[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vr / w / wg] [i / ic] [r9k] [s4s] [cm / hm / lgbt / y] [3 / aco / adv / an / asp / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / qst / sci / soc / sp / tg / toy / trv / tv / vp / wsg / wsr / x] [Settings] [Home]
Board
Settings Home
/a/


File: nvidia fire.jpg (66.33 KB, 800x600)
66.33 KB
66.33 KB JPG
>madVR v0.87.4 released

>* workaround added: NNEDI3 upscaling failed/froze with newer NVidia GPUs
>* fixed: NNEDI3 chroma upscaling produced wrong colors with 10bit sources

>All OpenCL features (except DXVA NV12 processing) should now also work with newer NVidia GPUs. However, the current official NVidia drivers are broken and always produce zero/black output when used with D3D9 interop inside of madVR. I've tried to make it work somehow, but failed. You have to downgrade to NVidia 327.23 drivers (or older) to make the OpenCL features work with NVidia. Would be nice if someone could report the OpenCL driver bug with the current official drivers to NVidia.

http://forum.doom9.org/showthread.php?p=1664667#post1664667

Time to downgrade if you want to enjoy NNEDI3 upscaling.

Also,

>720p -> 1440p image doubling
>64 neurons
>geforce 670gtx
>no OpenCL dither
83% GPU load
>>
Only seems to render the red channel for me when NEEDI image doubling is used, with some mildly creepy results.

>You have to downgrade to NVidia 327.23 drivers (or older) to make the OpenCL features work with NVidia
Well, that explains it. Not sure if I'm gonna bother downgrading. Jinc works great, and I mostly watch 720p on a 1080p screen so it's not like there's gonna be a huge difference.
>>
>>101038813
Woah
>>
File: 1390854317644.png (188.14 KB, 1320x828)
188.14 KB
188.14 KB PNG
>>101038813
Jesus.
>>
>>101038813
>>101039326
oh god that's horrifying
>>
File: 1272233604631.jpg (53.10 KB, 338x263)
53.10 KB
53.10 KB JPG
>>101038393
I never thought I would get to post these again.
>>
I need better testing material

http://www.screenshotcomparison.com/comparison/59669
>>
File: 1272233680286.jpg (59.61 KB, 639x480)
59.61 KB
59.61 KB JPG
>>
daiz inc in 5...4...3...
>>
what is nnedi3
google just returns doom9 bug reports and unrelated blog posts as top results
>>
>>101039863
neural net edge-directed interpolator. pretty much the gold standard for powers of 2 video upscaling. until recently it was so slow that it was only useable as offline upscaler, i.e. as part of some encoding chain in avisynth.

With the opencl implementation it's now viable for use in madvr if you got a good gfx card
>>
>>101039992
So on non-2^n scales it does what, scales up then downsamples with lanczos?
>>
>>101040076
Yes, or whatever algorithm you choose in the settings.
>>
File: nvidia gtx 480.jpg (237.17 KB, 1162x778)
237.17 KB
237.17 KB JPG
>>
>>101040076
depends on the source/target resolution and settings , it can do nnedi 2x/4x up + conventional downscale or nnedi 2x/4x up + conventional upscale.
>>
>>101040231
or only NNEDI3 if it's exactly 2x/4x.
>>
File: evans.jpg (52 KB, 644x478)
52 KB
52 KB JPG
be cool about fire safety
>>
File: military optometry.jpg (527.98 KB, 1513x1225)
527.98 KB
527.98 KB JPG
http://www.screenshotcomparison.com/comparison/59671

comparison between bicubic 75, no AR [something you would get with EVR-CP] and NNEDI3 64 neurons
>>
http://www.screenshotcomparison.com/comparison/59675
>>
>>101040464
Yeah, it's probably more noticeable at sharper sources than anime screenshots.

