madVR - high quality video renderer (GPU assisted) - Page 1838

Uoppi · 8th March 2016, 18:46

Quote:

Originally Posted by Werewolfy

My problems with interlaced DVD (I'm talking about DVD but I think it's related to all interlaced content, it's just that 99% of my interlaced content is DVD) are nothing related to D3D11 exclusive mode or the latest MadVr version, you're right I never mentioned it. I've always found it very hard to avoid dropped frames with deinterlacing activated and "don't rerender frames when fade in/out is detected" unchecked because I can't use high CPU or GPU queue, the queues don't fill completely with deinterlacing activated. The only way I've found so far is to use 23-14-6 queues but even with that combination, I'm not totally confident to avoid dropped frames with some DVD...

Maybe try upping the GPU queue size even more?

I need to use a GPU queue of at least 20 to avoid dropped frames with interlaced NTSC DVDs. Any less gives me occasional and "unexpected" dropped frames. After extensive testing, 20-20-6 works for me perfectly.

(My setup is Win10, factory-overclocked GTX 960 and a dual core Ivy Bridge Pentium, using D3D11 windowed mode.)

Px · 8th March 2016, 18:52

Quote:

Originally Posted by madshi

That's interesting. Can you please create a debug log from this situation with v0.90.15?

http://www.cs.iptcom.net/tmp/full-logs-9.0.15.7z - uploaded debug logs along with fresh crash log

Uoppi · 8th March 2016, 18:58

Quote:

Originally Posted by madshi

because I personally think watching any content with 3:2 pulldown is simply "wrong", and most users would prefer using smooth motion FRC (or minor flickering) over that. So I believe there's only a very small number of users who require optimized 3:2 pulldown smoothness.

I'm not sure if I understand correctly but why would I want to watch a 3:2 pulldown movie of 23.976/24 fps at 60 Hz instead? I never use smooth motion anyway (based on my quick tests I can't stand the ghosting) and wouldn't I need to use much lower PQ settings because of smooth motion's higher resource demands?

iSeries · 9th March 2016, 01:06

Hi, using 2d ssim causes a lot of dropped frames for me. In the stats its listed twice (like this 'ssim 2d100 ar < ssim 2d100 ar') and render times go way up. 1d ssim works fine and is listed in the stats only once as expected. Is that normal for 2d?

ryrynz · 9th March 2016, 01:13

Quote:

Originally Posted by iSeries

using 2d ssim causes a lot of dropped frames for me. In the stats its listed twice (like this 'ssim 2d100 ar < ssim 2d100 ar') and render times go way up. 1d ssim works fine and is listed in the stats only once as expected. Is that normal for 2d?

No. Could you post a screenshot of the OSD?

iSeries · 9th March 2016, 01:15

Quote:

Originally Posted by ryrynz

No. Could you post a screenshot of the OSD?

Sure, here or to an image hosting site?

huhn · 9th March 2016, 01:18

take an image host please

iSeries · 9th March 2016, 01:31

Quote:

Originally Posted by ryrynz

No. Could you post a screenshot of the OSD?

Here you go. Seems to only happen when supersampling 1080p > 1080p. Works as expected when doubling eg 720p and downscaling to 1080p.

http://imgur.com/vrtHks6

ryrynz · 9th March 2016, 01:43

Quote:

Originally Posted by iSeries

Here you go. Seems to only happen when supersampling 1080p > 1080p. Works as expected when doubling eg 720p and downscaling to 1080p.

http://imgur.com/vrtHks6

Can confirm that on my OSD too. Something's amiss. I think it's just the OSD though.

SSIM 1D is about twice as fast as SSIM 2D, that's normal. SSIM 2D is quite intensive because it uses Jinc kernel for downscaling.
My GTX 960 finds it too demanding generally as well, just stick to 1D until you get something faster. Better use of resources to have more NNEDI3 neurons than using SSIM 2D IMO.

iSeries · 9th March 2016, 02:10

Quote:

Originally Posted by ryrynz

Can confirm that on my OSD too. Something's amiss. I think it's just the OSD though.
SSIM 1D is about twice as fast as SSIM 2D, that's normal. SSIM 2D is quite intensive because it uses Jinc for downscaling.

In all other doubling / downscaling scenarios, 2d works fine (as does jinc), only adds around 2-3ms to render times. When supersampling render times shoot up into the 80s ms with 2d, from around 35ms with 1d. Jinc downscaling also shows the same behaviour when supersampling, its in the osd twice and render times shoot up into the 80s.

ryrynz · 9th March 2016, 02:17

Quote:

Originally Posted by iSeries

In all other doubling / downscaling scenarios, 2d works fine (as does jinc), only adds around 2-3ms to render times. When supersampling render times shoot up into the 80s ms with 2d, from around 35ms with 1d. Jinc downscaling also shows the same behaviour when supersampling, its in the osd twice and render times shoot up into the 80s.

