madVR - high quality video renderer (GPU assisted)

[leeperry]

Sadly I don't think finer grained neuron settings are possible. From the NNEDI3 docs
Oh you nail it!

BTW using monostatic ED1 LL, low debanding, CC AR LL downscaling, J3AR upscaling for chroma & luma, with my 1Ghz 7850 I used to be capped to:
-64x NNEDI luma for <=24p >=1.85 720p@1080p
-32x NNEDI luma for >=25p 720p@1080p

Now with the new bulds, I can do 64x NNEDI luma for <=25p 720p@1080p but that's still a no-go for 29.97fps.

[pie1394]

Tested with Core i5-3570K + Z77 + DDR3-2400 8GB + HD7970 (PCIe Gen3 x16, Catalyst 13.12) on Win7x64SP1. Both test build#1 and build#2 allow more stressful NNEDI3 settings if compared to 0.87.9.

Yet test build2 has obviously less impact on concurrent GPU deinterlacing job's performance than test build1.

Here is the test build's top playback settings with Debanding+Angle_detection, No_Dithering options on HD7970:

[720x480i60 to 1920x1080p60]
Chroma scaling: NNEDI3_32 (was NNEDI3_16)
Image Doubling: Luma with NNEDI3_32, Chroma with NNEDI3_32 (Chroma was NNEDI3_16)
Image Up-scaling: Jinc3_AR (was Bi-cubic 100 AR)

[1280x720p24 to 1920x1080p24]
Chroma scaling: NNEDI3_32 (was NNEDI3_16 with Luma NNEDI3 doubling)
Image Doubling: Luma with NNEDI3_64 (was NNEDI3_32 at max)
Image Up-scaling: Jinc3_AR
Image Down-scaling: Spline3_AR

[1440x1080i60 to 1920x1080p60]
Chroma scaling: NNEDI3_32 (was NNEDI3_16)
Image Doubling: Luma with NNEDI3_32 (was impossible)
Image Up-scaling: Jinc3_AR
Image Down-scaling: Spline3_AR

[1920x1080p24 P10 to 1920x1080p24]
Chroma scaling: NNEDI3_128 (was NNEDI3_64)

[1920x1080i60 to 1920x1080p60]
Chroma scaling: NNEDI3_32 (was NNEDI3_16)

[Anime Viewer]

Yes, but only with SLI on. SLI is much better with the test builds though still not as fast as without SLI.


Its weird that you would see a difference SLI or not Nvidia systems shouldn't have the interop issue. Given that you saw apparent improvement with the test builds I tried them on an Optimus system using the Nvidia gpu for mpc, and NNEDI3. It made no difference on the Optimus system. Nvidia users don't need to waste time trying this test build.

[sexus]

well who knows we may see some improvements coming our way for us nvidia users , sure would like to get more optimization for my titan over here , get some stuttery playback on some movies here and there even on 720p content sometimes , am using NNEDI64 for chroma upscaling, which seems to

be pretty hardcore on my titan with image doubling but with jinc3 for chroma upscaling i dont get any occasional stuttering , odd, my image doubling settings are NNEDI64 for both luma doubling at 1.5 scaling factor and quadrupling at 3.0 scaling factor and 16 NEEDI on both chroma doubling at 1.5 scaling and quadruppling at 3.0 scaling factor

error diffusion is set to option 1 and smooth motion set to only if motion judder without it, debanding is enabled medium to high , the trade quality for performance only has one option checked and that is dont use linear light for dithering which was preset no idea why

[jaju123]

Hey Madshi, your reply about the AMD interop problem has been forwarded to the AMD driver team. Hopefully in the future we will see some improvements relating to this issue and maybe a complete resolution.
I'll post any updates in this thread as they come :)

[seiyafan]

Seeing that 32 neurons is the default, is 16 neurons still better than none? or does it bring other issues? My card is not fast enough to do 32.

