36 Hours of Testing MEP Premier Settings For Playback and Export.

CubeAce wrote on 3/26/2020, 9:34 AM

Hi all.

Prelude.

This is going to be lengthy, so please bear with me. You may find it interesting or complete rubbish.

First I'd like to say that the moderators here and few regular contributors are very knowledgeable and have helped /are still helping me a lot.
Without them I doubt I would have my current PC built setup and install of Movie Edit Pro Premium working as well and hassle free as it has been working.
That the odd glitches I've had along the way have been duly sorted by their collective wisdom and knowledge.
There has been however one point of contention I have not been able to come to terms with regarding MEP's performance and handling of both smooth playback and export times.


I have neither ever got export times down to what others have experienced nor had easy times with smooth playback considering how much I have upgraded my PC. Particularly with regard to John EB's claim that an addition of using an NVIDIA card seems to make no difference to export times

I can't remember if John EB has the same view on playback performance or not so I can't comment on that at this time until I get some form of conformation.

What I can say after a lot of testing is that from John EB's viewpoint and experience he is probably and most definitely correct in his observations if my following findings are what the cause of his findings are based upon.   

So,for the sake of this test I tried to keep it simple and not too taxing for any PC setup capable of running MEP Premium.

Windows was set for MEP to use the Intel GPU as 'Power Saving' in the 'Display /  Graphics settings'.
I used the same project for all of my export and playback smoothness tests.
I configured preview and proxy rendering settings were set to 'Best performance for playback (MXV) with Extract audio when creating proxy files.
All video clips were copied into the project folder so as to have the files on my fastest drive to read from.


The Main test Video project was using normal 1080 x 1920 variable bit rate 60fps h264 clips with only cross fades applied from beginning to end. No additional effects added.
I then ran each test on each of the two monitors, one being powered by the Intel Graphics chip and then the following test on the second monitor powered by my
aging NVIDIA card.

This is where I ran into the first possible difference between me and John EBs findings. Where, if John's MEP setup had to follow my setup, John would be perfectly correct and valid about reporting his findings and drawn conclusions.

This is down to how MEP's 'Program Settings' Dialogue box opens in the 'Display options' box and could be why John EB is not seeing any activity or benefit from using the NVIDIA GPU. It is effectively blocked from being used for any task within MEP. I've no idea why but the Task Manager reports no effective activity at all.

[Edit] have changed image file as accidentally uploaded wrong result.

I want to repeat. At this point these are the findings with my PC.

I do not conclude from the past days experiments that the results will be the same for everyone as computer systems differ but I hope others may find this useful.

So at this point I can say is that when I used MEP on the Intel powered monitor the NVIDIA card could do nothing. Playback was slightly glitchier on cross-fades (slowed down)
but not by much.
Export rendering always took longer and by some margin. Up to three times the length of the play time of the video. Cross-fade sections slowed even more.

Note no NVIDIA activity and export time more than project length.

Having MEP placed on my NVIDIA powered screen made the whole experience of using MEP more palatable in every way. I could see the NVIDIA card doing something and varying
The amount of use depending on whether it was at a cross-fade point of not. The Intel GPU 3D rendering at that point always dipped but so did the Intel Video decode on export, but the slowdown was nowhere near as much as when on the other monitor.
In general playback was much better as was scrolling and exporting times fell to below the length of play time of the video. Sometimes up to a third less time.

When using either monitor I saw similar parts of the processing having the same effect on performance while other parts of processing had no effect at all in regards to the Intel GPU or CPU's performance.
It did not seem to matter whether the Intel GPU was processing 3D to over 80% or as little as 10% or less. It varied depending on whether it was dealing with a cross-fade
or when exporting what type of codec was in use.
If and when the NVIDIA card could be pressed into action the Intel GPU took a hit and decreased both 3D and Video decode activity.

How big a hit I can't say or judge as my card is very low powered for this but suspect a higher powered card may reduce the hit to the Intel GPU. I suppose I will just have to wait until such times I may be able to think about getting one.

What I can say is that in my observations, how high in performance I can get the Intel Video Decode line in the graph to stay,
the faster the export times are with all codecs that I used in the tests. So base MP4, H264 and H265. All tests used the best quality settings I could apply.

There were other settings that helped playback smoothness and response times to scrolling and the like.
Adding titling or some effects like masking objects and tracking them made little difference to playback or export times whereas a lot of the third party plugins did slow down exporting a lot during the sections those effects were applied but curiously didn't effect playback that much.