But it does look noticeably better in >>101040954
>>
>>101040954
Wow it's fucking nothing.
There's a small improvement, but is it really worth it?
>>
>>101040464

>literally no difference apart from the 1 pixel displacement you get from highlighting it
>>
>expecting people to downgrade their video driver because dumbasses broke the computer code

yeah, okay, GIT RITE ON DAT
but I don't think it's an issue for me in either case since I have an older vidya card.
>>
>>101041137
I'd need something with lots of fine lines to have a good comparison shot. E.g. something with hair in HD
>>
>>101040464
>>101040954
It's somewhat sharper. Do you have any higher res comparisons?
>>
Since november I had been in what you could call a downward spiral of bad anime taste. I think it began after watching Paranoia Agent.
I stopped following recommendations and started to watch anime only because it was "fun to watch". I started with rewatching random SnK episodes for about a week.
Then I watched bakemonogatari, nisemonogatari, oreimo, nekomonogatari, oreimo SII. I'm not saying they're bad but I wasn't watching them for the story or deepness,
the only thing I wanted was for them to be pleasant to watch. I think the only relly great thing I saw was EoE on the 15th dec. During this period I was also more active on /a/ than usual. I lurked for at least 3-4h/day. So, one day I saw a thread where the op said that the anime in pic related changed his taste in anime and he now couldn't watch stupid shonen and sucha anymore. I got almost scared and decided that I would never watch it. So in the days before Christmas Eve I watched Outbreak CompanyI chose it because it looked like a fun anime with a very original plot and then Valvrave, which I finished during the night between 2013 and 2014. I'm not saying it was bad, but god it was soo painful to watch. The characters changed from likable to unlikable EVERY 5 MINUTES. Almost like a girl on her period. The only two consistent and decent characters were L-Elf and blonde guy. He was also the only semi-competent member of the "government".
(Continued in next post)
>>
>>101041303
Well, nvidia is and has been notoriously shit when it comes to OpenCL, they want to force people to use CUVID. So it's not improbable that it's nvidia who's doing things wrong here.
>>
>>101041394
see >>101039566 for an SD upscale
>>
http://check2pic.ru/compare/34577/
http://check2pic.ru/compare/34576/
The ones from the doom9 thread.
>>
Now that most of the regressions seems to be fixed, can someone with doom9 account ask him to add a filename string value in the profile scripting thingy? It'd be nice to be able to have a "HorribleSubs" filter, for example. I guess he also needs to make a "contains" keyword for it to be useful, but it shouldn't be hard to do.
>>
>>101038813
>redruM
>>
when will madvr be able to handle 60fps videos
>>
>>101045195
should already be possible with a current-gen gfx card, depending on which algorithms you choose.
>>
>>101045580
and which algorithms are those?
>>
New drivers are total shit and caused me tons of big problems. I live on 314.22 just fine.
>>
>>101045708
lancoz 3tap for example. only 16% GPU load at 24fps for me, so 60fps should be easy.
>>
Saw a post about this yesterday but there didn't really seem to be a solid reason for this.

While running 8 bit video Playing [DXVA] appears on mpc-hc, In Lav Video it says Active decoder = CUVID. Which is right? should I switch to DXVA2 (native)
>>
>>101046663
I don't know shit about DXVA but people here seem to shit on it, fiddle with the options until it goes away.
>>
>>101047146
DXVA is for laptops or otherwise performance-constrained systems that can handle software-decoding.
>>
so if i have a newish radeon am i cool?
>>
>>101047278
yes
>>
>>101046500
i'm getting horrible stuttering on my hd5850, and the GPU usage never goes past 40%
>>
>>101047685
Try Fullscreen exclusive mode on
>>
My pitiful laptop can't even run madVR without videos shitting themselves.

