Rendering an 11 mintue video takes 8 hours +

TechWizard wrote on 6/25/2020, 11:46 PM

Hi there,

For a university project I need to make a video and edit evrything together. A few weeks back I had to make a 7 minute video which rendered in about 1.5 hours (about 4 fps). Now I needed to make an 11 minute video, however this is taking way longer to render. currently it is at 8.5 hours and is expected to take another 4 hours (currently averaging around 1 fps). There is almost no GPU usage, not a single core is maxed out and I am just utterly frustrated that most of my resources are not being used.

PC Specs:

Intel i8700k, IGPU enabled

Nvidia GTX 1080 Ti, Game Ready Drivers

32 GB of DDR4 2666 MHz RAM

960 EVO 1TB SSD where it is writing to, ST2000LM015-2E8174 Seagate Barracuda 2TB HDD where the material it is editing together is stored.

The video I am rendering is a 1080p 60 fps, HVEC format. It says in the overview that hardware encoding is enabled, though I have my doubts about it. Here is a screenshot of the export overview:

Any help is appreciated!

Back to post

Comments

johnebaker wrote on 6/27/2020, 11:43 AM

@TechWizard

. . . . Both of my monitors are connected to the 1080 . . . .

Try connecting one to the UHD 630, without a monitor attached it may be 'deactivated'. I have have mine connected one monitor to UHD 630 the other to the 2060

. . . . on the latest graphics driver, 26.20.100.7262 (for UHD 630) . . . .

You appear to be missing several drivers updates .7262 was released in Nov 2019. I am on 26.20.100.7985, and the latest appears to be 27.20.100.8336 (Windows 2004?)

John EB

CubeAce wrote on 6/27/2020, 1:16 PM

@TechWizard

Hi.

I make the latest Intel 630 UHD studio driver to be 27.20.100.8280.

I'm sure John EB will say, but I think you either need a ghost header for the Intel GPU output or connect a monitor to it. I may be wrong on that, hence my caution.

I am also running Windows 10 2004.OS build 19041.329

Ray.

[Edit]. John beat me 😁

Scenestealer wrote on 6/27/2020, 6:39 PM

@TechWizard

. . Both of my monitors are connected to the 1080 . . . .

I agree with John and Ray - connect each GPU to a different monitor. I have the Display Port from the Intel connected to my primary monitor and my Nvidia to my secondary monitor although which is connected to which doesn't seem to affect export speed.

What does affect export speed on my system is which one is selected in the "Playback in Arranger >Video Mode" selection. Fastest by 30% or more is with the Nvidia GTX1060 set in Video Mode.

No I meant I could not select it for the "Video Output".

You took my point though.....that Video output is not relevant to the discussion about HW acceleration? The fact that you could not select the GTX there, would probably be due to the fact that you do not have a separate output available to the program because you are only outputting from one GPU which has been set to extend the GUI over both monitors.

I do not have a dedicated camera, or one that can record in a decent resolution so I guess any camera material I might get my hands on is not going to be a representative sample.

As long as it is HEVC and is not from a device like a phone that uses Variable Frame rate Recording (check with Media Info) but best would be to use an HEVC export you have successfully rendered already from VPX. It would help a lot if you could please supply all the information we have asked for to date.

I think it is important to use a DCH Intel driver that is up to date with the latest Windows 10 version on your system.

Peter

TechWizard wrote on 6/28/2020, 4:11 AM

@johnebaker, @CubeAce, @Scenestealer

Hi all,

I have updated my iGPU drivers, the graphics command center told me I had the latest drivers though when I went to the Intel site I could download the new ones. I have also connected one of the monitors to the iGPU. This allows me to see activity on both GPUs (20% on the iGPU, 6% on the 1080), though I do not notice a significant speed increase, it is still hovering around 25 minutes.

I willl try to see if changing the power setting to high power has any effect on the render times and I'll edit this post with the results. Afterwards I will try encoding the render I am doing now (the project on one timeline) with a template thrown over it and see what results I get when using both options (power saving and high power).

EDIT: The results are in and I can confidently say that high power or power saving has no effect on render times (assuming the Video Mode is on the iGPU for both power settings). Both have a rendering rate of 24.33 fps. I did not test whether changing the Video Mode has a significant effect, I will do that with the HEVC test later on.

Interesting nonetheless to notice that nesting has such a significant effect on the rendering times. You would think it would work relatively equally but apparently it does not.

Anyway, thank you all for your suggestions and you will hear the results of both tests soon!

