MadVR Settings For 4K UHD

[mikeyjeff]

It seems like something else must be running every 10-20 minutes. That is way too many frames for it to be the normal suspects.

Is anything using your hard drive? What if you keep a monitoring program open, like Task Manager, Resource Monitor, and/or hwinfo?

Edit:
Do you have f.lux (https://justgetflux.com/) or something that might be like it? Anything that changes the color of your display?

Hi @asmodian - appreciate your thoughts.
I don't believe I have anything of that nature running on my PC apart from the bare essentials. The only thing I can think of would be my virus scanner (AVG Free) which I uninstalled last night and am using the stock standard version that comes with Windows 10 Pro (1803). Virus scanners are notorious for using plenty of resources. I have the Windows HDR mode on that's the only thing affecting the display and all the standard settings in the Intel Graphics HD panel, although would be happy to get some thoughts on what should be on there.

The one thing I'm not sure of (and will check this given your feedback) is to see whether the interruptions are at similar or exact time intervals to give extra weight to your hypothesis. I am suggesting this 10-20 min timeframe but need to do some digging to see whether there is any pattern to this. I'll do as you suggest with Taskman and see what I discover - it's pretty bare bones...

I tried to show some "Ctrl-J" images last time without success - I'll try to provide the link below as pure copy/paste to show the nature of what is happening, along with perhaps some other stats that might help!:

First Image - as it is playing:

https://drive.google.com/open?id=10qnCrctj2BPZYlDIsnfMduYNUiN5qbrQ

Second Image - after the frame drop ~10-20 min worth of playback

https://drive.google.com/open?id=1QijYDOAVi7BqinTT-snsqi3fptr0ZkGA

Note - I had it on 24hz for 23.976 fps hence, one frame repeat every minute or so, but honestly, I never notice it - it's purely the 10-15 frame drops that drive me crazy. If I choose 23hz in the Intel GPU then I get hours or days before frame repeats, but tried 24hz to see if that might solve the problem too!!

I've also tried dumping GPU decoding and pure CPU decoding and it improves render rates a great deal (Intel's GPUs must really suck) but doesn't solve the dropout problem. Have also tried removing GPU flushing and lowered queues too...

Thanks again!
Mike.

[Asmodian]

Everything looks good, do your max stats (5s) spike when you get the frame drops?

Intel's GPUs do suck, sticking to software decoding is probably a good idea, but I agree that it probably isn't the issue. Good performance for 10 minutes but with large numbers of dropped frames together really does not seem like normal performance or timing issues I am used to.

No temperature issues or power throttling? hwinfo (https://www.hwinfo.com/download.php) would help monitor, it even has a max column where you can see if your CPU or GPU has ever hit any performance limit (temp, power, voltage, etc.). "Sensors only" is a nice option when launching it. :)

[mytbyte]

Disabling "present a frame for every sync" did a lot for me to avoid dropped frames on a Atom Intel iGPU (that's only if you run @ 50 or 60Hz refresh)

[mikeyjeff]

Disabling "present a frame for every sync" did a lot for me to avoid dropped frames on a Atom Intel iGPU (that's only if you run @ 50 or 60Hz refresh)

Thank you both - I will try that diagnostic software to see what is going on. I run Core Temp in the background just to monitor the CPU temperature and I've never seen it get above 45 Celsius (it is winter here in Australia at present) - it normally sits around high-30s so I doubt the temperature is the issue, but love to be thorough, so will try HWInfo and see what it uncovers. I'll also try to use Task Manager to see what is going on (although hopefully that doesn't influence/interfere with playback as I'll have to have it on the same window "bring to front" as the playback itself).

I'll also try that Sync option and report back.

I really appreciate all of your help!
Cheers,
Mike.

[mikeyjeff]

Thank you both - I will try that diagnostic software to see what is going on. I run Core Temp in the background just to monitor the CPU temperature and I've never seen it get above 45 Celsius (it is winter here in Australia at present) - it normally sits around high-30s so I doubt the temperature is the issue, but love to be thorough, so will try HWInfo and see what it uncovers. I'll also try to use Task Manager to see what is going on (although hopefully that doesn't influence/interfere with playback as I'll have to have it on the same window "bring to front" as the playback itself).

