madVR - high quality video renderer (GPU assisted)

[dansrfe]

Originally Posted by mrkazador

Where did you get .3204 drivers? I'm using an Intel i3 NUC and just installed 9.18.10.3165 and the seekbar is not disappearing for me and can't find the drivers you listed.

What port does madVR listen on over LAN? Also, is there some documentation on how to use it?

[kasper93]

@dansrfe: See net-protocol.txt

[raul31]

Originally Posted by vivan

raul31,
rename "mpc-hc.exe" to something else (e.g. "game.exe"). And you have to force mpc-hc to use dGPU, not Madhcctrl.

Originally Posted by madshi

See vivan's reply.

thanks Vivian & mashi! (my apologies as well for late reply) who would have thought that itz that easy... cool!
when i play a film/movie, MPC-HC is now displayed in the nvidia GPU activity in the notification area, which can only mean that madVR is utilizing it.

i have a lot more questions for u madshi (& other nvidia videophiles) as i'm trying to get the correct set-up with every detail. i hope u can guide me in getting the correct settings with everything.

now that i have "high performance nvidia processor" selected as the preferred graphics processor for MPC-HC LITE (i'm now using the LITE version... just trying it out) ,

what exactly do i choose for the next sequential section "specify the settings for this program:" ?

these are the features that are listed:

ambient occlusion
anisotropic filtering
antialiasing - FXAA
antialiasing - gamma correction
antialiasing - mode
antialiasing - setting
antialiasing - transparency
CUDA - GPUs
maximum pre-rendered frames
multi-display/mixed-GPU acceleration
power management mode
texture filtering - anisotropic sample optimization
texture filtering - negative LOD bias
texture filtering - quality
texture filtering - trilinear optimization
threaded optimization
triple buffering
vertical sync

now what setting should be applied to each one so that madVR gains the best quality from the nvidia GPU?

[madshi (& other nvidia videophiles), i don't expect u to retype all the features when answering me, just copy & paste all the features that i typed out & then specify afterwards what setting is the best for each one]

thanks a million !!! i really appreciate it!

[Andy o]

madshi, have you read up on the new Haswell GT3 (Iris 5100) and GT3e (Iris Pro 5200) chips and have an idea to how they would fare with madVR's more GPU-intensive tasks? The difference between those two is that GT3e has 128MB on-die RAM. Anandtech called it effectively an L4 cache. I'm feeling this year is finally new laptop year, and I'm eyeing out the Asus Zenbook Infinity which will apparently come with a dual-core GT3 part (i7-4558U), and I'm wondering how much difference the lack of this on-die memory will make on madVR specifically, since I wouldn't buy a PC that doesn't run madVR a bit more than reasonably well! Also, this ultrabook has a 2560x1440 screen

[bugmen0t]

I will try watching movies with madVR OSD on but I can already assure you that there is no dropped frame or presentation glitch when there is a black frame since I checked that several times after it occured (since I thought in the past that it's one of the former). So there only remains the possiblity with the empty queues.

[ryrynz]

Originally Posted by mrkazador

Where did you get .3204 drivers?

http://asder00.blogspot.se/

Actually, I take it back, it's still doing it on certain content.

[sucht]

Originally Posted by DarkSpace

I'm fairly sure he did mean XySubFilter, as that it the version that sends the subpictures to madVR directly instead of blending them onto the video.
Unfortunately, as far as I'm aware, XySubFilter is still work in progress and a public beta for testing purposes is not yet released, so the only way to get it is to compile it yourself from the source.

sorry but where can i get the XySubFilter sources? (i'd like to play around a bit)
all i can find is xy-VSFilter :/

[kasper93]

@sucht: See this post http://forum.doom9.org/showpost.php?p=1631842&postcount=19039

[pie1394]

Originally Posted by madshi

That doesn't look so very much different from the defaults? FWIW, I would recommend to use a CPU queue slightly higher than the GPU queue. E.g. try a CPU queue of 24 with everything else the same.

The strange random frame dropping on HD7970 is somewhat annoying. Before using that setting, I did try several combinations and couldn't prevent the frame dropping when the GPU frequency is automatically changed from 300MHz to 950 MHz during the playback. Increasing the GPU queue size to 16 somehow does help on this particular system...

This issue is never noticed on GTX260+ and ION chipset. I feel that AMD driver's GPU frequency changing logic does not seem to run as smooth as nVidia driver's...

With the latest Catalyst 13.6 beta2 driver, it seems to work better than older drivers on some very high bit-rate samples which could stress the GPU at particular scenes.

[Vyral]

Originally Posted by madshi

The 6850 is quite fast, but even the 6850 should not be able to do Jinc3 AR chroma upsampling with 0% GPU. So I would say there's probably something wrong somewhere. Does the madVR OSD report "NV12" (YCbCr 4:2:0)?

Here, a pic with madVR OSD and GPU-Z :
http://img199.imageshack.us/img199/7749/ca0.png
Do we have to uninstall and reinstall madVR for each new version or just uncompressed the zip file on the folder where we install madVR ?
I always use the second method and maybe something happened between two versions.

[madshi]

Originally Posted by raul31

these are the features that are listed:

ambient occlusion
anisotropic filtering
antialiasing - FXAA
antialiasing - gamma correction
antialiasing - mode
antialiasing - setting
antialiasing - transparency
CUDA - GPUs
maximum pre-rendered frames
multi-display/mixed-GPU acceleration
power management mode
texture filtering - anisotropic sample optimization
texture filtering - negative LOD bias
texture filtering - quality
texture filtering - trilinear optimization
threaded optimization
triple buffering
vertical sync

now what setting should be applied to each one so that madVR gains the best quality from the nvidia GPU?