So in conclusion.


On my system at least, using an NVIDIA card and setting up MEP to only be on the monitor powered by the NVIDIA card gives me the best overall playback experience and fastest export times regardless of whether it is HEVC or not.

My other more limited testing using a 4K set of clips and exporting at the same resolution were roughly the same except export times lengthened to roughly just over twice that of the project length.    

During all the testing I took snapshots of the performances of all exports during crossover sections and of final render times.
Labeled which screen they were on and whether the NVIDIA card was selected or not.

I have a few more tests I've thought of to round this all off such as whether to put the editing timeline onto the Intel powered monitor or not but that has just occurred to me.
It is possible to configure MEP to show up in different ways on the monitors anyway and save them as presets so if it turns out it is better to have everything on one screen then I could make a preset for that.

I will let you all know later.

What else am I going to do with my spare lock-down time? 😇

Keep safe and healthy.

Ray.

 

Last changed by CubeAce

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

Comments

johnebaker wrote on 3/26/2020, 3:26 PM

@CubeAce

Hi Ray

. . . . John EB has the same view on playback performance or not so I can't comment on that at this time . . . .

With respect to the playback performance, when working with 4K video, the results tend to be mixed depending on the complexity of the timeline eg collages and Picture in Picture and effects and transitions used in the project.

In general I get smooth playback and occasionally some slight jerking with complex parts of the timeline.

There are several options and steps I take when the timeline has complex elements in it:-

  • The Preview lightning symbol is turned on with the reduced resolution and frame rate options checked
  • Pre-render, complex areas of the timeline to achieve smooth playback, eg a collage that involves 4K video clips - for images only I do not pre-render.
  • Complex animated titles are created independent of the main project either in MEP or Titler Pro and exported as finished video for use on the timeline eg complex titles like this noticably slow down performance if made in the project

I rarely use such effects from the NewBlue and Hitfilm packs so cannot say how they affect playback performance.

The one effect I have used, that does hit playback performance and export render times, is the NeatVideo denoiser - I have the Pro version, and it is supposed to work with the iGPU, however AFAICS it is using the CPU!

HTH

John EB

Lateral thinking can get things done!

VPX, MEP Premium, MEP 2016, and earlier versions, Music Maker Premium, Music Maker 2016, 2015.

Running Windows 10 64bit on Intel i7-8700K 3.2 GHz, 16Gb RAM, 1Tb + 2 x 2Tb internal HDD + 60Gb internal SSD, + 6 x 2Tb ext HDDs, Sony FDR-AX53 Video camera, Contour HD 1080 and Sony HDR-AS30V Sports cams.

CubeAce wrote on 3/26/2020, 4:06 PM

@johnebaker

Hi John.

I was more interested in what options you had in MEPs Display options setup if you had MEP open on your NVIDIA powered monitor as in my example to see if it activated your NVIDIA graphics card or not, and then whether it made any type of difference.

This is the first time I've managed to render anything in less time than the project length at a decent quality level.

I'm sure I'll find plenty within MEP to slow down rendering times but I'm looking to see if upgrading my graphics card will allow the Intel's GPU Video Decode to work more efficiently. I still don't understand how that relationship seems to work yet or whether my card is hindering that process or not. Taking the card out of the equation in that seems to slow it down a bit as the 3D render line in the graph goes up a bit. The CPU and Intel GPU 3D render amounts seem to have little impact on export speeds. The core temperatures just rise a few degrees, along with the power consumption. Doing the same but with the project on the Intel powered monitor also clobbers the Intel GPU rendering 3D but apparently leaves little processing over for the Video Decode. The result had been much longer render times and poorer playback.

HEVC export without NVIDIA card on Intel powered GPU monitor.

HEVC export with NVIDIA card on NVIDIA powered GPU monitor.

Notice on both examples that the Intel GPU 3D graph is almost maxed out and yet with the NVIDIA card in use the Video Decode remains higher throughout with a bit of a blip during a cross-fade.

Ray.

Last changed by CubeAce on 3/26/2020, 4:07 PM, changed a total of 1 times.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

pmikep wrote on 3/26/2020, 11:02 PM

If you can upload your assets and the project file somewhere, then perhaps some of us can run it on our boxes to give comparative encode times, and test for relative smoothness scrubbing thru the time line. I use jumpshare as a free host.

Started with MEP 11, then 17, then MX, then MEP 2013, 2015, then 2016. Changed to the fast competitor after that which worked fine with my non-Intel hardware. Then bought a used Dell with an Intel GPU, just to play with MEP again. Installed MEP 2020 Plus in March 2020, even tho I don't like loosing patches if I have to reinstall after a year.