I'll also try that Sync option and report back.

I really appreciate all of your help!
Cheers,
Mike.

Hi guys - I've used the HwINFO64 software as suggested, did 3 independent tests, but using the same content and same preconditions. A few observations:

1. There isn't a particular time period when the frame drops occur - on one occasion it was within minutes of starting the file, on the second occasion I got 10 minutes into the film and on the third occasion it was within a few minutes again. That doesn't mean that it isn't related to periodic system functions, I will look into that further, but it seems less likely.

2. On all three testing occasions, when the frame drops occur, there is a material difference in the following key measures:

GPU D3D Usage [%] GPU GT Usage [%] GT Cores Power [W] and GPU D3D Memory Dynamic.

During "good" playback, these figures fluctuate between:

GPU D3D Usage: 2-8% when not being used, 42-44 when the movie was playing, and at the point of frame loss - it spikes to 65.7. That was conclusive with test #2 where it went to 67, and again, 42-44 during normal usage - quite suspicious!!

GPU GT Usage: Again 2-8% when no movie showing, 42-44 when the movie plays well and again at frame loss it spiked to 64.9, 66.9 in the other test!

GT Cores Power [W] (obviously a function of the Usage) - 1.1-1.5 during normal playback, and spikes to 6.227W when the frame loss occurs.

GPU D3D Memory Dynamic: Without the movie playing, it is at 607-631MB, movie playing between 1056 and 1092 (it is either of these two figures) then at the point of frame loss failure, it spikes to 1136MB - My Intel i7 8700 comes with 1GB of GPU memory - am I quite simply running out of memory in my graphics GPU?

Note that I turned off hardware acceleration, but obviously it is still using the GPU but just not to decode HEVC clearly the CPU is doing that.

Also note that there were no temperature alerts, or any other alerts related to temperature, hard disk reading - all looked good.

I am happy to make the CSV files available if you wanted a closer look but on the surface of it, it seems as if the files I'm trying to play simply cannot be handled by this internal GPU and perhaps I should just bite the bullet and get a NVidia GTX or something that has more RAM and processing power? It might then have to flush its memory out, start again, run out of memory, spit the dummy, and so on the process goes. I'm not sure that's how this works exactly, but there seems to be some consistency in what happens when the frames are dropped.

What do you think guys? There are other columns that spike also but not by the same magnitude and in all honesty are probably driven by these metrics above.

The file was a ~30GB MKV movie file H.265 HEVC encoded, BT2020 colour gamut (HDR10) and 3840x2160 resolution with DTS-HD sound encoding.

Happy to make the CSV files available and provide a link and point you to the precise points where I'm confident the frame drop occurred. (It's not an exact science, as soon as the frame drops happened, I killed MPC-HC and then as quick as I could, stopped the logging on the HWinfo software!)

Welcome your thoughts guys and again, much appreciated.
Cheers,
Mike.

[Asmodian]

Do you have any BIOS options to raise the memory allocated to your GPU? Intel iGPUs share system memory and it should be possible to allocate it more.

Have you tried all the way down to the minimum GPU buffers and frames presented in advance? It does seem like you are running out of GPU memory, even with small buffers. :(

I agree, it doesn't sound like temperature or another application.

[mikeyjeff]

Do you have any BIOS options to raise the memory allocated to your GPU? Intel iGPUs share system memory and it should be possible to allocate it more.

Have you tried all the way down to the minimum GPU buffers and frames presented in advance? It does seem like you are running out of GPU memory, even with small buffers. :(

I agree, it doesn't sound like temperature or another application.

I am going to investigate that tonight to see if I have the ability to assign more memory to the GPU. I have 16GB of RAM on my Z370 motherboard, I'm also consulting the BIOS to see if I can do that - most modern day motherboards seem to allow this, so will look into that.