[cca]

For me the difference of the new test builds can be summarized as follows: with build 1 I can do NEEDI3 at 32 neurons scaling 1280x720p@29.97fps to 1080p. With build 2 I cannot, it just drops frames. It is a huge improvement for my aging Radeon 5850.

[Procrastinating]

Seeing that 32 neurons is the default, is 16 neurons still better than none? or does it bring other issues? My card is not fast enough to do 32.

NNEDI3 produces an entirely different set of artifacts compared to traditional upscalers, and increasing neurons decreases these artifacts. Having 16 neurons is generally considered to be comparable to Jinc 3 in terms of having "equivalent degrees" of distortion on an image, though these are often also considered "nicer" artifacts, particularly with anime. 64 neurons is where you really see the wonders of NNEDI3 though, rather than just looking at a "nicer" distorted image.

[Farfie]

Seeing that 32 neurons is the default, is 16 neurons still better than none? or does it bring other issues? My card is not fast enough to do 32.

For anime I'd say using NNEDI3 at any amount of neurons is worth it. For live content, I still like what it does, but it's more in the open. My opinions.

[n3w813]

@leeperry
Have you had a chance to test your Nvidia GTX 750ti OC yet? I'm thinking about replacing my current vid card with that one. :)

[leeperry]

Not yet, madshi caught me by surprise with yesterday's faster builds and I'm still kinda reluctant to buying a graphic card with a 128bit memory bus...Lucky me, my seller rarely checks his emails and it would appear that the white dots I was getting were due to a malfunctioning flat HDMI cable(thrown as a freebie with a TV), for some supernatural reason it would act up when the GPU reached 70°C and dots would disappear a few mins later...and that never happened on the windows desktop :confused:

hu1kamania was kind enough to post results using his o/c 750Ti (http://forum.doom9.org/showpost.php?p=1677596&postcount=25951), it's essentially on par with a 7850 without the interop lag lottery and saving ±70W in the process :)

[Asmodian]

Its weird that you would see a difference SLI or not Nvidia systems shouldn't have the interop issue. Given that you saw apparent improvement with the test builds I tried them on an Optimus system using the Nvidia gpu for mpc, and NNEDI3. It made no difference on the Optimus system. Nvidia users don't need to waste time trying this test build.

Given the large increase in PCI-E activity with SLI on I think it is safe to assume something like AMD's interop issue is going on with SLI enabled. Especially since an attempt to improve AMD's issue improved SLI performance so dramatically.

Optimus is not similar to SLI in any way.

I agree there is no need for single Nvidia GPU users to test these builds but I would love to see someone else test SLI performance. (And CF users, are there no CF users anymore?)

I noticed a lot of Optimus specific options when editing profiles with Nvidia Inspector, has anyone played with them to see if madVR behavior can be improved on an Optimus system?

[peplegal]

Hi guys...

...for some esoteric reason, when I overwrite those files of the last test version (AMD Interop) over my default instalation, MPC-HC stops loading MadVR (start using LAV render instead).

I rename "madVRinteropTest2.ax" to "madVR.ax" and put the ".ax", the ".dll" and the ".exe" files over previous files.

What am I doing wrong ?

Any advice ?

Thanks in advance.

[Stereodude]

Can you please try to find out which exact madVR build introduced this problem? Also a debug log might help figuring out why the switch fails. Please don't switch back and forth in the debug log. Just let it fail, then stop, I don't need to see it working in the log, I only need to see the fail. Please enable the OSD (Ctrl+J) while creating the debug log, because otherwise the log will not contain all important information.
FWIW, the MPC-HC 1.7.4 update seems to fix the issue. Not sure why some versions of MadVR behaved differently with the older versions of MPC-HC though.

Do you still want the debug log?

[Procrastinating]

MPC-HC stops loading MadVR (start using LAV render instead).


Uninstall everything, do a registry cleanup with CCleaner or whatever, reinstall everything, give your user account full control of the madVR folder and files. The latter part is probably the only real problem here, but the former is always nice to do once in a while.

[Osjur]

Hi all.

