madVR - high quality video renderer (GPU assisted)

[huhn]

I can only repeat what various people have told you before:

29i is supposed to be output as 59p. (Note that many people mix 29i and 59i, but generally mean the same thing, 29 interlaced frames/59 fields per second, ie. what is typically broadcast in NTSC countries)
If its interlaced, it should be output with twice the "frame rate". Its as simple as that.

That sample however is NOT 29i, its 29p. You renamed it with a "deint=video" tag in the filename, which FORCES madVR to deinterlace the video. If you do that, its going to treat it like a normal interlaced file, which as a result does indeed duplicate all frames - but thats only because you forced it to do that, it wouldn't do that on its own.

Interlaced video does NOT contain frames, it contains fields. For technical reasons, two fields are packed into one "frame". So what you do when deinterlacing is first split these two fields again - so your 29 fps interlaced stream becomes 59 fields per second, and then every field is individually deinterlaced, which results in 59 frames per second. Thats how proper deinterlacing HAS to work. If a stream is interlaced, that is what will ALWAYS happen.

i'm pretty sure his main quetion is about 29i with telecine 2:2 content if it is fine to watch this with half frame rate.

a 1080p29 interlaced stream with telecine is normally called telecine in this forum to make the difference clear but the stream is still interlaced.

of cause this doesn't change all the fakt you have stated above proper deinterlacing creates 50/60 fps even with a 3:2 telecine source. a very clear high quality BD with 1080p29i and 2:2 cadence is deinterlanced to 59p with duplicated frames thanks to the inverse telecine function of nvidia and thanks to the pulldown detection of amd.
but in the end a stream like this should be watched with MadVR IVTC or in very rare cases just as progressive.

[Procrastinating]

I figured that it might help madshi and those who want to keep track of the thread if a survey was created which allowed people to enter render times, settings and GPU in a formatted manner.

To achieve that, I created a public survey. The spreadsheet is linked upon submission of some render times. (https://docs.google.com/forms/d/1hx_F1PsC79yJ1lY1wJAB9-sfTz-qtYas4Ew9FYIxcrc/viewform) If anyone has questions, suggestions or problems with the survey, feel free.

I chose a default setting which is probably a realistic benchmark for everyday use of NNEDI3.

[ryrynz]

Madshi, how does madVR's NNEDI settings stack up vs the default settings in Avisynth filter? Was wondering if we'd see options available for nsize and qual etc.

[nevcairiel]

I chose a default setting which is probably a realistic benchmark for everyday use of NNEDI3.

Using NNEDI for chroma upscaling is rather wasted, as the benefit is rather miniscule. This does not seem to be a realistic setting.
Luma doubling is its real power and its real visible advantage over the other modes.

Also, madVR cannot make use of multiple GPUs, mentioning the whole CF/SLI stuff just overcomplicates it.

[ryrynz]

I figured that it might help madshi and those who want to keep track of the thread if a survey was created which allowed people to enter render times, settings and GPU in a formatted manner.


Is it possible to use drop down boxes for options? People can just test what combinations they want which are likely to be what others will want to run anyway.

[jkauff]

Using NNEDI for chroma upscaling is rather wasted, as the benefit is rather miniscule. This does not seem to be a realistic setting.
I think you mean chroma doubling, right?

[madshi]

I figured that it might help madshi and those who want to keep track of the thread if a survey was created which allowed people to enter render times, settings and GPU in a formatted manner.

To achieve that, I created a public survey. The spreadsheet is linked upon submission of some render times. (https://docs.google.com/forms/d/1hx_F1PsC79yJ1lY1wJAB9-sfTz-qtYas4Ew9FYIxcrc/viewform) If anyone has questions, suggestions or problems with the survey, feel free.

I chose a default setting which is probably a realistic benchmark for everyday use of NNEDI3.
This is interesting, but I would suggest to wait for the next madVR build before doing the survey, because I'm currently trying to reduce the AMD OpenCL interop cost.