EDIT2:

whilst exporting to HEVC using an unaltered VPX template

@Scenestealer could you please elaborate what you mean with an unaltered VPX template? Is this just the default when you import the clip or some template from the "Templates" tab?

- TechWizard

CubeAce wrote on 6/28/2020, 5:16 AM

@TechWizard

Hi.

Did you make sure that you did not tick the calculate video effects on GPU box?

That seems to be an important step in reducing export times and does not stop the GPUs from being used.

No idea why that should be but it makes a difference.

Ray.

TechWizard wrote on 6/28/2020, 5:51 AM

@CubeAce

Hi,

I did not tick that box, in both of the tests. And there was GPU activity on both GPUs so it did indeed not stop the GPUs from being used.

- TechWizard

johnebaker wrote on 6/28/2020, 7:44 AM

@Scenestealer, @CubeAce

@TechWizard commented

. . . . Maybe something has changed in the Windows graphics settings since the newest update . . . .

and that VPX was starting with the NV GPU selected.

Something has changed in Windows 10 2004, the issues I had with VPX and MEP not starting with the NV selected and Windows 10 1909 is fixed, the GPU drivers did not change between Windows versions!

@TechWizard

. . . . Interesting nonetheless to notice that nesting has such a significant effect on the rendering times . . . .

It would be interesting to see what the export time is when everything is on one timeline.

John EB

Scenestealer wrote on 6/28/2020, 7:51 AM

@TechWizard

By an unaltered export template I mean choose "File >...>Export HEVC" then in the export window tick "display all"and from the dropdown choose a template that best matches your movie settings which from your first post would be 1920x1080 60P - then export your movie / clip(s) without altering any of the Advanced encoder settings.

I still think your export time seems a little slow and would still like to see the full Media Info view of the HEVC clips that you have resized against the background JPEGs. If you are not familiar with "Media Info" it is a free utility that can be downloaded and used to analyse video clips for Frame Rate, Bitrate, GOP, Audio data, etc.

Peter

browj2 wrote on 6/28/2020, 9:21 AM

@TechWizard

Are you using any special effects like Neat Video, by any chance?

John CB

TechWizard wrote on 6/29/2020, 12:47 AM

@johnebaker

Hi all,

It would be interesting to see what the export time is when everything is on one timeline.

All previous export times I mentioned where for one timeline (with the exception of the 26 hours, that was using nested timelines), I have not tested it again with the nested timeline as I was not impressed with the intermediate results.

@browj2

Are you using any special effects like Neat Video, by any chance?

No I do not, I only use transformations and resizing of the video clips from the effects tab (and a panorama effect to get the audio balance right). I have not used any kind of filter or other special effects.

@Scenestealer

I will report back later in the day with the test results using the settings you have given me, including the "Media Info" you described.

- TechWizard

johnebaker wrote on 6/29/2020, 3:26 AM

@TechWizard

Can you post a screen shot of the timeline please.

John EB

TechWizard wrote on 6/29/2020, 5:56 AM

@CubeAce, @Scenestealer, @browj2, @johnebaker

Test Results

Hereby a post to detail the test results from some tests I have run with various settings. (EDIT: Added the Media Info files as @Scenestealer requested [10]).

Setup

Intel I7 8700k hexacore running at 3.70 - 4.30 GHz
- Intel UHD Graphics 630, Driver 27.20.100.8336
Gigabyte NVidia GeForce GTX 1080 Ti waterfoce 11 GB VRAM
- Studio Driver 451.48
32 GB 2666 MHz DDR4 HyperX RAM
ST2000LM015-2E8174 Seagate Barracude 2TB HDD (Source material on this drive, EDIT this is the D drive and has 1.5 TB free)
960 EVO Samsung 1 TB SSD (Rendered video written to this drive, EDIT this is the C drive and has 426 GB free)
Windows 10 Home Version: 2004 19041.329
Magix Video Pro X 18.0.1.77 (UDP3)

Test Project Setup

Timeline with only 1 clip [1], a 10 minute 57 seconds 1920x1080 60 fps HEVC with 48000 audio sample rate. This is then exported as an HEVC video with the unaltered MP4 FullHD 1920x1080 59.94 render template.

Parameters being Tested:

Windows Graphics Setting (Power Saving mode versus High Power Mode)
VPX Display Options -> Video Mode -> Grpahics Card (Intel UHD versus GTX 1080)

Results

Keep in mind that for most results the seconds are not a significant factor, except for the GTX test where the rendering process is faster than realtime playing the video (i.e. the rendering is completed before the video playback is completed if both processes start at exactly the same time).