CCCP is the best I can do.
>>
>>101047875
If you've got a laptop res of 1336 x 768 then madVR wouldn't even be that noticeable anyway. Minus glorious smooth motion of course.
>>
>>101047786
that helped a little, now i get whooping 10fps
GPU usage under 20%
>>
>>101048030
Software or hardware decoding? Does it work fine with EVR-CP?
>>
>>101048073
hardware decoding of course, video works perfectly fine on software
>EVR-CP
whats that?
>>
>>101048129
> video works perfectly fine on software
If it does, why don't you just use software decoding whenever you play 60 fps videos then? And if it can be played perfectly fine with software decoding + madVR, why are you saying madVR can't handle it? I'm honestly confused.
>whats that?
Another video renderer.
>>
>>101048353
i thought that mad VR only works with hardware decoding
>>
>>101048427
Uh, no.
>>
File: heatenings.jpg (25.16 KB, 638x693)
25.16 KB
25.16 KB JPG
>Newly purchased nvidia apu
>>
File: Untitled.jpg (182.27 KB, 1298x845)
182.27 KB
182.27 KB JPG
>>101045195
60fps videos work fine.
>>
>>101048679
PC specs?
also i'm playing 1920*1080@60 downscaled to 1280*800
>>
File: Speccy.jpg (108.37 KB, 868x532)
108.37 KB
108.37 KB JPG
>>101048953
>>
>>101049007
guess it's a radeon problem then
>>
>>
File: 1358098099698.gif (1.19 MB, 690x200)
1.19 MB
1.19 MB GIF
I still laugh at Nvidia fire jokes and PS3 nogames jokes.
>>
File: speccy.png (59.45 KB, 674x532)
59.45 KB
59.45 KB PNG
>>101049007
That reminded me of the weird temperature display Speccy always shows for my motherboard, it randomly jumps from like -30 to over 80 degrees within seconds of eachother. Must be some kind of bug.
>>
>>101049781
They never get old
>>
>>101049982
>>
Now AMD will be the ones causing house fires with vidya cards operating at 90c. A few more degrees and you could use your rig to fry an egg.
>>
>>101050029
>>
File: nvidia brand cereal.gif (891.54 KB, 240x192)
891.54 KB
891.54 KB GIF
>>101050104
>>
File: 1368573698193.png (2.01 MB, 920x1270)
2.01 MB
2.01 MB PNG
>>101050107
>>
>>101048030
post a screenshot of the ctrl+j info thing
>>
>>101049897
its an ASUS z68 problem, got that too
>>
>>101050104
>A few more degrees
protein already coagulates at 40-60°C (depending on the proteins). that's why you can fry eggs on black surfaces during summer
>>
File: manly tears.jpg (40.19 KB, 447x599)
40.19 KB
40.19 KB JPG
>>101049897
>>101050496
Your motherboard is shielding you from the wrath of your NVIDIA cards
>>
How the hell is this /a/ related?
>>
>>101050576
madVR is what everyone should be using with their media players to watch their animu.
>>
How does NNEDI3 compare to Jinc for chroma scaling? I assume that's where it will be used more because of the whole powers-of-2 thing.
>>
File: 1317934438498.png (18.47 KB, 418x303)
18.47 KB
18.47 KB PNG
>>101050576
>streaming
gtfo

>horriblesubs
confirmed for ADD

>10bit encode
acceptable

>madVR
how it is intended to be viewed
>>
I'm surprised Daiz isn't in here yet
>>
There's nothing wrong with the Nvidia drivers, madshi can't code for shit
>>
>>101050761
chroma is not as visually important, and there are two steps involving chroma:

a) bringing 4:2:0 chroma up to 4:4:4 [chroma upscale settings]
b) bringing the whole 4:4:4 picture up to target resolution (NNEDI can be used for luma here, jinc&co for chroma-only) [doubling & regular upscale/downscale settings]
>>
File: 1386648737556.png (247.23 KB, 760x572)
247.23 KB
247.23 KB PNG
>>101050902
>>
>Apple may be trying to not-so-delicately nudge everyone to move their code from CUDA to OpenCL, but I’ve seen a first-hand failure of AMD’s OpenCL support with V-Ray RT for Maya. Chaos Group built V-Ray RT on OpenCL, but after extensive work trying to get the GPU variant of its RT render engine running on AMD hardware and an effort by yours truly to light a fire under Apple and AMD, Chaos Group gave up and ported it to CUDA instead. So V-Ray RT’s GPU mode only works with OpenCL and CUDA—on Nvidia hardware.

AMD literally with no drivers
>>
>>101050947
>Being this buttmad AMDFAG

All OpenCL software works fine, only madshi's buggy piece of junk doesn't work, go figure
>>
>>101051005
I'm >>101049007
>>
>>101050761
http://screenshotcomparison.com/comparison/59697
>>
>Drops frames with it enabled

Well bugger.
Guess that's what happens when you cheap out on hardware.
>>
What does - use a separate device for presentation actually do?
>>
>>101051921
I don't know but some people get performance increases with it on/off.
>>
>>101051393
If you know, what was the original resolution of that? Thanks.
>>
>>101052113
Of the video file? 720p, only chroma is scaled there.
>>
NNEDI3 is a lot more noticeable on large resizes. Like 240p->960p. It would be an amazing emulator filter if someone would port it to a glsl shader.