[huhn]

you find a way to lower data size, so NN is performed on the opencl side?
or are you going with something like d3d -> direct compute -> opencl?
time will tell

[nevcairiel]

I think you mean chroma doubling, right?

His one pre-defined settings string suggested NNEDI for chroma upscaling, and thats what I meant.

Chroma doubling is probably in a similar boat, though.

[Procrastinating]

This is interesting, but I would suggest to wait for the next madVR build before doing the survey, because I'm currently trying to reduce the AMD OpenCL interop cost.

I figured I would let people select any particular build they want at this stage, any data is good data. If you want only a subset of information in the survey though, then I can change that.

Is it possible to use drop down boxes for options? People can just test what combinations they want which are likely to be what others will want to run anyway. Google docs seems to be impressively inflexible, but I'm sure I could try and find a way to turn all the convoluted typing into a series of drop-down boxes, which I think everyone would prefer. I need sleep though.

Using NNEDI for chroma upscaling is rather wasted, as the benefit is rather miniscule. This does not seem to be a realistic setting.
Luma doubling is its real power and its real visible advantage over the other modes.

Also, madVR cannot make use of multiple GPUs, mentioning the whole CF/SLI stuff just overcomplicates it.
I have heard a few people find that chroma upscaling actually benefits more than luma with NNEDI3 in some cases, but if the semi-general consensus is something entirely different from what I understand, then I can change that.

I included crossfire because madshi said that while he isn't trying to work on it, he is not trying to prevent it either. I think it would be interesting nonetheless to see if crossfire performance improves or degrades over time, for when/if madshi wants to start working on crossfire.

The main draw for this I think, besides providing more efficient data for madshi, would be to track performance in various scenarios and hardware configs over time. If I can sort out a nice way to both input and parse information quickly and painlessly with this, it could become quite a handy tool for everyone. I'm thinking about seeing if I can parse specific GPUs and compare their characteristics against rendering times with this system. Of course, the spreadsheet is free for anyone to make use of as well.

[DragonQ]

This is interesting, but I would suggest to wait for the next madVR build before doing the survey, because I'm currently trying to reduce the AMD OpenCL interop cost.
I suppose I will be your prime guinea pig for that? :p

[michkrol]

I figured that it might help madshi and those who want to keep track of the thread if a survey was created which allowed people to enter render times, settings and GPU in a formatted manner.

To achieve that, I created a public survey. The spreadsheet is linked upon submission of some render times. (https://docs.google.com/forms/d/1hx_F1PsC79yJ1lY1wJAB9-sfTz-qtYas4Ew9FYIxcrc/viewform) If anyone has questions, suggestions or problems with the survey, feel free.

I chose a default setting which is probably a realistic benchmark for everyday use of NNEDI3.

Seems like a good idea, however to make it actually comparable you'd need to impose all settings, possibly some predefined presets to chose from and impose the file played. Best to use some freely available material like Big Buck Bunny (www.bigbuckbunny.org), available under Creative Commons rights. That way you can be sure that the testing environment is similar for everyone.
I'd also include GPU usage apart from rendering times, since with NNEDI you get a penalty from just enabling it and increasing neurons doesn't increase the render times that much, especially on stronger GPUs.
Then you just take the results, group them by target (screen's) resolution and settings (presets) to compare them pretty easily.

That's of course just my opinion, but I hope it proves useful for you.

[JonnyRedHed]

Would it be possible to have the OSD display which profile is active when playing please. And to also highlight said active playing profile in the MadVR settings window.

I have to currently look at my GPU usage and temp to work out what's going on. I'm not a very technical user so something as simple as a line on the OSD and a highlight over the profile in settings would be really useful.

[michkrol]

Would it be possible to have the OSD display which profile is active when playing please. And to also highlight said active playing profile in the MadVR settings window.

I have to currently look at my GPU usage and temp to work out what's going on. I'm not a very technical user so something as simple as a line on the OSD and a highlight over the profile in settings would be really useful.

The active profiles are bolded in settings, sometimes you have to close and re-open the settings for this to work, especially after changing auto-select rules.