I have been testing build 1 with overclocked 290X and I think these are the best settings so far for me without dropped frames:

1920x1080 10bit @ 23.97fps to 2560x1600
Debanding on, Error Diffusion 1 dithering, all trade quality settings unticked and Smooth Motion on

Chroma & Image upscaling: Jinc 3 taps + ar

NNEDI3 Luma doupling: 64 neurons
NNEDI3 Luma quadrupling: 64 neurons

Way faster compared to normal build where I was dropping frames like no other.

In which order should you enable those NNEDI3 algorithms?
I'm guessing Luma 2x, 4x, Chroma 2x, 4x, Chroma upscaling

[QBhd]

Hi all.

I have been testing build 1 with overclocked 290X and I think these are the best settings so far for me without dropped frames:

1920x1080 10bit @ 23.97fps to 2560x1600
Debanding on, Error Diffusion 1 dithering, all trade quality settings unticked and Smooth Motion on

Chroma & Image upscaling: Jinc 3 taps + ar

NNEDI3 Luma doupling: 64 neurons
NNEDI3 Luma quadrupling: 64 neurons

Way faster compared to normal build where I was dropping frames like no other.

In which order should you enable those NNEDI3 algorithms?
I'm guessing Luma 2x, 4x, Chroma 2x, 4x, Chroma upscaling

Quadrupling for that content seems to me like a complete waste of resources... And personally I want Chroma Upscaling before Chroma Doubling.

Also you did not mention what Image downscaling you use

QB

[Qaq]

Test 1 works faster for me = better. Win7, 7750, PCIe 2.0 x16.

[Osjur]

Quadrupling for that content seems to me like a complete waste of resources... And personally I want Chroma Upscaling before Chroma Doubling.

Also you did not mention what Image downscaling you use

QB

CRARLL for downscaling

Just testing how much my card can take before choking.

Still trying to figure out the best compromise between image quality and speed.

[leeperry]

There seems to be just as many ppl who prefer one build or the other, maybe madshi will come up with a version sharing the best of both builds? Or are we gonna end up with a sub-option in mvR's config panel in order fix AMD's sloppy drivers?

I'm still surprised to see that they haven't released WHQL drivers in +4 months now, if we're lucky they'll have the kindness of releasing some by the end of the year. I'm not installing beta drivers, especially when I see that the latest ones might brick R9 290X Lightning boards due to bogus fan speeds under load...that's what I call beta.

[kasper93]

New drivers or not we won't get fix for interop bug. We need someone who can actually reach the person in AMD who could put that in motion. Maybe posting here http://devgurus.amd.com/community/opencl would bring someone's attention. Or at least get some comment whether they fix that or not :)

Tapatalk 4 @ GT-I9300

[Vyral]

There seems to be just as many ppl who prefer one build or the other, maybe madshi will come up with a version sharing the best of both builds? Or are we gonna end up with a sub-option in mvR's config panel in order fix AMD's sloppy drivers?

I'm still surprised to see that they haven't released WHQL drivers in +4 months now, if we're lucky they'll have the kindness of releasing some by the end of the year. I'm not installing beta drivers, especially when I see that the latest ones might brick R9 290X Lightning boards due to bogus fan speeds under load...that's what I call beta.

I still have the 13.9 version installed on my system (first version with openCL). 13.12 brings no new feature or important bug fix, so I have no reason to update. Same thing with the two lastest beta. I hope they will improve the overall performances of their drivers and softwares (especially CCC) in the next version, but I have low expectations.

[jaju123]

New drivers or not we won't get fix for interop bug. We need someone who can actually reach the person in AMD who could put that in motion. Maybe posting here http://devgurus.amd.com/community/opencl would bring someone's attention. Or at least get some comment whether they fix that or not :)

Tapatalk 4 @ GT-I9300

I am in contact with an AMD representative. The issue was forwarded to the driver team a few days ago. Hopefully we will seen some results soon.

