actually suspect that the measured points in my 3DLUT will be the same, or roughly the same as before, it's just that each level between them is much better defined now.
It is quite possible that the measurements also change a bit. So you might want to redo the calibration, just to be sure.

-------

After seeing all those NNEDI3 benchmarks I want to post a settings recommendation. I understand that you guys want to test how much your GPU can handle, so you dial up all the new settings as much as possible, and then you're a little bit let down by how slow it all gets. But most of you really do seem to go overboard with the settings IMHO. So here's my recommendation:

(1) In my tests I've found little reason to enable NNEDI3 for the chroma channels. So I would recommend to leave the right side of the image doubling settings page unchecked.

(2) The neuron number is more important for doubling than it is for quadrupling. I'd recommend 32-64 neurons for doubling and 16-32 for quadrupling. I wouldn't go lower than 16 for doubling, unless your GPU absolutely can't handle 32 neurons and you still prefer NNEDI3 with 16 neurons over Jinc/Lanczos.

(3) When using image doubling with NNEDI3, I'd recommend to use Lanczos3 AR for "image upscaling", and not Jinc3 AR.

Since NNEDI3 costs so much performance, we have to accept some compromises. Of course if you have an almighty GPU, you can dial up everything to your heart's content. But the above recommendation is what should IMHO give the best quality/performance ratio, when using NNEDI3.
madshi is offline   Reply With Quote
Old 28th January 2014, 11:19   #22111  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
still problems with hd4000 and deint split drops first, gpu load still at 84-95 % with 87.4. with a old test 86.11 was at ~82 %, now it is at 88-95% but working.
back buffer is set to 8 and fullscreen exclusive is disabled else it isn't working on 86.11 too rest is totally default.

Last edited by huhn; 28th January 2014 at 11:33. Reason: it's 87.4
huhn is offline   Reply With Quote
Old 28th January 2014, 11:26   #22112  |  Link
turbojet
Registered User
 
Join Date: May 2008
Posts: 1,840
Quote:
Originally Posted by turbojet View Post
A few issues on 87.4 that I don't think have been reported:

1. With nvidia 250 and 650 and error diffusion on 87.4 the screen goes blank when paused and takes a few seconds to come back. Is this intentional or a bug?
Can anybody reproduce this? Doesn't seem to occur here.
Seems to happen always in film mode not in video mode.

Quote:
Originally Posted by turbojet View Post
2. On a 60mhz display when ivtcing with smoothmotion on there is a lot of good frames removed and blended frames left. Almost as if smoothmotion is running before ivtc.
No, IVTC is definitely done first. It must be because it's done on the CPU and I don't copyback, so I couldn't accidently swap the order of these two operations.

My best guess would be that your source video is not telecined film, but already contains those blended frames. Forced film mode only works correctly with properly telecined content. For field-blended or frame-blended sources you need to use video mode deinterlacing.
For testing you can use http://www.mediafire.com/download/5l...ce_banding.mpg set display at 60hz, turn on film mode and smooth motion. It's soft telecined so must decode with lav video to find correct cadence. Maybe it's just my setup, setting a profile helps but it would be nice to use smooth motion and ivtc together.

if (srcFps=29) and (filmMode=true) "interlaced" else "progressive"
29.97 ivtcd in film mode = progressive
29.97 in video mode= interlaced

is this a bug?
__________________
PC: FX-8320 GTS250 HTPC: G1610 GTX650
PotPlayer/MPC-BE LAVFilters MadVR-Bicubic75AR/Lanczos4AR/Lanczos4AR LumaSharpen -Strength0.9-Pattern3-Clamp0.1-OffsetBias2.0