If it is of interest, attached are two of the CSV files that the monitoring software generated. The time series where the frame drop clearly took place is row 46 in the file "mpc test" (I exit the MPC-HC software immediately, hence the next two entries bottom out) - and row 204 in "mpc test 2" - just for interest - you may discover something else I don't.

It did occur to me that given the huge size of these files (~30gb) that this might be the reason why it is problematic - it is obviously trying to process such large files and perhaps 1gb of RAM for the GPU is insufficient. Who knows!

I'll do some GPU RAM / BIOS analysis tonight and report back.

File links attached if there is some interest on sleuthing further! :)

Test 1:
https://drive.google.com/open?id=1xWX-JGgC4ANeWwSIBFeOFbq88Zjerxct

Test 2:
https://drive.google.com/open?id=1JEa6i6Es9ceZsTq40x22MjNgqnplv8VX

Cheers guys!
mike.

[Asmodian]

The size of the video file is not important, it is the size of each input and output video frame that is important. I suspect memory use is the culprit, 1 GB is simply not quite enough for a 4K display. That said your test two has a similar effect but GPU memory only jumps from 602 to 894 MB, not above 1 GB, you also see a spike in GPU usage and power at the same time. There is even a 122 MB jump in physical memory used then, significant only because it had been so stable. Both CPU and memory load show a similar effect, a small jump that shouldn't be an issue.

Using madVR defaults I see 1599 MB of video memory used. Using the defaults but with 4 GPU buffers and 1 frame presented in advance I see 982 MB of video memory used watching 1080p bluray on a 4K display and 1067 MB watching UHD HDR bluray on a 4K display. Using my normal settings watching UHD HDR on a 4K display I see 3111 MB.

With 1080p at 1080p I only see 745 MB of video memory used even with madVR's defaults (8 GPU queue and 8 frames presented in advance), 1014 MB watching UHD HDR at 1080p, so 1 GB is plenty for a 1080p display. Hopefully you can increase the memory available to the GPU.

I do get stable playback even with only one frame presented in advance (no presentation glitches), so you might be able to get away with 1 GB if you only want to watch 1080p video. :(

[nsnhd]

Using madVR defaults I see 1599 MB of video memory used. Using the defaults but with 4 GPU buffers and 1 frame presented in advance I see 982 MB of video memory used watching 1080p bluray on a 4K display and 1067 MB watching UHD HDR bluray on a 4K display. Using my normal settings watching UHD HDR on a 4K display I see 3111 MB.

With 1080p at 1080p I only see 745 MB of video memory used even with madVR's defaults (8 GPU queue and 8 frames presented in advance), 1014 MB watching UHD HDR at 1080p, so 1 GB is plenty for a 1080p display. Hopefully you can increase the memory available to the GPU.

I do get stable playback even with only one frame presented in advance (no presentation glitches), so you might be able to get away with 1 GB if you only want to watch 1080p video. :(

These are very interesting numbers that no one posted before, that may help to chose GPU cards with 3GB or 4GB Vram. Some users just claim that more than 3GB is needed for 4K display. If I understand correctly, just reducing GPU queues and frames presented in advance a little you can be safe with 3GB Vram ?

[huhn]

you can use a 3GB card for UHD savely with default settings.

lowering queues may break playback completely and may break deinterlancing.

subtitles incresses Vram usages and they are without any doubt needed for some movies like avatar
NGU needs a lot of Vram.
the Vram used by the OS can very alot.

there are quite some issues if you look at this.
in the past we had a 1060 3GB a powerful card that can do a lot of stuff in madVR and not really because it will sooner or later use to much Vram.

than we had the 1050 which would be fine with 3GB because you can not really stress this GPU enough to use more than 3 GB of Vram but it was limited to 2 GB which is clearly not optimal.

and here we are with a 1050 3GB card now that has quite a punch but really slow memory which may or may not result in a problem.

this case it about a iGPU they rarely really care if they run out of "Vram" the ram has the same speed anyway and they usually just take as much Vram as needed.