[Soukyuu]

For anime I'd say using NNEDI3 at any amount of neurons is worth it. For live content, I still like what it does, but it's more in the open. My opinions.Could you please give some concrete examples? In all my previous tests I was unable to see any improvement with NNEDI3 yet I keep hearing how good it works for anime. Is it something you can see on a screenshot or is it an overall impression while viewing? Which scalers do you use to up/downscale?

[madshi]

Thanks for all the feedback so far, guys!

Quite interesting results. Originally I had expected test build 2 to provide a very big improvement, and I had expected test build 1 to only bring a very small improvement. And development time for test build 2 was several days while development time for test build 1 was just a couple of hours. But on my PC test build 1 actually works better than test build 2. I still decided to let you guys try both, and as usual, one build works better for some users and the other build for other users. <sigh> Guess I'll have to add an option to choose which method to use. FYI, test build 2 does the same thing which test build 1 does, but adds a whole new rendering logic modification (which eats additional GPU RAM) on top.

-------

Build2 is memory hungry, 3.6GB of "system commit" was freed after closing player. I needed to close some programs, because I got only 6GB RAM, 3GB pagefile, 1GB of gpu mem which is full, but I used to it already ;p Windows notified me during playback that I run out of memory. But I had already around 5GB used.
Hmmmm... Build 2 is supposed to consume more RAM, however it should be a fixed amount more and should not increase over time. Could you please check whether that is the case? Or do you have slowly increasing RAM consumption all the time? That would indicate a leak which is not supposed to happen...

Forcing ivtc with deint=ivtc on film in 59 fps source detects 6:4 cadence but doesn't remove frames and gpu load remains high.
Oh ok, I guess it will not remove enough fields in that situation to get from 59[p] to 23[p]. Well, as I said, 59p decimation is on my to do list, but it's probably not going to come soon.

29i works fine with force film mode now, it didn't last I checked months ago. Unfortunately 59i doesn't even when double framerate deinterlacing allowed it still detects 2:2 and plays at 29 fps.
29i and 59i are just two different names for the same thing.

Scaling from 720x480 to 1920x1200 with ED2 and NNEDI doubling at 32 neurons on luma using a Radeon 5850 I am getting:

- 52 frame drops/refresh with 87.9
- 40 frame drops/refresh with test 1
- 31 frame drops/refresh with test 2

I don't know why others with the 5850 are getting so much better performance though.
You are probably using DXVA2 deinterlacing? That turns your 720x480 source into 60fps. Which means that madVR has to process 60 frames per second, while a movie usually only has 24 frames per second. Probably the other users are mostly testing with movies (24fps). Rendering 60fps needs 2.5x more power than rendering 24fps. That might explain why the other 5850 users are getting better results. If they test with 24fps content their GPU has to do 2.5x less work compared to your tests.

If your test content is movie content, too, try forcing madVR into film mode. That will reduce the frame rate to 24fps, too. So you'll suddenly have 2.5x as much power compared to before. But this will only work properly for movies. If you do this for sports or music concerts (or for badly interlaced content), you'll get heavy combing artifacts.

Do the test builds improve anything on Nvidia hardware at all?
I don't expect a big improvement, but maybe a very (very) small one.

Yes, but only with SLI on. SLI is much better with the test builds though still not as fast as without SLI.
Hmmm... Interesting. Are you sure these improvements come from the interop optimizations and not from the new windowed mode (which 87.9 doesn't have yet)?

Yet test build2 has obviously less impact on concurrent GPU deinterlacing job's performance than test build1.
Can you clarify what exactly you mean with that? Are you saying that build 2 works better when using DXVA deinterlacing?

FWIW, the MPC-HC 1.7.4 update seems to fix the issue. Not sure why some versions of MadVR behaved differently with the older versions of MPC-HC though.

Do you still want the debug log?
Good to hear, and no, in that case I don't need a debug log.

Hey Madshi, your reply about the AMD interop problem has been forwarded to the AMD driver team. Hopefully in the future we will see some improvements relating to this issue and maybe a complete resolution.
I'll post any updates in this thread as they come :)
Thanks - that would be quite awesome if they could bring interop up to speed. I believe it would bring another performance jump to NNEDI3 scaling for AMD users.