Showing active profiles in OSD would be kind of impractical, since you can have multiple profiles groups, hence multiple active profiles at once (up to 9 right now).

[nevcairiel]

Would it be possible to have the OSD display which profile is active when playing please.
Depending on your settings, you could have a multitude of profiles active, since you can define multiple profile groups for different settings.
So this is rather impractical for the OSD.

And to also highlight said active playing profile in the MadVR settings window.

The active profile is highlighted in bold right now.

PS:
michkrol was faster :)

[JonnyRedHed]

OSD, Ah ok I understand.

Bolded in settings, ah yes close and re-open. Thank you for explaining.


I can see them working now, thanks.

[Farfie]

Regardless of whether or not I'm upscaling from 720p to 1440p, NN looks like crap. The image is fuzzy and aliased as hell. I realize the theory behind NN for a 1:1 mapping in this scenario, but it does not work out for me.

In my opinion, for anime especially, NNEDI3 is the obvious champion. Even at only 16 neurons, it increases sharpness drastically while keeping aliasing low, even if it is undoubtedly more aliased than jinc, to my eyes. Still as I said, I believe it keeps the aliasing to a minimum, so only if you find yourself extremely sensitive to aliasing should you bounce back and forth between NNEDI3 and Jinc to determine what you find more pleasing. Spline and Lanczos are still great, but I believe they should be reserved for weaker hardware setups. To me, everything else is deprecated. And again, this has been my opinion :)

[Ceremony]

This is interesting, but I would suggest to wait for the next madVR build before doing the survey, because I'm currently trying to reduce the AMD OpenCL interop cost.

about that test:
I am unable to run it on my A10-7850K. It crashes with the following error instantly:
The application has failed to start because its side-by-side configuration is incorrect. Please see the application event log or use the command-line sxstrace.exe tool for more detail.

do i need to install anything before running the benchmark your provided at http://madshi.net/OpenClSpeedTest.zip

[Fullmetal Encoder]

Why? What is worse about 480p/576p -> 2160p compared to 480p/576p -> 1080p?

Watching a DVD on a 70" screen I can understand some shudders but that is due to the size not the resolution, the extra resolution only helps.

It would be worse since the larger the screen resolution is compared to the resolution of the source image the larger the number of pixels that must invented to fill that higher resolution in the process of interpolating the image. Generally speaking I think that an image that has thousands of guesses as to what should be in those pixels is less desirable than an image that has hundreds of guesses. Unless, of course, there is something seriously awry with my understanding of how scaling is done.

[John Carmack]

about that test:
I am unable to run it on my A10-7850K. It crashes with the following error instantly:


do i need to install anything before running the benchmark your provided at http://madshi.net/OpenClSpeedTest.zip

I had this error too. I downloaded an obscure VCredist 2005 from MS and it worked.

[Ceremony]

I had this error too. I downloaded an obscure VCredist 2005 from MS and it worked.nope, didnt work. installed these two:
http://www.microsoft.com/en-us/download/details.aspx?id=3387
http://www.microsoft.com/en-us/download/details.aspx?id=5638

no change, same error... :(

[Asmodian]

It would be worse since the larger the screen resolution is compared to the resolution of the source image the larger the number of pixels that must invented to fill that higher resolution in the process of interpolating the image. Generally speaking I think that an image that has thousands of guesses as to what should be in those pixels is less desirable than an image that has hundreds of guesses. Unless, of course, there is something seriously awry with my understanding of how scaling is done.

This is not true; the more resolution at the destination the better. It isn't "guessing" it is math.

Some examples thinking about vertical resolution only through the same reasoning holds true to for horizontal.

480 -> 1080
source lines want to display on 2.25 lines of the destination.

480 -> 2160
source lines want to display on 4.5 lines of the destination.

4 out of 5 (80%) of the lines are the same values as the source at 2160 while only 2 out of 3 (66%) of the lines are the same at 1080.