madVR default setting give me these numbers:
1080p at 1080p ~1.2 GB
1080p at UHD ~1.9 GB

[mikeyjeff]

The size of the video file is not important, it is the size of each input and output video frame that is important. I suspect memory use is the culprit, 1 GB is simply not quite enough for a 4K display. That said your test two has a similar effect but GPU memory only jumps from 602 to 894 MB, not above 1 GB, you also see a spike in GPU usage and power at the same time. There is even a 122 MB jump in physical memory used then, significant only because it had been so stable. Both CPU and memory load show a similar effect, a small jump that shouldn't be an issue.

Using madVR defaults I see 1599 MB of video memory used. Using the defaults but with 4 GPU buffers and 1 frame presented in advance I see 982 MB of video memory used watching 1080p bluray on a 4K display and 1067 MB watching UHD HDR bluray on a 4K display. Using my normal settings watching UHD HDR on a 4K display I see 3111 MB.

With 1080p at 1080p I only see 745 MB of video memory used even with madVR's defaults (8 GPU queue and 8 frames presented in advance), 1014 MB watching UHD HDR at 1080p, so 1 GB is plenty for a 1080p display. Hopefully you can increase the memory available to the GPU.

I do get stable playback even with only one frame presented in advance (no presentation glitches), so you might be able to get away with 1 GB if you only want to watch 1080p video. :(

Hey guys - I've been doing a bit of reading on this and it seems that Intel Graphics HD actually doesn't have any dedicated RAM of its own - it simply uses system memory but the 128mb that sits there is fictitiously put there: - straight from the Intel site:

By default, the Intel graphics driver will report 128 MB of fictitious Dedicated Video Memory for compatibility with applications that don’t correctly comprehend a fully unified memory architecture. See Dedicated Memory Reporting for more information.

So therefore, by checking my Display settings, I have access to heaps of RAM:

https://drive.google.com/open?id=1WKJDyf_9i44kGjfjokEqLOj5q1u7n7rD

This kinda leaves me scratching my head - the 1GB GPU memory I read is clearly not right, I have access to heaps of RAM which still doesn't explain why I'm getting these frame drops.

Unless there's something I'm missing - well and truly searching for some troubleshooting tips!

Best wishes,
Mike.

[Asmodian]

Maybe it is something else launching or running in the background, that would explain the small bump in memory usage, % cpu, % memory bandwidth, GPU memory usage, etc. :confused:

[mikeyjeff]

The size of the video file is not important, it is the size of each input and output video frame that is important. I suspect memory use is the culprit, 1 GB is simply not quite enough for a 4K display. That said your test two has a similar effect but GPU memory only jumps from 602 to 894 MB, not above 1 GB, you also see a spike in GPU usage and power at the same time. There is even a 122 MB jump in physical memory used then, significant only because it had been so stable. Both CPU and memory load show a similar effect, a small jump that shouldn't be an issue.

Using madVR defaults I see 1599 MB of video memory used. Using the defaults but with 4 GPU buffers and 1 frame presented in advance I see 982 MB of video memory used watching 1080p bluray on a 4K display and 1067 MB watching UHD HDR bluray on a 4K display. Using my normal settings watching UHD HDR on a 4K display I see 3111 MB.

With 1080p at 1080p I only see 745 MB of video memory used even with madVR's defaults (8 GPU queue and 8 frames presented in advance), 1014 MB watching UHD HDR at 1080p, so 1 GB is plenty for a 1080p display. Hopefully you can increase the memory available to the GPU.

I do get stable playback even with only one frame presented in advance (no presentation glitches), so you might be able to get away with 1 GB if you only want to watch 1080p video. :(

Folks, something that might be quite interesting here - I had Task Manager and the GPU graph open whilst playing a 4K file and at the precise point of frame drop out, have a look at the graph. I don't know what causes it to do this, but the graph spikes to 100% then drops down to 50-odd % and then works its way back to its nominal range. For what it's worth, with normal playback, it almost locked at 70% and simply doesn't budge up or down, until frame drop.