[flashmozzg]

madshi, have you looked into OpenCL 2.0? Shared Memory looks like it might improve things.

[turbojet]

Originally Posted by turbojet View Post
29i works fine with force film mode now, it didn't last I checked months ago. Unfortunately 59i doesn't even when double framerate deinterlacing allowed it still detects 2:2 and plays at 29 fps.
29i and 59i are just two different names for the same thing.

What I mean by 29i is when bob deinterlaced (double rate) it just duplicates the frame, this is pretty common in taped shows. Madvr under force film mode now detects 2:2 cadence and plays it correctly, it didn't last I checked a few months ago. 59i when bob deinterlaced has 2 unique frames, typical on live broadcasts. Madvr in film mode detects 2:2 cadence and plays at 29 fps, half the frames are missing. it would be nice if it played these at 59 fps when double deinterlaced is allowed in trade quality settings, then you could probably get rid of the film/video mode option.

If you need some 59p video samples when you are working on it, I have a bunch here.

[pie1394]

Can you clarify what exactly you mean with that? Are you saying that build 2 works better when using DXVA deinterlacing?


Yes, it is about the DXVA deinterlacing time... if it is still accurately reported on the test builds.

With build#1, it is more than 2ms. But I didn't memorize the accurate number. Since I noticed that for interlaced samples and none of both builds showed very big difference on stressful NNEDI3 settings, I took the testing focus on build#2 yesterday.

With build#2, it is around 1.2~1.4ms. If my memory is correct, this value is the regular build's number with 1440x1080i / 1920x1080i contents on HD7970.


Actually you have done an impressive job to workaround this inter-op issue! :thanks:

The good part is about the card's fan speed! Both builds obviously produce less GPU loading / heat than the regular build with NNEDI3 settings. So the 3-fan sound of Lantic HD7970 is now at least quieter with some regular NNEDI3 settings. The room temperature is about 25 ~ 30'C recently in the evening -- already 10~15'C higher than the number in the winter. With the regular build + NNEDI3 settings, the fan noise is quite loud.

[Tyestor]

Does anyone know where I can find a detailed (or a semi-detailed...brief even) explanation of (literally) every single option, sub-option, sub-sub-option etc... in madVR (that is current with the latest version (v0.87.9 as of this post))? Thanks.

[DragonQ]

FWIW, changing from 640x480p/25 to a 720x576i/25 source doesn't change my results. Test 2 still lets me use 16 neurons but no more.

[kasper93]

Hmmmm... Build 2 is supposed to consume more RAM, however it should be a fixed amount more and should not increase over time. Could you please check whether that is the case? Or do you have slowly increasing RAM consumption all the time? That would indicate a leak which is not supposed to happen...

Nope, not leaking it just use "a lot" of memory. http://i.imgbox.com/BdMqaEv9.png Funny that using system memory doesn't really slow down the process. So it really must be doing copy back anyway.

[bozokaydin]

My GTX 690 gave a very poor performance when using image doubling. i can not use chroma upcaling, luma and chroma image doubling even at 16 neurons for 720p.it drops frames. Am i doing something wrong?

[6233638]

Does anyone know where I can find a detailed (or a semi-detailed...brief even) explanation of (literally) every single option, sub-option, sub-sub-option etc... in madVR (that is current with the latest version (v0.87.9 as of this post))? Thanks.It's not current, but it covers a lot: http://yabb.jriver.com/interact/index.php?topic=80253.msg546015#msg546015

The main things that aren't covered are the new profiles, NNEDI3, and dither options.

[Anime Viewer]

My GTX 690 gave a very poor performance when using image doubling. i can not use chroma upcaling, luma and chroma image doubling even at 16 neurons for 720p.it drops frames. Am i doing something wrong?

What do you have set for your dithering method? I found that when I tried any type of Error Diffusion my GPU couldn't handle it, but if I set it to Ordered dithering (random should work too) I was then able to get NNEDI3 to run without dropping frames.