I think you may underestimate how demanding Jinc and AR are, only madshi can comment on the performance and OSD though really. No problems have been reported about the performance of Jinc or AR, so I expect that everything is fine here.

iSeries · 9th March 2016, 02:22

Quote:

Originally Posted by ryrynz

I think you may underestimate how demanding Jinc and AR are, only madshi can comment on the performance and OSD though really. No problems have been reported about the performance of Jinc or AR, so I expect that everything is fine here.

If it was so demanding it would be the same downscaling 4k to 1080p, but with the 4k samples i have i can use 2d ssim downscaling absolutely fine (GTX950), well below maximum render times (and its only in the osd once). This issue only happens with supersampling.

CarlosCaco · 9th March 2016, 02:23

Quote:

Originally Posted by madshi

What does the "present a frame for every vsync" option have to do with this?

i was just saying that this option increase a little my render times, without this option checked i having some glitches...

ryrynz · 9th March 2016, 02:30

Quote:

Originally Posted by iSeries

If it was so demanding it would be the same downscaling 4k to 1080p, but with the 4k samples i have i can use 2d ssim downscaling absolutely fine (GTX950), well below maximum render times (and its only in the osd once). This issue only happens with supersampling.

No it wouldn't, you're not factoring in image doubling.

madshi, this will probably be addressed regarding the OSD, but Jinc AR shows up in the OSD when supersampling and downscaling using Jinc with no AR. It shows Image doubler = Jinc AR < Jinc.

iSeries · 9th March 2016, 02:36

Quote:

Originally Posted by ryrynz

No it wouldn't, you're not factoring in image doubling.

madshi, this will probably be addressed regarding the OSD, but Jinc AR shows up in the OSD when supersampling and downscaling using Jinc with no AR. It shows Image doubler = Jinc AR < Jinc.

When supersampling, doubling 1080p and downscaling back to 1080p with 1d, using super-xbr to double, and using superres 3 and crispen edges to enhance, i get around 35ms render times. Downscaling with 2d it shoots up into the 80s. That can't be normal. Like i said, all other doubling scenarios 2d only adds around 2-3ms to what 1d costs, and same goes for downscaling 4k native to 1080p, no trouble at all using 2d, again only adds 2-3ms to what 1d costs.

ryrynz · 9th March 2016, 03:13

Quote:

Originally Posted by iSeries

When supersampling, doubling 1080p and downscaling back to 1080p with 1d, using super-xbr to double, and using superres 3 and crispen edges to enhance, i get around 35ms render times. Downscaling with 2d it shoots up into the 80s. That can't be normal. Like i said, all other doubling scenarios 2d only adds around 2-3ms to what 1d costs, and same goes for downscaling 4k native to 1080p, no trouble at all using 2d, again only adds 2-3ms to what 1d costs.

Only 2ms difference between 1D and 2D? I think something's broken, you should maybe look into resetting your settings as my GTX 960 has about 8ms render times when downscaling 4K content and 16-18ms when using 2D (with AR and LL)

tp4tissue · 9th March 2016, 07:11

Hey guys.. so , with Madvr.. If I want to use it, I have to stick with the 32-bit.

What do I use to decode 10bit hevc 4K to then connect with Madvr, because my downstairs computer is not fast enough using the 32bit Lav filter to CPU decode that.

I'm a Madvr diehard.. I just thought this would be the place to post this, since you guys are also sticking with 32bit for madvr..

huhn · 9th March 2016, 07:19

madVR has a 64 bit version for a very long time so no a lot of us are using 64 bit.

tp4tissue · 9th March 2016, 08:09

Quote:

Originally Posted by huhn

madVR has a 64 bit version for a very long time so no a lot of us are using 64 bit.

what.. what............................ wait, is it the same download in the OP ?

ryrynz · 9th March 2016, 08:37

Quote:

Originally Posted by tp4tissue

what.. what............................ wait, is it the same download in the OP ?

The zip contains both the 32 bit version and the 64 bit version. Make sure to run install.bat after extraction.

9th March 2016, 07:11	#36757 \| Link
tp4tissue Registered User Join Date: May 2013 Posts: 712	Hey guys.. so , with Madvr.. If I want to use it, I have to stick with the 32-bit. What do I use to decode 10bit hevc 4K to then connect with Madvr, because my downstairs computer is not fast enough using the 32bit Lav filter to CPU decode that. I'm a Madvr diehard.. I just thought this would be the place to post this, since you guys are also sticking with 32bit for madvr.. __________________ Ghetto \| 2500k 5Ghz

9th March 2016, 01:06	#36744 \| Link
iSeries Registered User Join Date: Jan 2009 Posts: 625	Hi, using 2d ssim causes a lot of dropped frames for me. In the stats its listed twice (like this 'ssim 2d100 ar < ssim 2d100 ar') and render times go way up. 1d ssim works fine and is listed in the stats only once as expected. Is that normal for 2d?

9th March 2016, 01:18	#36747 \| Link
huhn Registered User Join Date: Oct 2012 Posts: 7,925	take an image host please

9th March 2016, 07:19	#36758 \| Link
huhn Registered User Join Date: Oct 2012 Posts: 7,925	madVR has a 64 bit version for a very long time so no a lot of us are using 64 bit.