madVR - high quality video renderer (GPU assisted)

[Swede]

Sadly this thread, with it's +66.000 posts, is killing the server so I have to delete it and just keep an archived version at: https://forum.doom9.org/archive-perm/index.php/t-146228.html

[Framerate]

Once you calibrate your display for SDR, are there any test patterns or calculations to help you determine your madvr HDR settings? Is everyone mostly doing trial and error with the display peak luminance (nits) setting and dynamic target nits setting, or if you can measure your display's brightness, contrast, and color gamut, is there a more scientific way to input the correct numbers to get the most accurate HDR image?

Also what MadVR settings are most powerful in terms of rendering, sharpening, enhancing the image, undoing compression artifacts, etc? If I want to max the overall PQ, but still keep the image accurate, what should I use? My PC has a 4070 Super GPU, a 12700K CPU and 32 GB DDR5 ram.

Also if I want to use Reclock, do I have to use MPC-HC 32 bit version, and if so, can it handle all the same madvr settings as the 64-bit (NGU antialiasing etc?)? I want to try Reclock out with MPC-HC 32 bit, but how do I it without messing up my MPC-HC 64 bit settings? Among other issues, my "profile auto select rules" code that google AI wrote for me may be incorrect because my 32-bit Profile is in bold letters even when I am using the 64 bit player. The code is "if (bitDepth == 64) "64-Bit High Quality"
else "32-Bit ReClock Low Overhead"

I tried putting an mpc-hc.ini document in both the 64-bit folder and the 32-bit folder, so that each mpc-hc version would keep its settings separate, however I did not know that you had to allow that in the general settings, so changing a setting on one is still changing it on the other in some cases. Do I have to do a clean uninstall and reinstall of either/both MPC-HC versions, or is that not necessary? There was some warning when I first tried this that it detected a .ini file and said therefore none of my settings would be saved at all, but Google AI gave me some advice to change something and then I did not get that warning anymore. Maybe it was permissions. However I'm wondering if I messed something up. I don't remember exactly what Google AI told me to do and the whole Google AI conversation is now giving me an error and not loading. A new convo is also telling me to change the .ini file in MPC-HC 64 bit folder to "mpc-hc64.ini" instead and also, quote, to "Run as Administrator Once to Activate... Because these folders are protected by Windows, you must manually give the player permission to write to these files the first time: Go to your 64-bit MPC-HC folder, right-click mpc-hc64.exe, and select Run as administrator.Open View > Options > Player. Check the box that says Store settings to .ini file. Click Apply and close the player."

Thanks for the help

[huhn]

Once you calibrate your display for SDR, are there any test patterns or calculations to help you determine your madvr HDR settings? Is everyone mostly doing trial and error with the display peak luminance (nits) setting and dynamic target nits setting, or if you can measure your display's brightness, contrast, and color gamut, is there a more scientific way to input the correct numbers to get the most accurate HDR image?
you use the measured brightness
Also what MadVR settings are most powerful in terms of rendering, sharpening, enhancing the image, undoing compression artifacts, etc? If I want to max the overall PQ, but still keep the image accurate, what should I use? My PC has a 4070 Super GPU, a 12700K CPU and 32 GB DDR5 ram. for an accurate image >none<.

Also if I want to use Reclock, do I have to use MPC-HC 32 bit version, and if so, can it handle all the same madvr settings as the 64-bit (NGU antialiasing etc?)? I want to try Reclock out with MPC-HC 32 bit, but how do I it without messing up my MPC-HC 64 bit settings? Among other issues, my "profile auto select rules" code that google AI wrote for me may be incorrect because my 32-bit Profile is in bold letters even when I am using the 64 bit player. The code is "if (bitDepth == 64) "64-Bit High Quality"
else "32-Bit ReClock Low Overhead"
the GPU doesn't care it does neither run in 32 bit cpu code or 64 bit cpu code it uses it own code. i'm pretty sure the bitdepth is not about what type of x86 is used... a 32 bit program can use 64 bit precision just as a 64 bit program it is just slower. cause videos are not 64 bit if ever it will fall back to the else. but that's what you get for using an AI. an AI that struggles with player executable and video "bitdepth"

in 32 bit you have very limited ram available and most codecs are plain slower when CPU decoded but that's about it. i do not know if cpu -> gpu up and download is affected.

[Framerate]

Help plz!!

I am experiencing a number of pecularities with madVR. Is there a bug with the control J information on the latest beta?

First, I tried a test where I played a 23.976 FPS source file outputting at 120 Hz. The frame drop prediction counter said "one frame drop every 41 seconds" or so, however, I left the movie running for 3-4 minutes and did not get a further dropped frame.


Second, I am trying to achieve more accurate framerate fractions. However, when I go into Custom Resolution Utility (CRU) and lower the frame rate by five one-hundredths, madVR still spits out exactly the same numbers, as if it does not recognize my new custom resolution at all.

Even before this, when I tried 47.952 Hz and 95.904 Hz custom resolutions, the system exhibited strange behavior. At first, it seemed to round close to an even number. Then, when I enabled Variable Refresh Rate (VRR) on my display, I got a more exact number. After that, I tried turning VRR off on my display while madVR was still running. I got a black screen and had to force a shutdown of my computer. However, afterdoing that, I am now getting more exact values even with VRR turned off, as if this process somehow modified what madVR or the display was doing.