Bottom line: Something is causing the GPU utilisation to spike to 100% and I'm trying to figure out what is causing that - because whatever that is, is causing this frame drop. I'm not certain that it's the content in the particular scene - it was the same scene, so it's not as if it was a drastically different set of graphics than seconds before, or seconds after.

Anyway, perhaps this might add some evidence to the collection!

See link attached:

https://drive.google.com/open?id=1aFeSY1r-JE2em-pydBQLshG7rxKG2OtN

I have circled the "frame drop" bit in red!

Cheers guys
Mike.

[foxyshadis]

I strongly suspect that a regularly occurring process is really messing up your watching. Aside from CPU, watch the disk activity and queues too, that will often give you a better idea of the actual performance culprit, since disk contention is generally the #1 cause of performance problems that aren't driver-related these days. Check task scheduler, and look through the full task manager list and just try killing things that aren't built-in to Windows. Also, turn off Windows Defender real-time; when you're on the ragged edge of viewing capability, you need every edge. (Downside: It'll pop up windows over your movie whining at you to enable it, potentially also messing up performance.)

[mikeyjeff]

I strongly suspect that a regularly occurring process is really messing up your watching. Aside from CPU, watch the disk activity and queues too, that will often give you a better idea of the actual performance culprit, since disk contention is generally the #1 cause of performance problems that aren't driver-related these days. Check task scheduler, and look through the full task manager list and just try killing things that aren't built-in to Windows. Also, turn off Windows Defender real-time; when you're on the ragged edge of viewing capability, you need every edge. (Downside: It'll pop up windows over your movie whining at you to enable it, potentially also messing up performance.)

Thanks for this - yes, I'm beginning to think this is worth pursuing.

I will assemble some software to do some deeper monitoring. I'm getting MSI Afterburner, GPU-Z as well as Process Explorer and will do a set of tests not using MPC-HC content whatsoever and see if we get that spike (if we do, then we're well on the way to figuring out what it is!) - if not, then clearly something is causing MPC-HC/LAV Video/MadVR to spike given if it was the content itself, then it would happen in the same spot every time, which it simply doesn't.

Any other thoughts on software I should use for this kind of detailed investigation?

Thanks all!
Mike.

[mikeyjeff]

I strongly suspect that a regularly occurring process is really messing up your watching. Aside from CPU, watch the disk activity and queues too, that will often give you a better idea of the actual performance culprit, since disk contention is generally the #1 cause of performance problems that aren't driver-related these days. Check task scheduler, and look through the full task manager list and just try killing things that aren't built-in to Windows. Also, turn off Windows Defender real-time; when you're on the ragged edge of viewing capability, you need every edge. (Downside: It'll pop up windows over your movie whining at you to enable it, potentially also messing up performance.)

Hi guys - I can happily say that this problem is now solved!

As suspected by many of you above (and thank you for your input!) the culprit was the rather poor way Windows 10 allocates GPU processing power. I did some baseline testing in that every time that I had frame drops, the GPU would spike significantly, and did the test with no movie running. It turns out that the GPU spikes significantly when my windows background changes image.

Clearly the way that Microsoft have implemented this code makes use of the GPU in such an inefficient way and moreover, prioritises this function over anything else it is doing. I did some further research and turns out many gamers are irritated with this, as they also get frame drops and lower FPS whenever this happens.

The obvious solution in my case is to simply turn off my desktop background slideshow, cycling through nice background photos.

As an aside, I find it astonishing that such a powerful PC (16gb RAM, i7-8700 with HD Graphics 630) with Windows 10 is unable to handle both of those functions without frame dropping but this is the case.

In any case, I hope this provides some insight to anyone else who has scratched their head trying to figure out what was going on with this one.

As the old saying goes, "All things being equal, the simplest solution is usually the right one". Couldn't be more true here.

Thanks again all - really appreciate everyone's feedback.

Best wishes,
Mike.