Last edited by turbojet; 28th January 2014 at 12:40.
turbojet is offline   Reply With Quote
Old 28th January 2014, 11:32   #22113  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Which Sammy TV is that? Can it do 4:4:4? I might be looking into getting a new monitor for my development PC. I'm tempted with misusing a TV for that. But 40" does sound a bit large. 30" would be ideal...
there are hybrid displays out there: german list: http://geizhals.at/de/?cat=monlcd19wide&xf=103_Tuner~99_26#xf_top

i highly recommend Philips tvs for pc usages they normally all support unlimited rgb in pc mode on >all< refresh rates.
huhn is offline   Reply With Quote
Old 28th January 2014, 11:34   #22114  |  Link
omarank
Registered User
 
Join Date: Nov 2011
Posts: 187
Profiling works nicely here, but keyboard shortcuts for profiles don't work for "processing" and "rendering" settings. The shortcuts work only for "scaling algorithms".

I am using a profiling script for smooth motion too, and I found that the profile which enables smooth motion, when enabled, doesn't consider the option selected in the smooth motion settings page and always enable smooth motion.

I was wondering how to use "command line to execute when this profile is activated/ deactivated". Can someone please show me an example of using this functionality?
omarank is offline   Reply With Quote
Old 28th January 2014, 11:41   #22115  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Just as a heads-up madshi:

Last night (at 4 oīclock in the evening/morning!) Blaire from 3dcenter already contacted ManuelG of Nvidia, who requested a step by step guidance of the problem. Since Iīve linked your last and more detailed post to him, he was actually able to reproduce the problem and gave all required information to ManuelG in step by step form.

They are looking into it right now. Only thing left now is to look for a new beta driver, where they fix the problem, but I guess Blaire will let me know soon enough. Going to keep you updated on this as soon as I get some feedback.
iSunrise is offline   Reply With Quote
Old 28th January 2014, 11:45   #22116  |  Link
Qaq
AV heretic
 
Join Date: Nov 2009
Posts: 422
MadVR 87.3. ATI 7750, driver 13.12, Win7 SP1 x86. OS and video driver are set to high perfomance.

NNEDI3 image doubling: dropouts ~30%
OpenCL error diffusion: dropouts ~30% (only works fine for small size videos without upscaling to full screen)
NNEDI chroma upsampling: dropouts ~30% (this one worked fine for me in 87.1b, btw)

Timings are fine, its just present queue goes to zero. Need more fine tuning of the new CCC 13.12 maybe.
Qaq is offline   Reply With Quote
Old 28th January 2014, 11:53   #22117  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Quote:
Originally Posted by 6233638 View Post
I may have to edit my 3DLUT now, as values near black seem to be darker than before when using error diffusion.
...
As you should be able to see though, with error diffusion enabled each level is much more distinct from the last. When brightened like this, 20-25 all look roughly the same with random dither.
That is actually very interesting. Never thought that error diffusion (havenīt checked it myself, yet, though) would improve something in such drastic form.

When I hardware calibrated my Eizo CG243W and checked that pattern after calibration in madVR, I also wondered why at some certain brightness levels (especially the lower ones) it seemed to miss a lot of differentiation, when outside of madVR, everything was perfectly fine. At that point I actually thought that I must have been doing something wrong with the madVR settings, but doing it again carefully didnīt change the outcome. Back then I reported my results, but I guess it was quite hard for others to understand what exactly I was seeing, since you need to have a screen that is able to perfectly differentiate levels in the first place.

I am really surprised by these results now. If that translates well to the actual picture, this could mean a lot more defined and natural apperance of the source and also a lot more details in darker (and brighter?) areas. And it would also mean that every good calibrated LCD/CRT would benefit a lot from it. It would actually make it even more worth it to calibrate your screen now than it already has been before.

Are we really sure that this is only the result of random dithering vs. error diffusion? And what are the reasons for this? Why should dithering change differentiation steps in the brightness levels? Yes, it is kinda expected, that less noise would lead to a lower noise floor, and as such, would lead to somewhat more perfect and darker blacks (and probably also whites) but why would it suddenly change the differentiation between the levels? Wouldnīt there have to be some gamma change involved to come to this result? I donīt understand this at all.