[StinDaWg]

Test build 1 is faster than 2 for me.

7850 pci-e 2.0 x16
720p24->1080p
B75AR chroma/luma, Spline4 downscaling
NNEDI3 64 neurons
83% GPU usage build 1
87% GPU usage build 2

I wasn't even able to use 64 neurons before, it topped out at 32.

I can also now do 720p60 with NNEDI3 doubling 16 neurons which I couldn't do before. 82% GPU. Overall a huge improvement!

[Tyestor]

It's not current, but it covers a lot: http://yabb.jriver.com/interact/index.php?topic=80253.msg546015#msg546015

The main things that aren't covered are the new profiles, NNEDI3, and dither options.

Thanks. Now I just need information on artifact removal, image doubling, NNEDI3 vs Jinc 3 taps for chroma upscaling and downscaling, openCL to process NV12, I still have no idea what the flush options do and dithering.

[ryrynz]

I still have no idea what the flush options do and dithering.

Flush options don't need to be changed unless you're having problems (glitches mostly) the dithering (http://en.wikipedia.org/wiki/Dither) affects the image quality, creating smoother transitions and more accurate color reproduction.

I recommend the default ordered dithering setting unless you have the GPU power to select one of the DX11 options and can see a difference over the default setting.

[seiyafan]

Having 16 neurons is generally considered to be comparable to Jinc 3 in terms of having "equivalent degrees" of distortion on an image, though these are often also considered "nicer" artifacts, particularly with anime.

So if I use Jinc 3 as upscaling algorithm, by adding 16 neurons doubling it wouldn't help much right? Because my GPU cannot do 32, so in this case am I better off without doubling?

[StinDaWg]

So if I use Jinc 3 as upscaling algorithm, by adding 16 neurons doubling it wouldn't help much right? Because my GPU cannot do 32, so in this case am I better off without doubling?
What resolution are your files? If you are upscaling 720p to 1080p, image upscaling setting won't matter much when using image doubling, because it's only upscaling chroma.

[James Freeman]

IMO, after NNEDI you should not need Jinc; Lanczos 3 will do.
Especially because Jinc 3 is as heavy as Lanczos 8 (if not more) without noticeable visual improvement.

You can also save power on Dithering and select OD (Ordered Dithering), because in 8-bit (or as low as 4-bit) the visual difference between OD and ED is very small but the performance difference is huge.
I myself use Lanzos 3 AR for upscaling (Chroma included) and Catmull-Rom AR LL for downscaling, OD for dithering with both boxed checked.

I think I prefer running cool and quiet rather than squeezing almost invisible quality at the cost of heat and noise, short life of the GPU, and bigger electricity bill.
IMHO it is not worth pushing the hardware for that minuscule (almost invisible for me) quality improvement.

To be honest, with Hollywood DVD movies I don't even see the improvement of NNEDI Doubling over any other Upscaler like Lanczos.
Maybe NNEDI is more prominent in anime or cartoons (sharp lines)?

Anyway, I'm very grateful for madVR, whether I use its bombastic features or not.

[annovif]

James, probably you have got low end electronics, because the differences are very noticeable in all that settings.

[annovif]

Madshi , for people that use jriver, can you add lumasharpen in madvr? It is just a pixel shader, and thoretically it would be a fast work for you. I don't know if it is good enough for madvr, but i read that all people , here, talk about it . :thanks:

[James Freeman]

James, probably you have got low end electronics, because the differences are very noticeable in all that settings.

I consider myself a very critical observer, I think my bottleneck is my slow AMVA monitor (the most important part) not the GPU or CPU (listed in my signature).
It will be replaced with a faster IPS display ASAP.

I also agree with you that madVR lacks a good Sharpening algorithm.
lumasharpen is good, but I think if anything goes into madVR it has to be second to none, therefor I think madshi will completely rewrite it and make it several time better.

[e-t172]

I consider myself a very critical observer, I think my bottleneck is my slow AMVA monitor (the most important part) not the GPU or CPU.
It will be replaced with a faster IPS ASAP.