Of course the math doesn't really work like that, no original source pixels are left after either resize (except for NN which is why tp4tissue likes it), but the error in the scaling does decrease this way.

Another way to phrase it; If you are trying to approximate an image it is easier to do if you have more pixels to use to do the approximation, the error at each pixel can be smaller. It doesn't matter if the image is a real image or a 720x480 set of pixels, more pixels still allow a better approximation.

[vaporizer]

nope, didnt work. installed these two:
http://www.microsoft.com/en-us/download/details.aspx?id=3387
http://www.microsoft.com/en-us/download/details.aspx?id=5638

no change, same error... :(

openclspeedtest needs this vc redist 2005 sp1 x86 pack: KB2538242
http://www.microsoft.com/de-de/download/details.aspx?id=26347

[Procrastinating]

Updated the survey/spreadsheet (https://docs.google.com/forms/d/1hx_F1PsC79yJ1lY1wJAB9-sfTz-qtYas4Ew9FYIxcrc/viewform) To be easier to input information. You can enter settings relatively painlessly now. It looks like you can't do too much with the forms, unless you created a whole separate google app, which is possible I guess.

[Asmodian]

Updated the survey/spreadsheet (https://docs.google.com/forms/d/1hx_F1PsC79yJ1lY1wJAB9-sfTz-qtYas4Ew9FYIxcrc/viewform) To be easier to input information. You can enter settings relatively painlessly now. It looks like you can't do too much with the forms, unless you created a whole separate google app, which is possible I guess.

It looks very nice but I have to agree with michkrol. We need more standardization to get easily comparable data. Especially in settings and video clips, that way you reduce user error and confusion a lot. Just an idea, I know collecting and hosting video samples is a lot more work. Maybe just specify horizontal dimensions too; 4:3 720p is much easier on NNEDI3 than 16:9 720p. edit: Maybe ask users to use software decoding too? One less variable. :)

Edit: at your default set of settings my GPU declocks so I get... less than comparable rendering times. I am not sure how to deal with this except using default settings that do not down clock on anyones system. That means 200ms+ rendering times on some systems though so probably isn't a solution. Maybe include a field for GPU clock speed?

Also, madVR cannot make use of multiple GPUs, mentioning the whole CF/SLI stuff just overcomplicates it.

Sadly SLI or not does change the results a lot. I turn off SLI when watching <=720p content so I can turn the NNEDI3 neurons up. I would love to know about CF.

I wonder what is actually happening with SLI on? The second card gets some significant GPU usage but much less than the main card.

1280x720p24 -> 2560x1440 @ 72Hz, Bicubic75 AR chroma, Bicubic75 AR image, NNEDI3 128 Luma doubling, No smooth motion, no debanding, Ordered Dither, 3DLUT calibration, Windowed Overlay.

SLI on 41.2ms*
GPU0 81% @ 1097 MHz, 19% PCI-E, 7% memory controller
GPU1 18% @ 836 MHz, 16% PCI-E, 0% memory controller

SLI off 29.9ms*
GPU0 67% @ 1097 MHz, 5% PCI-E, 7% memory controller
GPU1 00% @ 324 MHz, 0% PCI-E, 0% memory controller

*rendering times measured without nvidia inspector open but it didn't seem to affect them.

GTX Titans, 3770K @ 4.6 GHz, Z77 chipset, each GPU is on PCI-E 3.0 x8, 32GB DDR3-2133CL9.

Note: This is with Zoom Player forced to "Single-GPU" SLI rendering in the drivers, not that the setting seems to help vs nothing set for Zoom Player (forcing AFR rendering or something on Zoom Player flashes black frames trying to give you a seizure).

edit: I cannot seem to change the SLI behavior anymore, windowed overlay, FSE, Old Windowed, or New Windowed. They all act the same no matter what I set in the drivers (except disabling or enabling SLI completely of course). This isn't bad per say as the only thing I could do before was break playback.