Power Saving - Intel UHD

Intel UHD usage: 57%
GTX 1080 usage: 6%
Render Time: 20 mins
Render FPS: 32.8
Media Info: [10]

Power Saving - GTX 1080

Intel UHD usage: 11%
GTX usage: 14%
Render Time: 9 mins 20 seconds
Render FPS: 70.3
Media Info: [10]

High Power - Intel UHD

Intel UHD usage: 55%
GTX usage: 7%
Render time: 19 mins
Render FPS: 34.5
Media Info: [10]

High Power - GTX 1080

Intel UHD usage: 11%
GTX usage: 14%
Render time: 9 mins 30 seconds
Render FPS: 68.8
Media Info [10]

Conclusions

Turns out that Power Saving or High Power have neglible effect. Though the results indicate that Power Saving is slightly faster, it does not necessarily have to be since multiple factors can have a small effect on rendering times like the load of the computer while rendering (if the computer is for instance playing music during rendering or displaying a PDF file). Interesting though is eventhough the consensus in the forums is to use the Intel UHD GPU in the VPX settings, it turns out that using the GTX brings significant speed gains. Though this might be a new feature in the newest VPX version (released on 08/06/2020).

Testing the Problem Project

Based on the results of the previous test, I decided to test the new found settings and see what the results were for the project for which I initially made this thread. So let me specify the changed Project Setup.

Project Setup

A timeline with 6 movies on it [2], crossfading into each other (to ease the transition between them). Each movie [3][4][5][7]has some resizing of 1 long or 2/3 shorter 1920x1080 60 fps MP4 (most likely H.264) clips. Audio may or may not have the Panorama effect and all clips have a silver border around them. Each clip is put on top of a PNG image, which is a slide from a slideshow. Part 4 [6] and Part 6 [8] are special as they have at some point 3 clips playing at the same time (same resolution, framerate and encoding) as well as some titling. The main timeline [1] is then exported as an HEVC video with the unaltered MP4 FullHD 1920x1080 59.94 render template.

Another project the above is tested against is the same project but then all the movies merged together into one timeline [9]. This is exported with the same settings as above.

Both projects have been tested on the High Power - GTX 1080 combination of settings. In the results section the first project is referred to as nested, whereas the second project is referred to as merged.

Results

Nested

Intel UHD usage: 10%
GTX usage: 4%
Render Time: Rendered 17413 frames out of 39381 in 90 mins
Render FPS: 3.2
Media Info: [10]

Merged

Intel UHD usage: 10%
GTX usage: 5%
Render Time: 27 mins
Render FPS: 24.3
Media Info: [10]

Conclusions

It seems to be that a nested project takes significantly longer to render than a project that has everything on one timeline. Why that is, no clue. Probably a mistake in the nested render algorithm.

Screenshots

Note that all faces in the clips have been redacted manually to conserve anonimity and privacy of the people involved.

[1]

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

https://mega.nz/file/EFwRBITY#0jsKTvHdcVyg1IpUe2gGV03MS3sjIh4_2DHZQYNg0yM

CubeAce wrote on 6/29/2020, 7:15 AM

@TechWizard

Hi.

You say you exporting the project to your C: drive.

Currently how much free space is on your C: drive and how large is the file size of the nested project in total please?

Ray.

TechWizard wrote on 6/29/2020, 8:15 AM

@CubeAce

Hi,

The C drive has currently 426 GB free. The nested project when (nearly fully) rendered is 566 MB. The project itself (when copying everything into the project folder) is 1.43 GB on the D drive (HDD).

- TechWizard

johnebaker wrote on 6/29/2020, 8:50 AM

@TechWizard

Thanks for the images of the timelines and the results.

However we are still missing the MediaInfo data on all the source files, the only MediaInfo data provided in the zip file is the finished export files which is not relevant to the issue.

The export results are more in line with what I would expect, especially now it is clear that there is a lot more going on then was originally indicated in your previous posts in particular:

multiple video file formats - HEVC, MP4 (h.264?) and MKV [6]
resizing of mp4 (h.264?) in movies [3 - 5] and [7]
re-encoding of mp4 (h.264?) and MKV
3 clips playing at the same time, and some titling [6] and [8]

All of which will contribute to longer render times.

John EB

TechWizard wrote on 6/29/2020, 10:05 AM

@johnebaker

Hi,

You are completely correct, It did not come to mind when writing my very first post. Please find the Media Info for all the source files in the link at the bottom of this post. Is it expected though that nesting extends the rendering time significantly? If it is then I completely missed that when browsing the forum before and only discovered that when I tested it. But yes, 27 minutes for an 11 minute video is definitely not bad, especially given the effects applied in Part 4 and Part 6.

