madVR - high quality video renderer (GPU assisted)

[SamuriHL]

Originally Posted by brazen1

Uhm, I'm not sure what you're trying to educate me/us about? Quite simply, using my GTX 960 4GB for the past two years, if I turn off FSE, no matter what I do, 3D enables on the display but not the player using madVR. The moment I turn FSE back on, everything works perfectly and I attribute that to the FSE setting I just toggled per the results. Now I don't know how it works for you especially since you are using an AMD card and as you wrote haven't tested Nvidia. Fwiw, I too wrote a script to engage stereoscopic system wide a year ago when I discovered HDR playback disables it for some odd fluke of a reason no one has addressed. The difference between my script and the one you copy and pasted of Manni's is his requires a manual input to engage it and mine is automated via software front end. I've just seen you insist a bunch of times info that doesn't correlate with my real world tests and usage. This often leads to misinformation for others. This isn't to say perhaps stereoscopic engages properly for others without FSE, but I can assure it doesn't work over here. No hard feelings but before you correct me or anyone else, perhaps you should set up an identical environment and do the actual test yourself. It seems to me you've been having problems for weeks here with the most basic of things and suddenly you've become an authority. Again, no offense and with all due respect that's just how I see it.

Not to add fuel to a fire and my ONLY reason for responding is to provide more data here, however, on my Windows 10 insider build HTPC with a 1060, I am able to run 3D in windowed mode. That is NOT to discount your issue with having to use FSE, simply to provide more data that it's not at all consistent between users.

[ryrynz]

Calm down with the quoting, brazen1 don't quote entire posts with batch files.. You can edit the quote to one line..

[Razoola]

I got a quick question about DXVA2 copyback in relation to profile 2, UHD and the GTX 1030 (and maybe 1050) and LAV / madVR.

I'm sure I am correct to think that in a dual GPU PC setup using LAV and DVXA2 copyback that I could use a GTX 1030 to decode video profile 2 that is then output on another GPU to the display (which does not have provile 2). Now, in a setup like this am I right to assume that the memory on the 1030 does not come into play when it comes to the stuff madVR does and the only important memory madVR is intrested in from a GFX card standpoint would be that on the card with the port outputting GFX to the display?

So if for example I have a GTX980 and a GTX1030, madVR would not touch the 2gig memory on the GTX1030 and would only be limited by the 4gig memory on the GTX980? At the same time LAV would only HW decode on the GTX 1030 which is then passed (copyback) to the GTX 980?

Has anyone got a setup running like this (does not have to be those two GFX cards) and if so, is there a big hit in time for the cards to copy data from one to the other?

[mytbyte]

Originally Posted by ryrynz

Don't get the 1030, it's too slow. My 960 handles hdr only just.. 60fps is no go.. Plus no extra image processing and lower end chroma options.. If you're taking 4K/HDR seriously it's 1060+ otherwise 1050ti minimum imo. 950/960/1050 great for 480-1080p on the cheap.

It's me who wants to get the 1030 and it's for a second, old rig in another location, no sense in anything dearer than 1030/R550 (not in US so prices don't compare)...I have the 960 in my main rig and it runs a certain HFR movie by Ang Lee in 60 fps + others just fine with any MadVR Options, basically only the Error diffusion kills it...and DXVA downscale in copy-back mode...

[huhn]

Originally Posted by Razoola

So if for example I have a GTX980 and a GTX1030, madVR would not touch the 2gig memory on the GTX1030 and would only be limited by the 4gig memory on the GTX980? At the same time LAV would only HW decode on the GTX 1030 which is then passed (copyback) to the GTX 980?

d3d11 can do this better than dxva2. for DXVa2 you need a active display connection to the 1030 for d3d11 you don't need one.

the biggest possible problem in such a setup should be the PCIe speed. i'm not sure if UHD 60 FPS can be done with PCIe 3 x8 but 4x could is dangerously low.

so if your board support 2 GPU with PCIe 3 8x it may work if it only supports 16x 4x you maybe in trouble.