Testing on a Dell Vostro, i3-8100, UHD 630, 16 GB 2400 DDR4 (CL15), Win10 Home, heavily NTLite'd. Now with GTX-1650 Super OC'd. Added a WD Blue SATA SSD for OS, Apps, Temps and Video-In. 2 Monitors.

pmikep wrote on 3/27/2020, 12:12 AM

Regarding "Update in background." I took it that it was the timeline thumb nails that was set to update in the background as the timeline scrolls during playback.

Is this correct? Or it is something else that is updating?

Started with MEP 11, then 17, then MX, then MEP 2013, 2015, then 2016. Changed to the fast competitor after that which worked fine with my non-Intel hardware. Then bought a used Dell with an Intel GPU, just to play with MEP again. Installed MEP 2020 Plus in March 2020, even tho I don't like loosing patches if I have to reinstall after a year.

Testing on a Dell Vostro, i3-8100, UHD 630, 16 GB 2400 DDR4 (CL15), Win10 Home, heavily NTLite'd. Now with GTX-1650 Super OC'd. Added a WD Blue SATA SSD for OS, Apps, Temps and Video-In. 2 Monitors.

pmikep wrote on 3/27/2020, 12:14 AM

It seems to me that Video Preview during encode can only use more resources. Yes? (That is, better to not enable the preview.)

Started with MEP 11, then 17, then MX, then MEP 2013, 2015, then 2016. Changed to the fast competitor after that which worked fine with my non-Intel hardware. Then bought a used Dell with an Intel GPU, just to play with MEP again. Installed MEP 2020 Plus in March 2020, even tho I don't like loosing patches if I have to reinstall after a year.

Testing on a Dell Vostro, i3-8100, UHD 630, 16 GB 2400 DDR4 (CL15), Win10 Home, heavily NTLite'd. Now with GTX-1650 Super OC'd. Added a WD Blue SATA SSD for OS, Apps, Temps and Video-In. 2 Monitors.

CubeAce wrote on 3/27/2020, 3:46 AM

@pmikep

Hi Mike

It's amazing what I can forget to add in the way of information after three days of testing where I hadn't made notes of everything along the way. I have not heard of jumpshare but have looked at it. I will evaluate it and let you know but it does look promising so will let you know. Really it doesn't matter what the project consists of though as long as it's HD and only consists of a simple effects here and there, I used blanking and tracking of car number plates of cars and a few cross-fades. You only have to monitor comparative export times and smoothness from using one monitor over another making sure you checked the Display options in the program settings each time as these will be lost each time you bounce back and forth between monitors. But if you really need the project of course I will consider it.

No real idea of what Update in background really does but projects loaded slightly quicker on my machine and seemed to me at least to play back and scroll more smoothly. In fact scrolling could be done via the slider showing x99% speed. I didn't suffer any crashes or freezes doing that or even pulling the cursor across the project timeline either when using the NVIDIA powered monitor. Though to be fair in that respect using the Intel GPU powered monitor was not that much less responsive using the same settings where possible. Mainly it was render times that suffered most.

If you look at the image with the four boxes showing Program Settings then I never changed any of the audio settings or arranger speeds or anything in the 'Video/Audio' section I only ever changed the things mentioned in that image. I then did one highest quality export in MEPs own MP4 format and one in HEVC. Using the same settings for each test. I did try a few other settings but they had no impact on export time differences.

Having the preview window open and closed was also another test I failed to mention. On my system the export times were as far as I could see ( and some rendering times were very fast to react to) where practically identical.

I promise to look into Jumpshare as it looks like to be a very useful tool. Thanks for the info.

What I am really trying to find out is does a higher end NVIDIA card make a difference to export times? I personally am convinced having one makes a difference when set up as in my experiment but how much can it effect the performance if I upgrade from my lowly card?

Ray.

Last changed by CubeAce on 3/27/2020, 3:54 AM, changed a total of 1 times.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

johnebaker wrote on 3/27/2020, 5:03 AM

@CubeAce

Hi Ray

. . . . MEP open on your NVIDIA powered monitor as in my example to see if it activated your NVIDIA graphics card or not, and then whether it made any type of difference. . . .

It made no significant difference within the limits of the testing method.

On your latest test what was the source video resolution and the export setting resolution you used?

John EB

 

Lateral thinking can get things done!