[Ceremony]

openclspeedtest needs this vc redist 2005 sp1 x86 pack: KB2538242
http://www.microsoft.com/de-de/download/details.aspx?id=26347

thanks, that did the trick! Here are the measurements of a A10-7850K APU (with 2400MHz CL11 DualChannel RAM):
https://forum.doom9.org/attachment.php?attachmentid=14140&stc=1&d=1397456902

[Procrastinating]

Added clocks and changed a few things, Google doesn't like moving columns linked to forms around, apparently. Learned regex.

What sort of default would you guys recommend, namely madshi? One which has sub-40ms render times for the high end say, and above-40 for low-end. Maximizing nnedi of course.

Tears of steel seems like a pretty reasonable standard video. If all is decided tomorrow, I can reorganise the survey.

[6233638]

480p content is 720x480, not 848x480. (and it should actually be scaled to 853x480)
A large amount of the SD content I watch is PAL, which would be 720x576.

40ms as a target would only be suitable for film content.
I watch a lot of videos which require rendering times to be below 16ms. (60fps)


Does 10-bit content put any additional strain on the GPU for scaling, or is it just more demanding because decoding has to be done in software? (thus irrelevant)
I wouldn't know, as I thought that format was only used with pirated content.

[nevcairiel]

Does 10-bit content put any additional strain on the GPU for scaling

It does not. Although since its twice the size, it might put additional load on the memory controller during the upload stage, which may however not be all that important.

[James Freeman]

Do you guys think that NNEDI3 for Chroma is a waste of resources?

[michkrol]

480p content is 720x480, not 848x480. (and it should actually be scaled to 853x480)
A large amount of the SD content I watch is PAL, which would be 720x576.

40ms as a target would only be suitable for film content.
I watch a lot of videos which require rendering times to be below 16ms. (60fps)

Does 10-bit content put any additional strain on the GPU for scaling, or is it just more demanding because decoding has to be done in software? (thus irrelevant)
I wouldn't know, as I thought that format was only used with pirated content.

You do realize, that depending on where you live you might be allowed to make personal copy/backup of your media for your personal use?

It might make sense to deinterlace/ivtc and scale the video to correct aspect ratio if you want to watch it on an older TV with built-in video player or a (cheap) hardware media player with very limited format support and you are re-encoding it anyway, since it was broadcasted at some crazy bitrate and you want to (re)watch it later. And once again it's perfectly legal in some countries.

As for the 10-bit H264, the need to decode in software is the biggest difference, so perhaps a "DXVA-N/DXVA-CB/software decoder" option and always using 8-bit 4:2:0 content (most common) would be good for this tests.

Once again it's not used for pirated content only, you might opt to use 10-bit while encoding your streams to save space or improve quality/size ratio. I hear it helps even with 8-bit sources. It is useful with computer-rendered movies/animations to avoid banding with lower bitrate.

[6233638]

You do realize, that depending on where you live you might be allowed to make personal copy/backup of your media for your personal use?Of course, but why would you not just leave it in its native format? Conversion takes time and resources, and results in worse image quality.

It might make sense to deinterlace/ivtc and scale the video to correct aspect ratio if you want to watch it on an older TV with built-in video player or a (cheap) hardware media player with very limited format supportI don't see the relevance to madVR.


Do you guys think that NNEDI3 for Chroma is a waste of resources?I wouldn't say that NNEDI3 chroma scaling is a waste, but there are plenty of other things you might want to enable first.

As far as I can see, there is zero reason to enable chroma doubling though.

[nevcairiel]

Do you guys think that NNEDI3 for Chroma is a waste of resources?

I use NNEDI3 chroma scaling for 1080p content on a 1080p screen, just because the GPU is idle otherwise, but if I need to scale the image itself, I would prefer higher neurons of Luma doubling any day.

[huhn]

Of course, but why would you not just leave it in its native format? Conversion takes time and resources, and results in worse image quality.

this forum is about creating backups from dvds. and if you want to IVTC your disc (and there are very very good reason to do this) you need to reencode it. MadVR IVTC is good no question but it got problems with VFR, cadence break/unusal cadences. removing CM out of tv streams is also a very good thing to do you also got deblocking debanding more types of IVTC NNEDI3 deint and lots lots more. 848x480 is mod of 16 this is very important (or was...) for encoding and thanks to croping left and right 2 pixel is comes down very very close.