- TechWizard

https://mega.nz/file/0EJhQY4b#Wto8QkwlFeTD56Z_rUnLTqHpqoDf6v2H3epNORg4JYk

johnebaker wrote on 6/29/2020, 11:36 AM

@TechWizard

Thanks for the source video Media Info.

I can see why the rendering is somewhat slower then we originally anticipated for the following reasons:

all source videos are AVC codec and exporting to HEVC means every video clip is being transcoded
there is a mix of frame rate modes - constant and variable. For the variable frame rate clips the program is having to recalculate new frames to replace the 'missing' frames.
there is a mix of 1080p and 720p resolutions - rescaling is having to be done on all the 720p clips
there is a mix of framerates 29.97, 30 and 60fps - exporting at 60 fps means the lower frame rate videos are taking longer to export as frames must be doubled up or recalculated
there is a mix of bitrate modes - variable and constant.
a mix of TV standard PAL and NTSC
a mix of colour space BT601 and BT709

The combination of the above all add up to extra work for the GPU. This will become more relevant if you have to create longer projects.

I assume this is a collaboration project and that some of the clips have been through a conversion program eg Handbrake, or exported from an animation package (Blender?).

If so I would make the following suggestions for future collaboration projects:-

Define among the project members a standard file format for the video components ie

container format - eg mp4
codec - h.264 or h.265
bitrate mode - always use variable - perceived quality will be better
resolution - 720p or 1080p
Standard - either NTSC or PAL
frame rate - mode: constant,
framerate : NTSC 30 or 60 fps, PAL 25 or 50 fps
Colour space - BT709
export format, unless it was specified that the submitted video must be HEVC format (to save on storage space?) then export to mp4, h.264 with the framerate and resolution matching the source video - this is the most universal format for playback.

HTH

John EB

TechWizard wrote on 6/29/2020, 11:49 AM

@johnebaker

Thank you for the tips, I will definitely take them into account for the next project. I will mark the large "test" post as the solution as it contains all the necessary info for other forum members to look at and see why it was taking so long in my case.

Thanks you, @browj2, @Scenestealer and @CubeAce fo all your insights and tips. It is all very much appreciated.

- TechWizard

Scenestealer wrote on 6/30/2020, 7:38 AM

@TechWizard

Thanks for your comprehensive set of tests and reporting of results - very interesting and @johnebaker has made the same points I would have mentioned. The only thing to add there is that if the end requirement for the export can handle HEVC then I would use that because VPX is set up for the maximum performance boost via the discreet Nvidia card, especially if the source are clips are HEVC. VPX11 introduced much accelerated rendering, especially when there are multiple tracks of HEVC, such as a multiple PinP or collage, during preview and export by utilising the NVDEC HW decoding chip and NVENC HW encoding chip on the Nvidia cards. VPX12 has extended this support for H.264 decoding via AMD and Intel HW also but "I believe" HW encoding H.264 is not available on any of the discreet GPU's hardware encoder layers.

Interesting though is eventhough the consensus in the forums is to use the Intel UHD GPU in the VPX settings, it turns out that using the GTX brings significant speed gains. Though this might be a new feature in the newest VPX version

I mentioned earlier that like you I found setting my GTX1060 in Video mode consistently give an encoding speed boost but no one has figured out why this should be as this setting is only supposed to affect preview smoothness (even according to Magix support!).

As far as the GPU selection for VPX in the Windows display settings goes - we will just have to assume that something has either changed in Windows or changed in the latest VPX12 that has changed how the program reacts to this setting. Depending on which installs first on my system - Win10 2004 or VPX12- I might be able to tell more later.

I would still like to see the Media Info for the single 10min clip HEVC clip that you encoded to HEVC along with the graphs from Task manager for each GPU, showing 3D, Video Decode and Video Encode, to see what might be occurring with VPX12 as I have not yet installed this version. But I fully understand if you can not be bothered as we have asked quite a bit from you in this thread.

Thanks again

Peter

TechWizard wrote on 7/2/2020, 10:01 AM

@Scenestealer

It took a while but it has been done (exams got in the way). Please find the zip file at the following link: https://mega.nz/file/RFIESaQT#FrbLHUSwzDc0rX5xPAlpFhNu8x3g8oPy-rg0FWZ7EyE

Hopefully this gives you some insight. Do note that the two spikes in the iGPU graph (3D) are most likely caused by taking a screenshot of the task manager window.

- TechWizard