But now, when I try to adjust them further, madVR changes absolutely nothing, and the custom resolutions do not impact anything—even when messing with frame rates that are native to the PC and display, like 23.976 Hz. I also tried keeping the frame rate at 23.976 Hz but changing the porch times, but again, it changed absolutely nothing in madVR's numbers. How can I actually get these tweaks to work?

Additionally, I am trying to output 23.976 FPS content at exactly five times the frame rate, which should be 119.880 Hz, but madVR is outputting at 119.885 Hz instead. Mathematically, this should cause a full frame drop or repeat every 3-4 minutes. However, what actually happens on the madVR OSD (Ctrl+J) screen at the start of playback is this: the clock deviation starts in the positive range, and I might get a prediction of one frame drop every X seconds or minutes. But then, as the movie plays, the frame drop prediction quickly gets longer and longer. As this is happening, the clock deviation trends toward zero and begins moving into the negative range. As it goes further into the negative range, madVR predicts frame drops less frequently not more frequently. After leaving the movie running for half an hour, it settles into a frame drop prediction of one frame drop every 5 hours or so but may keep going I have not tested that.

I am not 100% sure, but I believe this was originally happening when outputting 23.976 FPS at 23.976 Hz as well. I know when I tested madVR a year ago on Windows, the prediction for 23.976 FPS at 23.976 Hz would very quickly settle at one frame drop every hour and a half or so. But this time on Windows 11, I think it was always shifting and growing, until about a week ago when it suddenly changed. Now, for 23.976 FPS at 23.976 Hz, it quickly settles at 1 frame drop every 1 hour and 5 minutes or so. However, for 23.976 FPS at 119.88 Hz, it still shifts over time and does so in a way that seems to make no logical sense. What is going on here?

Lastly, when I am outputting 23.976 FPS content at 119.88 Hz, what does the frame drop/repeat glitch prediction counter actually mean? Does it mean I will get one full 23.976 FPS source frame drop every 5 hours, or does it mean I will get one 8.3-millisecond VSync drop every 5 hours? In other words, does it mean that one of the 119.88 Hz display refreshes will drop every 5 hours, or does it mean that one of the actual 23.976 FPS source frames will drop every 5 hours? And if it is the latter, will it wait to drop that frame all at once after 5 hours, or does it mean that it will drop an 8.3-millisecond VSync once per hour because 119.88 Hz is exactly five times 23.976 Hz and the prediction is one every 5 hours?

[QBhd]

Note that the temp of your room will also affect the drop/repeat frame time shown... Well, the temp of your system...

If you're getting hours reported, you are good!

QB

[rod93]

Hi everyone, the infamous stuttering issues with madVR are back. This time, the render queue eventually drops to 0 and stays there. I’ve been trying to fix it for two days now. Is anyone else experiencing this at the moment? I managed to isolate the cause: it’s coming from HDR tone mapping.

[tp4tissue]

Sup madvr peoples,

Tp4 has gotten a new intel b390 igpu laptop. It's pretty nifty. the question is, is B390 fast enuff for Madvr + Tonemapping.

Happy to report that it is.

This laptop has a pathetic non 4K resolution, merely sufficient for 16inch, 2880x1800, 120hz, 600nit full field screen qdoled, use ATNA60HU01 CCSS from displaycal database for calibration.

The Acer tracks the eotf accurately at around 74 brightness, hdr content slider 50. Measured with i1d3. if you are measuring, use v113 mtpg, latest versions are bugged for hcfr.

Through Lav, you must use dxva2 native decode to minimize latency, Dx11 for whatever reason adds alot of ms, and it will drop frames.

Do not use DXVA scaling, it's bugged, use any other algo, SSIM for downscaling looks great. well within the ms envelope.

Copyback for cropping is fast enough in dxva2 for 2:1 aspect ratio movies, but it adds alot of ms, and for example, there's not enough headroom to crop a full 16:9 frame into 16:10, it will exceed 41ms. Maximum with copyback, this gpu will top out around 16:8, You'll need a build that pumps the GPU another 15-20%.

This refers to a difficult to tonemap movie with lots of grain like, starship trooper 1997. Garbage movies like "Obsession" which is completely vasaline smooth will crop easily, because there's not alot to tonemap.

You can use dxva2 native and zoom in instead to achieve crop at 37ms-ish.

acer swift 16, it has 65watt peak, 45watt sustained, even many more expensive rigs clamp to 30 sustained, be aware, Madvr with tonemapping goes at least 37watt between cpu and gpu.

Even if you get one with rated 45watt, you have to test, because for example, the msi costco prestige, this variant overheats like crazy, and will drip to 24watts.

The strongest B390 you can buy right now is the HP omnibook x variant with 63watt sustained. It's a bit heavier at 4lb vs 3.5, The problem is they only offer it with a 2048x12** res panel, which is basically just 1080p for $1500, GTFO. The other more expensive omnibook chassis clamps even lower.

If you guys get b390, CHECK the wattage, for example The Dell $2000 one also have 30ish sustained. No bueno.

[Sunspark]

dxva2-native decoding is supposed to be iffy with madvr but apparently not so with other renderers (to my understanding).

How is the latency if you set decoding to "None" given that this is a brand new machine?