Watching a 3 gb TV stream with 2:2/3:2/4:2:2:2 telecine deint by MadVR doesn't look nice at all reencoding. it to about 200-1000 mbsaves a lot of space.

[iSunrise]

@madshi:
Blaire linked to your recent workaround (your changelog for 0.87.9) for the NV driver issue and asked me via PM, if there still is a driver fix needed.

I kinda feared this would happen, since your woraround takes the pressure off of NV to fix an issue no one else (besides you and people that use NV hardware with madVR) seems to care about (that's how I interpret it).

It looks to me that they were in the process of working on the fix, but they re-checked if they have the newest madVR version to test against.

Now, what should I tell him? Some technical details would probably be helpful. Also how you (if memory serves right, a madVR user actually came up with the idea) worked around the bug, so NV knows where and what to search for.

[James Freeman]

I wouldn't say that NNEDI3 chroma scaling is a waste, but there are plenty of other things you might want to enable first.


I use NNEDI3 chroma scaling for 1080p content on a 1080p screen, just because the GPU is idle otherwise, but if I need to scale the image itself, I would prefer higher neurons of Luma doubling any day.

Yes, I mean for scaling 1080p video (Chroma) on a 1080p or 1440p Display.

I CAN see a small difference in chroma aliasing with NNEDI3 vs Lanczos, but the question is whether its worth the extra 30% GPU hit?
I'm also using Ordered Dithering instead of Error Diffusion (another 30% GPU hit).

I guess I prefer a cool running system (and a low electricity bill :D) instead of squeezing every last drop of "almost invisible" performance from the system.

[Asmodian]

I CAN see a small difference in chroma aliasing with NNEDI3 vs Lanczos, but the question is whether its worth the extra 30% GPU hit?

Only you can answer this for sure.

I'm also using Ordered Dithering instead of Error Diffusion (another 30% GPU hit).

I guess I prefer a cool running system (and a low electricity bill :D) instead of squeezing every last drop of "almost invisible" performance from the system.

But this makes me think not. ;)

[leeperry]

Humm, this page really makes a strong case for GPGPU on maxwell: Nvidia’s new Maxwell-powered GTX 750 Ti is hyper-efficient, quiet, a serious threat to AMD (http://www.extremetech.com/gaming/176785-nvidias-new-maxwell-powered-gtx-750-ti-is-hyper-efficient-quiet-a-serious-threat-to-amd/3)

Apparently miners are pretty happy with its low TDP, as they said in that url "GCN’s advantages in the GPU compute market are not so great that superior power consumption and per-core scaling can’t close the gap".

I just saw a sale on this 760 (http://www.gigabyte.com/products/product-page.aspx?pid=4663) but it's very hot and pretty sluggish with GPGPU apparently.

FWIR in a local review, a slightly o/c 750Ti is only barely slower than my o/c 7850 in some games(saving 70W in the process) and some ppl claim that AMD's success in GPGPU was more a matter of luck with their choice of architecture than anything else, very eager to try that highly o/c 750Ti (http://www.gigabyte.com/products/product-page.aspx?pid=4984), hopefully it'll get here by the end of the week :)

[Asmodian]

That 750Ti looks like the better option (vs 760) to me. A very nice madVR card!

I will wait for the giant die version of Maxwell, if they ever release it. :D

[hu1kamania]

I've been using the 750 Ti since early March, and I feel like I have it pretty dialed in for my needs. I'd like to share my current MadVr configuration so folks out there can get an idea of it's capability, and determine if it's the right card for them.

In addition to MadVr, the GPU is also responsible for LavFilters via MPC-HC. Lav set to CUVID HW Deinterlacing, 50p/60p with High-Quality Processing enabled. I'm also using LumaSharpen as a post-resize shader. This is all under a native resolution of 1920x1080.