@mytbyte
HDR->SDR conversion needs a lot of processing power feel free to ignore it and buy a card that can't do it. HDR passthrough is fast HDR->SDR conversation is not.

like a said before the 550 is not worth it's money get a 560 for little more that is 2x as fast.

if you want YT HDR you need either a beefy CPU or a 1030-1050 ti.
if you want a GPU that can do HDR->SDR conversation you need a GPU with enough processing power to do that.

[mclingo]

Originally Posted by brazen1

Uhm, I'm not sure what you're trying to educate me/us about? .......

Here's the thing, I too thought this way, I was convinced FSE was required on NVIDIA until I tested it myself. When I said I hadnt tested it I meant I had not tested this batch file, 3D was working on my system with FSE turned off but it wasnt always working due to the 3D turning itself off all the time, if I deployed the batch file 3D should be working fine. I went back to AMD because I was getting system hangs and far more black screen loss of handshake than I was with My AMD card, and my AMD card always works in 3D.

No idea why your NVIDIA card doesnt work with it off then, very odd.

EDIT - have you tried windowed mode?

[mclingo]

Originally Posted by brazen1

This question for example has been asked and answered more times than I can count and not only in this forum either. Use the search function.

tried that, too many variables, yes or no would suffice, hardly putting anyone out.

[huhn]

luma is more important than chroma.
use your eye to see what works best for you.

i personally will never go lower than super XBR on chroma.
and i personally would never use a signal that subsamples or a TV that can't do 4:4:4 because the image quality differences are obvious to me.

[Siso]

Originally Posted by huhn

i personally will never go lower than super XBR on chroma.

How about NGU-AA-medium for chroma, and why you like super-xbr-100 for chroma?

[huhn]

correct me if i'm wrong but it is cheaper that's why. NGU low was comparable(faster) in speed to super xbr.

[oldpainlesskodi]

Nvidia Driver 390.65 out - MadVR HDR fixed (as madshi said it would - nice).

K

[mclingo]

Originally Posted by huhn

luma is more important than chroma.....

Thanks, you're probably more sensitive to PQ differences than me, I only see a difference on sub 1080 material and only a small fraction of my collection is this size.

[Siso]

Originally Posted by huhn

correct me if i'm wrong but it is cheaper that's why. NGU low was comparable(faster) in speed to super xbr.

That's good, but how about in terms of quality?

[aufkrawall]

Originally Posted by huhn

the biggest possible problem in such a setup should be the PCIe speed. i'm not sure if UHD 60 FPS can be done with PCIe 3 x8

Of course it can be done with Nvidia, I had no problems with 16x PCIe 2.0 (which is the same bandwidth).

[huhn]

well i hope you tested that with heavy subtitles which could increases bandwidth need by by let's say "a lot".

the x4 is the real problem you need a 8x board in the first place...

@Siso
try it out it's fine.

[aufkrawall]

Originally Posted by huhn

well i hope you tested that with heavy subtitles which could increases bandwidth need by by let's say "a lot".

Why would subtitles increase bandwidth by a lot? Is the whole subtitle container (or how you'd call it) resend for every frame?

[mclingo]

I too dont think the PCI express 2.0 will be a bottleneck, if a PCI-E SSd can surpass 500mb/s thats more then enough bandwidth.

[Asmodian]

Originally Posted by aufkrawall

Why would subtitles increase bandwidth by a lot? Is the whole subtitle container (or how you'd call it) resend for every frame?

They are rendered to images on the CPU at the display resolution.

[clsid]

That is not true for text based subtitles. Only PGS subs sometimes are encoded at resolution as video, which then gets scaled.

[huhn]

They are rendered to images on the CPU at the display resolution.

that would add zero bandwidth and vsfilter does it this way.
the "new subtitle interface" it uploading image that are scaled or even uploaded at the target resolution.
remember the nnedi3 copyback issue which was at best 120 FPS for FHD.

I too dont think the PCI express 2.0 will be a bottleneck, if a PCI-E SSd can surpass 500mb/s thats more then enough bandwidth.

a 8 bit UHD 60 fps file needs 746 mb/s ignoring overhead and stuff like this...
the point is not pcie 16x 2.0 but if pcie 4x 3.0 works.