Here is an example scaling a SNES screenshot: http://screenshotcomparison.com/comparison/59698
>>
>>101052796
I noticed that too. Doesn't make much of any difference (to my eyes at least) unless there's some serious upscaling going on. Which I guess makes sense, given what it's doing.
>>
>>101052796
>http://screenshotcomparison.com/comparison/59698
NNEDI3 is pretty amazing.
>>
>>101052796
is that jinc3 with or without AR? because AR has a special mode for computer-generated gfx.
>>
>>101050872
>I'm surprised Daiz isn't in here yet

Daiz is firmly convinced that madVR scaling does not offer an appreciable benefit over vanilla bicubic.
>>
>>101041600
>http://check2pic.ru/compare/34576/

Looking kinda QTEC there, nnedi3.

Disgusting. I'll keep muh Jinc.
>>
>>101054541
When you're scaling dimensions up to between 100% and 200%, (eg 480p to 720p or 720p 1080p) to you might as well use gaussian interpolation, because you're losing the top half of frequencies no matter what anyway.
When you're scaling down, this kind of interpolation isn't useful anyway.
>>
>>101054795
Have you ever tried upscaling something yourself? Or even just tried comparing upscaled material?
>>
>>101053660
It is with anti-ringing enabled. I don't see an option for any special computer-generated gfx mode in the madvr settings.

Here is Jinc3 without anti-ringing on: http://screenshotcomparison.com/comparison/59702

The checkerboard patterns in the background still kinda bleed together, but that just seems to be the way Jinc works. It doesn't happen with other filters like Spline or Lanczos, but those have way more aliasing and ringing.
>>
>>101054967
Yeah, I do it all the time as an OCD thing.
>>
>>101055021
The AR filter is a scary thing.
>>
>>101055021
>The checkerboard patterns in the background still kinda bleed together, but that just seems to be the way Jinc works.
Because it's removing frequencies at and above the image's supposed nyquist frequency.
>>
>>101055021
>I don't see an option for any special computer-generated gfx mo
It's supposed to be automatic when it detects clean images or something. Just something i read in those thousand of pages of the thread
>>
File: 1363312238033.gif (535.36 KB, 850x478)
535.36 KB
535.36 KB GIF
>mfw 95% GPU usage with OpenCL and NNEDI3 at 32 neurons on a GTX 760
>>
>>101053270
>>101055021
>not point-resizing your emuladores
>any year
>>
>>101056082
>not using an anti-aliased nearest neighbor filter
>>
>>101056328

You're disgusting.
>>
Is a 470GTX enough for this?
>>
File: dfgjfghj.png (178.30 KB, 2094x785)
178.30 KB
178.30 KB PNG
>>101056385
>>
>>101056386
sufficient for some of the upscalers. but i don't think there's a single card in existence that could handle everything on max.
>>
>>101056501
>no integer scale factor

You're FUCKING disgusting.
>>
The OpenCL option has a solid 25% increase in usage on my GPU when turned off.
>>
>>101056820
>integer scaling sources with non-square PAR
YOU'RE disgusting.
>>
>>101056877
>not emulatorating on a CRT

Nay thee.
>>
Friendly reminder from /sci/ that upscaling cannot introduce new information, and that the only source of more information are from better encodes and higher souce resolution.

Enjoy your placebo.
>>
>>101057470
the difference is how the intermediate pixels are calculated. NNEDI basically invents new pixels by guessing at edges.

It gets a better visually appealing result out of the data that's there.
>>
>>101057470
Friendly reminder from /commonsense/ that inferior upscaling algorithms will actually destroy information and introduce unwanted artifacts.
>>
Someone post a debanding profile I can use.
>>
>>101038393
>660ti
>try out NNEDI
dropping every 5th frame
>>
>>101056790
You forgot about 780's in SLI
>>
>>101058546
I'm not sure whether madvr can utilize SLI
>>
are there any comparisons between different numbers of neurons?
>>
OK, I get OpenCL not working but why isn't DXVA2 working with the update?
>>
Well, at least with the latest revision madVR has stopped rendering nothing but blackness for me.
>>
>>101059549
if you got an intel or amd try general settings -> use open cell to process dxva surfaces
>>
>>101059754
>my brain in charge of converting thoughts into typing
I meant
>try general settings -> use OpenCL to process dxva surfaces
>>
>>101059922
Still nothing. No video playback when resizing using DXVA2.

You'd think that if OpenCL isn't working with the latest Nvidia drivers then disabling that option would remedy the issue. Doesn't seem to be the case.
>>
>>101060078
oh, resizing with dxva. i think madshi mentioned an issue with that. few people use that option though



Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.