Image upscaling: Lanczos 3 AR
Image downscaling: Catmull-Rom AR+LL
ChromaUpscaling: 1080p NNEDI3 32 neurons / Everything else Bicubic 75 AR
Image Doubling: I will share my profile so you can see the range with which each profile operates:

1080p: 16 neurons (Not with NNEDI3 ChromaUpscaling, and it's not used often.)
720p < 24fps: 64 neurons
720p > 24fps: 32 neurons
720p > 30fps: 32 neurons
SD < 24 fps: 128 neurons
SD > 24 fps: 128 neurons
SD > 30 fps: 32 neurons

if (srcWidth >= 1900) and (srcHeight >= 721) "1080p"
else if (srcWidth >= 956) and (srcWidth < 1900) and (srcHeight >= 500) and (deintFps <= 23.98) "720p < 24fps"
else if (srcWidth >= 956) and (srcWidth < 1900) and (srcHeight >= 500) and (deintFps > 23.98) and (deintFps < 29.7) "720p > 24fps"
else if (srcWidth >= 956) and (srcWidth < 1900) and (srcHeight >= 500) and (deintFps > 29.7) "720p > 30fps"
else if (srcWidth <= 956) and (srcHeight <= 576) and (deintFps <= 23.98) "SD < 24fps"
else if (srcWidth <= 956) and (srcHeight <= 576) and (deintFps > 23.98) and (deintFps < 29.7) "SD > 24fps"
else if (srcWidth <= 956) and (srcHeight <= 576) and (deintFps > 29.7) "SD > 30fps"
Debanding: high
Dithering: ED, option 2 with both boxes checked
Nothing checked under trade quality for performance


I have my card overclocked to 1250Mhz, and boosts stable up to 1388Mhz. GPU Memory is at 1463.

I've been following madvr for quite a long time now, and I've noticed a lot of conjecture towards where the GTX 750 Ti stands in terms of performance. Hopefully this helps reign in some of that conjecture.

i7 3770k @ 3.7Ghz GTX 750 Ti Windows 8.1 64 bit

[leeperry]

Goodie, goodie! Well, that seems faster than my 7850 :)

Using monostatic ED1 LL, low debanding, CC AR LL downscaling, J3AR upscaling for chroma & luma, with my 1Ghz 7850 I'm capped to:
-64x NNEDI luma for <=30fps SD to 1080p
-64x NNEDI luma for <=24p >=1.85 720p@1080p
-32x NNEDI luma for >=25p 720p@1080p

Which model is it please? How hot does it get under load?

[hu1kamania]

Goodie, goodie! Well, that seems faster than my 7850 :)

Using monostatic ED1, low debanding, CC AR LL downscaling, J3AR upscaling for chroma & luma, with my 1Ghz 7850 I'm capped to:
-64x NNEDI luma for <=30fps SD to 1080p
-64x NNEDI for <=24p >=1.85 720p@1080p
-32x NNEDI for >=25p 720p@1080p

Which model is it please? How hot does it get under load?

I'm using the EVGA GTX 750Ti 2GB. Just the normal, not SC or FTW. I've never seen it hit 80C while heavily overclocked. Normally it reaches about 70C, and the fan is still quite low. If it's not overclocked it sits in the low 50's.

[QBhd]

Yes, I mean for scaling 1080p video (Chroma) on a 1080p or 1440p Display.

I CAN see a small difference in chroma aliasing with NNEDI3 vs Lanczos, but the question is whether its worth the extra 30% GPU hit?
I'm also using Ordered Dithering instead of Error Diffusion (another 30% GPU hit).

I guess I prefer a cool running system (and a low electricity bill :D) instead of squeezing every last drop of "almost invisible" performance from the system.

