madVR - high quality video renderer (GPU assisted)

[huhn]

if you can find a 560 4Gb you could use that one.

[Dodgexander]

The best I own is a 660 which is 2gb VRAM sadly.

[huhn]

an RX 560 4gb.

you need at least an maxwell, pascal or newer card to get an HEVC 10 bit decoder(not all of these cards have this decoder)

[nsnhd]

Originally Posted by huhn

if you can find a 560 4Gb you could use that one.

Rx 560 4Gb doesn't have VP9 profile 2 hardware decoder and is slow on madVR NGU, right ? What else as minus in comparison to a Gtx 1050ti ? Is the Rx560 less powerful than the Gtx1050ti in general ?

[Dodgexander]

Originally Posted by huhn

an RX 560 4gb.

you need at least an maxwell, pascal or newer card to get an HEVC 10 bit decoder(not all of these cards have this decoder)

Thats what I read and you confirm my findings. Basically the 1050ti is the way forward.

From my research, 7xx series have a very limited hardware decoder for HEVC.

9xx series have full decoder.
10xx series same.

But its not that thats the issue here, MadVRs scaling requires more shaders and for that it seems a 1050ti is optimal right now.

[huhn]

Originally Posted by nsnhd

Rx 560 4Gb doesn't have VP9 profile 2 hardware decoder and is slow on madVR NGU, right ? What else as minus in comparison to a Gtx 1050ti ? Is the Rx560 less powerful than the Gtx1050ti in general ?

generally yes it was slower (not sure if this is the case for madVR outside of NGU) but supposed to be cheaper but well GPU prices are not what they used to be... the 1050 ti is not heavily affected yet.

it doesn't have a VP9 decoder in general sorry but a hard to use hybrid decoder is not what you want if you are going to use madVR (AFAIK you can't use it out side of browser anyway).

AMD was usual faster with madVR but the polaris NGU problem and the fact that you can't buy a performance AMD GPU for a reasonable price.

[aufkrawall]

Originally Posted by huhn

debanding is nothing a RX 460 can't do at 1080p50 it's so cheap.
my 960 doesn't really care about debanding and this card is only ~50% faster.

Deband increases render time here by ~30% with RX 560 in WQHD 8 bit. Not exactly very cheap to me.

[sauma144]

Is it normal that black bar cropping feature doesn't work with youtube streams?

[el Filou]

Originally Posted by James_b

MadVR meets old school CRT-projector from the 90's (1920 x 816 / 71.952Hz) 2,2 m screen

Waw...

Nvidea edge enhancement 47% 0-255 + Nvidea Video gamma ramps at 1.24

just like Asmodian said: use madVR's post-processing instead of NVIDIA's. Both use the compute units anyway so you don't gain anything by using NVIDIA's.

[Asmodian]

Originally Posted by aufkrawall

Deband increases render time here by ~30% with RX 560 in WQHD 8 bit. Not exactly very cheap to me.

Debanding happens at the source resolution, are you watching WQHD content?

Debanding is probably the fastest remove artifacts option, nothing would be cheap to you.

[aufkrawall]

I originally tested with 4k material.
With WQHD material, the relative performance hit is even bigger. Are you sure that deband is applied before downscaling? Or perhaps it just happens before upscaling, but after downscaling?

I also remember that when I applied deband after downscaling in MPDN, I could save a ton of performance while not really sacrificing quality compared to applying it before downscaling.

[huhn]

depending is done after chroma scaling and it is hard to measure.

and it looks to me that it cost less than bicubic for chroma.
https://abload.de/img/debend2ru4p.png
so it is as free as it can get.

[aufkrawall]

I don't think we agree on the definition of "free". Also looking at single rendertimes is likely just academical since only the total rendertime counts in the end.

[huhn]

ok now i use deband with NGU very high so it only adds 2 % rendertimes

[aufkrawall]

Originally Posted by huhn

ok now i use deband with NGU very high so it only adds 2 % rendertimes

And that is supposed to tell me what? That your GPU didn't run with the same clocks?

[huhn]

you talk about 30 % of all rendertimes that means what?

so to put it into prospective i posted a screen that shows debanding is as fast if not faster than bicubic for chroma.

my clocks doesn't matter if my GPU clocks higher both will be render faster and the ratio will stay about the same.

but yeah i can create screen where debanding is increasing rendertimes by over 140 %. so what's the point?
https://abload.de/img/debend2vtj2z.png

so the only thing that matter is if we can agree that bicubic chroma is a very cheap scaling algorithm so debanding is very cheap too or we disagree about that.

[aufkrawall]

D3D GPU usage without deband (full clocks forced via ClockBlocker): stable ~53%
D3D GPU usage with deband (again full clocks forced): stable ~87%

Very, very far from free.

[huhn]

feel free to post the screen with advanced rendertimes.

if your numbers are correct it should be ~4.8 ms for debanding on a 60 fps source.
edit: can you please check if you use the default trade for quality settings?

[aufkrawall]

Or perhaps a practical example?
Jinc chroma upscale: ~75% GPU usage, no dropped frames
Jinc chroma + deband: ~95% GPU usage and lots of dropped frames (even with "dont analyze gradient angles for debanding" checked)

[huhn]

can you please post a screen with advanced rendersteps plz.