Leave all to default. Things like anisotropic filtering and anti-aliasing should be set to "let application decide" or "application preference" or however that option is called.

Originally Posted by Andy o

madshi, have you read up on the new Haswell GT3 (Iris 5100) and GT3e (Iris Pro 5200) chips and have an idea to how they would fare with madVR's more GPU-intensive tasks?

Read up: yes. Idea how they would fare: no. Guess we need to wait for tests.

[madshi]

Originally Posted by Vyral

Here, a pic with madVR OSD and GPU-Z :
http://img199.imageshack.us/img199/7749/ca0.png
Do we have to uninstall and reinstall madVR for each new version or just uncompressed the zip file on the folder where we install madVR ?
I always use the second method and maybe something happened between two versions.

Uninstall/reinstall is usually not needed. I don't know why GPU-Z shows a 0% GPU usage. Looks like a bug in either GPU-Z or in the NVidia driver to me.

[madshi]

madVR v0.86.5 released

http://madshi.net/madVR.zip

* fixed: calibration test pattern colors were sometimes incorrect
* added support for internal VideoLUT processing
* added remote control ability for internal VideoLUT processing

Nothing important for "normal users". Just a fix and some improvements for developers of calibration software...

[QBhd]

Originally Posted by madshi

Uninstall/reinstall is usually not needed. I don't know why GPU-Z shows a 0% GPU usage. Looks like a bug in either GPU-Z or in the NVidia driver to me.

It is the same here with a 5770... it always been that way, it will show usuage when interlaced content is played, but not 720p

I have always just chalked it up to a little quirk and nothing to worry about. I know the GPU is being used since memory usage increases as well as temps

QB

[karamancho]

Originally Posted by vivan

raul31,And you have to force mpc-hc to use dGPU, not Madhcctrl.

how do you do that?

[dansrfe]

Backbuffer queue keeps cycling between 0/8 and 0-7/8 in windowed mode with 0.86.4/5. I've tried pausing/playing, going back and forth in the timeline, and resizing above and below 1080p (original video resolution).

[leeperry]

Having more thoughts about 3DLUT gamut mapping, would that be possible to have the "primaries" tag also activated via a dummy PS script by any chance please? Something like "// primaries=709|SmpteC|EBU|sRGB|NTSC|PAL|470M|240M|170M"?

It seems rather clear that the PS script gamut mapping days are counted and it would still be highly appreciated to have the ability to override mVR's automatic logic, because IME +75% of 24p HD movies are still SMPTE-C(they look terribly greenish in REC.709) and I like anything 25fps to be treated as EBU, even HD(following the ISF recommendations).

I know it can be rolled with a hotkey but I only have two unused buttons on my mouse, one is currently used for matrix rolling and the other for TV/PC levels.

It would also be useful for ppl with funky encodes or gamuts that aren't properly detected automatically, one automatic profile in their media player et voilà. I realize that feature requests are a no-no but this would finally nail down the whole 3DLUT management matter, hope you'll consider it

Originally Posted by madshi

The key reason for this release is the added support for remote controlled calibration test patterns, in the hope that ArgyllCMS + HCFR will make use of that.

Oh wow, another dream coming true: the ability to run a test proof verification pass without the major hassle of rolling manual test patterns. If HCFR supported this, we could end up with a lot of dead-flat curves to drool at

Originally Posted by madshi

Jinc3 (without AR) needs to read 36 input pixels to produce 1 output pixel. For each of those 36 input pixels Jinc also needs to access a Jinc 1dlut. Add to that a lot of shader math.

Ahhh, no wonder things quickly get out of hand once you reach >540p source files. That would also explain why J3AR looks so impressive, OTOH sometimes I prefer bicubic on noisy/low-res files in order to act as a poor man's noise reduction.

BTW, is there any picture viewer that would support Jinc3? I guess AR won't be possible as you seem to have come up with this feature on your own, but ACDSee's resizing looks really ugly compared to mVR

[zoyd]

Test version of HCFR with auto-generated madVR patterns is available here.

[andybkma]

Anything us Zoom Player users can do to help with Bug 30? My ZP constantly gets stuck on top as per the bug

[pie1394]

Originally Posted by madshi

Read up: yes. Idea how they would fare: no. Guess we need to wait for tests.

Does the madVR code require a lot of "random" memory read access for each pixel group's processing? If not, I guess the performance is still stream-processor-count limited.

If I don't remember it wrong, nVidia's Kepler-based and AMD GCN-based (like Tahiti, Pitcairn, Cape Verde) 28nm GPUs have somewhat HUGE on-die SRAM cache than any previous generation in order to reduce the probability of external memory access. The on-die cache accessing bandwidth is at least TB scale, not just 100 ~ 200GB scale. That is why nVidia can cut GTX660 ext-mem to 192-bits from GTX560Ti's 256-bits, but still with better performance on even heavily access at big multi-texture memory fetching.

The GT3e 128MB wide-ported EDRAM definitely helps for the performance to reduce the accessing chance with main memory -- shared with CPU / peripherals. Its design is logically identical to a stand-alone GPU...

In Computex 2013, some vendors have demostrated the dual 4K output with Haswell. Although the performance is below the requirement of video playback, it still indicates Haswell capability -- up to 2 * 4K 30/60 Hz display via DP ?