Not sure if that's a good idea, because AMVA panels typically have much better contrast.

[James Freeman]

Not sure if that's a good idea, because AMVA panels typically have much better contrast.

Believe me,
I went from Dell U2410 to Asus VN279Q (IPS to AMVA).
I rather suffer from low contrast and IPS glow than off-center contrast/gamma shift.
You have no idea how disrupting it is, much worse than any IPS glow.

Besides, at the cinema the ANSI contrast is not bigger than about ±200:1, so 900:1 with IPS is enough to enjoy a movie and/or edit some photos.
IPS (PLS, AHVA) or OLED, don't compromise for less.
*I see you have a Dell U3014, DON'T SELL IT!

[Shiandow]

Madshi , for people that use jriver, can you add lumasharpen in madvr? It is just a pixel shader, and thoretically it would be a fast work for you. I don't know if it is good enough for madvr, but i read that all people , here, talk about it . :thanks:

If I recall correctly madhsi did mention that it would be somewhat easy to let madvr load a shader from the registry or a text file, which would probably be an easier fix. He also mentioned that it should be incredibly easy for the jRiver devs to add pixel shader support for MadVR.

[*Touche*]

Believe me,
I went from Dell U2410 to Asus VN279Q (IPS to AMVA).
I rather suffer from low contrast and IPS glow than off-center contrast/gamma shift.
You have no idea how disrupting it is, much worse than any IPS glow.

Besides, at the cinema the ANSI contrast is not bigger than about ±200:1, so 900:1 with IPS is enough to enjoy a movie and/or edit some photos.
IPS (PLS, AHVA) or OLED, don't compromise for less.
*I see you have a Dell U3014, DON'T SELL IT!

I have the opposite experience. I haven't really noticed AMVA gamma shift but can't stand the IPS glow on my new monitor.

Cinema's PQ is quite poor but size, sound and atmosphere make up for it. The songle most important PQ factor for me is contrast (deep blacks) so anything less than new MVAs for LCDs is a no go for me.

[James Freeman]

The single most important PQ factor for me is contrast (deep blacks) so anything less than new MVAs for LCDs is a no go for me.

Lets agree to disagree.

Contrast Ratio is important, but far less than what we've made to believe.
The human eye can't even see above about 1000:1 with fixed iris (static contrast).

I just can't stand the AMVA contrast/gamma Shift when I move my head.
Its a 27" monitor so the contrast ratio when looking dead center is around 3000:1 but if I turn my head too look at the edge of the screen the contrast is VERY poor.
That's the problem with AMVA/VA, the high contrast ratio is ONLY in the center (or measured with a calibration device),
but move your head an inch left or right, or just look with your eyes to one of the corners of the display, its FAR less from 3000:1, even far from 1000:1.

This is where IPS shines, it's rock solid from all typical viewing angles.
Have you ever seen/read about Video Reference Monitor or Professional Photo Editing VA/AMVA display?
There is a reason why all of them are CRT, IPS or OLED.

[Procrastinating]

So if I use Jinc 3 as upscaling algorithm, by adding 16 neurons doubling it wouldn't help much right? Because my GPU cannot do 32, so in this case am I better off without doubling?

doubling does exactly what it says; it doubles the luma/chroma resolution. If your source resolution is more than half of your screen resolution, then image upscaling doesn't do anything.

It's unlikely that your screen resolution is that much larger than double your source resolution, so it is recommended that you switch to something "cheap" for the last few pixels(ie bicubic 75 or lanczos).

It depends on the source, but for anime at least, if you're not dropping hundreds of frames a second, you're definitely better off with any doubling at all rather than no doubling at all. As such, you would want to reduce other settings (dither to ordered if you haven't already, then image, then chroma) until you can get it to work.

[seiyafan]

What resolution are your files? If you are upscaling 720p to 1080p, image upscaling setting won't matter much when using image doubling, because it's only upscaling chroma.

They are mostly 1080p, upscaling to 1440.