Could this be a result of your 3DLUT only, which needed to compensate for the shortcomings of your display? But why did I see the same thing and outside of madVR, everything is fine. Thatīs confusing me to no end right now. But I think I really need to test this to see it for myself.

Last edited by iSunrise; 28th January 2014 at 12:16.
iSunrise is offline   Reply With Quote
Old 28th January 2014, 12:11   #22118  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
So is there any way to avoid the black frames flashing during FSE/FSW transitions? This is happening whenever I want to open a context menu and that never occured on XP.

All this said, FSE was mandatory on XP but isn't it a bit of overkill on W7? The OS knows how to handle VSYNC for a change.

Quote:
Originally Posted by madshi View Post
depending on the video, even 16 neurons can look almost identical to 64 neurons. And 16 neurons only consume a quarter of the preformance of 64 neurons. That said, I recommend not to go lower than 32 neurons for luma upscaling, because 16 neurons leaves too many artifacts in the image for my taste with *some* videos. Every higher neuron setting costs almost twice the performance, and only brings a minor improvement (but an improvement nonetheless).
Does it make sense to use 32 neurons for 720p to 1080p? Or would J3AR be sufficient and NNEDI only necessary for SD to HD? I guess that calls for a number of tests but you might have run some on your own.

Quote:
Originally Posted by madshi View Post
Which Sammy TV is that? Can it do 4:4:4? I might be looking into getting a new monitor for my development PC. I'm tempted with misusing a TV for that. But 40" does sound a bit large. 30" would be ideal...
Yep, personally I find 27" too small and 32" a tad too big, this 30.5" OLED would be perfect: CES 2014: Chinese Brands Get In On The OLED Action

It's the Sammy UE32F5000, it does support 4:4:4 but then no more BFI so it becomes blurryland and there's the usual panel and backlight homogeneity lotteries....luckily you can check the panel maker in the factory menu, there's a thread on that matter on AVS if you can find the patience of trying your luck: Samsung Panel Version Thread

Many ppl claim that the Sammy panels are better but not a single one shows any measurement, all I know is that the AUO panel provides 0.4 cd/mē black, acceptable blurring once BFI is enabled and 3500:1 native CR.

The best choice for 4:4:4 BFI and 5K:1 native CR is that new UVēA monitor from EIZO but it costs 500€ and it's 24" only. This Sammy TV will fit the bill until the chinese flood the market with OLED

Quote:
Originally Posted by huhn View Post
i highly recommend Philips tvs for pc usage
AFAIK these don't provide any kind of R/G/B gain/offset settings so you'll have to use a CLUT on the PC in order to reach D65.

Last edited by leeperry; 28th January 2014 at 12:13.
leeperry is offline   Reply With Quote
Old 28th January 2014, 12:13   #22119  |  Link
DragonQ
Registered User
 
Join Date: Mar 2007
Posts: 934
Who can actually use error diffusion though? It seems as if it's about as demanding as NNEDI3 from all the reports so far.
__________________
TV Setup: LG OLED55B7V; Onkyo TX-NR515; ODroid N2+; CoreElec 9.2.7
DragonQ is offline   Reply With Quote
Old 28th January 2014, 12:24   #22120  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by DragonQ View Post
Who can actually use error diffusion though? It seems as if it's about as demanding as NNEDI3 from all the reports so far.
This is why I inquired about a CPU version.. unfortunately no answer from Madshi though :P So many goodies.. so much GPU..

Hey Madshi, was wondering how a 2x upscale with Jinc3ar + 2x upscale with nnedi then a downscale with Lanczos3 would look in MadVR. Basically getting that nnedi sharpness and antialiasing without too much thinning happening.. could be a nice blend of both. Do you think it's worth having a preupscaling option? I've played a round a bit in Avisynth with pre Jinc and Lanczos upscaling and from what I see things look quite nice when set up before nnedi. Thoughts?
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 03:31.


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