I personally rank the resource hogs like this:

Luma Doubling==> Chroma Upscaling==> Dithering==> Chroma Doubling

If I have resources to spare I will toss in chroma doubling neurons (I don't pay for electricty and my system's stupid cool) just to make sure I am getting every last drop of performance.


if (srcWidth >= 1900) and (srcHeight >= 721) "1080p"
else if (srcWidth >= 956) and (srcWidth < 1900) and (srcHeight >= 500) and (deintFps <= 23.98) "720p < 24fps"
else if (srcWidth >= 956) and (srcWidth < 1900) and (srcHeight >= 500) and (deintFps > 23.98) and (deintFps < 29.7) "720p > 24fps"
else if (srcWidth >= 956) and (srcWidth < 1900) and (srcHeight >= 500) and (deintFps > 29.7) "720p > 30fps"
else if (srcWidth <= 956) and (srcHeight <= 576) and (deintFps <= 23.98) "SD < 24fps"
else if (srcWidth <= 956) and (srcHeight <= 576) and (deintFps > 23.98) and (deintFps < 29.7) "SD > 24fps"
else if (srcWidth <= 956) and (srcHeight <= 576) and (deintFps > 29.7) "SD > 30fps"


You can clean up your script by reversing the order of things and going from small to lager values. Something like mine:

if (srcHeight < 481) and (deintFps < 31) "SD 30fps"
elseif (srcHeight < 541) and (deintFps < 26) "1/2 HD"
elseif (srcHeight < 541) and (deintFps < 31) "1/2 HD 30fps"
elseif (srcHeight < 721) and (deintFps < 26) "720p24"
elseif (srcHeight < 721) and (deintFps < 31) "720p30"
elseif (srcHeight < 721) and (deintFps > 31) "720p60"
else "1080p24"

It makes it easier to insert new lines and edit if you need to

QB

[leeperry]

I'm using the EVGA GTX 750Ti 2GB. Just the normal, not SC or FTW. I've never seen it hit 80C while heavily overclocked. Normally it reaches about 70C, and the fan is still quite low. If it's not overclocked it sits in the low 50's.
OK, thanks for the infos!

Are you using PCI-E 2.0 or 3.0?

What's your ASIC quality (http://www.ozone3d.net/public/jegx/201201/gpuz-058-asic-quality-menu.jpg) in GPU-Z BTW?

[hu1kamania]

Thanks for the suggestion QBhd

OK, thanks for the infos!

Are you using PCI-E 2.0 or 3.0?

What's your ASIC quality (http://www.ozone3d.net/public/jegx/201201/gpuz-058-asic-quality-menu.jpg) in GPU-Z BTW?

I am showing an ASIC quality of 78% with my current configuration.

Edit: oh yeah, and it's PCI-E 3.0

[mindbomb]

i'm having trouble conceptually trying to imagine what nnedi3 artifacts would look like when used in chroma scaling. It would be colors coming outside of the objects?

[Procrastinating]

nnedi3 artifacts tend to connect details where there was no connection previously, so I would imagine you could possibly notice colors appearing outside of their outlines.

[baii]

Not sure if mentioned previously.

Currently running a bay trail z3740 tablet,
stable (that is, until it start throttling due to temperature xD).24fps 1080p downscale to 1200 x 800 using quick sync/DXVA decode, DXVA downscaling, random diethering and FRC. Havent had a chance to get a profile for color management, hopefully it can handle it when 3dlut is added.

struggles with 1080i, drop frame if deinterlanced to 60fps, repeated frame if trade quality option "half frames" is used.

[turbojet]

madshi: Do you have any plans for using ivtc with 50/59/60 fps sources? It can be a large performance gain, in my case would allow nnedi3 doubling of 720p. I have some clips of 23, 25, 29 and 59 progressive in 720p59 if they are of any use.

[StinDaWg]

madshi: Do you have any plans for using ivtc with 50/59/60 fps sources? It can be a large performance gain, in my case would allow nnedi3 doubling of 720p. I have some clips of 23, 25, 29 and 59 progressive in 720p59 if they are of any use.
+1

Lots of stuff I have from Fox HDTV is 720p59.94 but actually 720p23.976 filmed.