VPX, MEP Premium, MEP 2016, and earlier versions, Music Maker Premium, Music Maker 2016, 2015.

Running Windows 10 64bit on Intel i7-8700K 3.2 GHz, 16Gb RAM, 1Tb + 2 x 2Tb internal HDD + 60Gb internal SSD, + 6 x 2Tb ext HDDs, Sony FDR-AX53 Video camera, Contour HD 1080 and Sony HDR-AS30V Sports cams.

CubeAce wrote on 3/27/2020, 5:31 AM

@johnebaker

Hi John.

Is your NVIDIA card connected to a monitor or ghost card?

Are both your monitors the same resolution, refresh rate etc?

Apart from the last test which used 4K material to see if there was a difference in the percentage rendering times between the use of different monitors, all projects were as stated earlier.

H264 1080 x 1920 60fps (59.940 fps for those pedantic people) variable bit rate encoded. Max Bit rate 80Mb/s. constant frame rate. They were from my Osmo Pocket mini cam.

[Edit]

I forgot the export settings.

for MP4

For HEVC

 

All projects used the same set of files and fades effects etc. Only the settings were changed within MEP after checking the settings of an export and for playback on each of the respective monitors.

I had to change where the NVIDIA card setting was placed when opening on the Intel GPU powered side as shown in the images. Each time I put MEP on the Intel powered monitor MEP lost the video card order and had to be reset.

On the NVIDIA powered monitor the settings were remembered and did not need resetting.

Closing and opening MEP only on the Intel powered side allow MEP to retain the Video output activated and Video mode settings as they were when I closed the project.

Every time MEP was switched from monitor to monitor it was closed and re-opened on the new to be used monitor. So, to be clear. End one test. move it to opposing window to close so the next time MEP is opened for a repeat test on the NVIDIA monitor (or other way around) it opens in the monitor the test is to be taken on.

That could be a useful bit of info I forgot to mention. I'm quite good at working out tests (at least in my own mind 😉), but not so good it seems on being methodical in my reporting of how such tests have been carried out. Sorry.

As I'm not using analog recordings, the base alternative mode 2 selection makes no difference to any of the tests as far as I'm aware. Maybe that could be a deciding factor as well if someone uses such files in their tests. I've no idea.

So many variables, not enough info 😅

Ray.

 

Last changed by CubeAce on 3/27/2020, 6:00 AM, changed a total of 2 times.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

CubeAce wrote on 3/27/2020, 6:01 AM

@johnebaker

Have added info requested above.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

johnebaker wrote on 3/27/2020, 8:26 AM

@CubeAce

Hi Ray

Thanks for the source video specs and export settings.

Monitors are a matched pair LG 24" IPS same frequency - 60 fps, one connected to the UHD the other the NV.

Will test further.

John EB

Lateral thinking can get things done!

VPX, MEP Premium, MEP 2016, and earlier versions, Music Maker Premium, Music Maker 2016, 2015.

Running Windows 10 64bit on Intel i7-8700K 3.2 GHz, 16Gb RAM, 1Tb + 2 x 2Tb internal HDD + 60Gb internal SSD, + 6 x 2Tb ext HDDs, Sony FDR-AX53 Video camera, Contour HD 1080 and Sony HDR-AS30V Sports cams.

CubeAce wrote on 3/27/2020, 9:43 AM

@johnebaker

Yes. Sorry for the omissions John.

I suspect less demanding export settings as well as less fps could give even shorter export times but it's that switch over of where the NVIDIA card Intel GPU goes in the Program Settings at the Display options screen that I think is making the main difference between GPU sharing and usage.

Each project however set up is going to give different results depending on the files used and settings. All I am trying to find is what is worth being in the system and how to configure it for smoothest playback and export vs quality of output times. There could even be say, effects that rely entirely on using the CPU that is slowing things down. While I have some enforced downtime and no footage to edit I thought I'd give it a shot.

The results I've got have given me the best rendering times I've had on my machine as well as smoothest playback. I do expect variations of performance from others as processor and motherboard architecture may have influences, as well as drivers, and operating system updates may also effect things. I have the latest updates I can find for everything on my setup. It could even be down to bios differences, it's hard to say. If nothing else I can now at least do real time exports at 1080 x 1920 and twice and a quarter that time for 4K when not using effects like New Blue.

At least with that, I'm more than happy and hope some of this may help others to at least experiment more.

Ray.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

Scenestealer wrote on 3/31/2020, 6:37 AM

@CubeAce @johnebaker

I feel like I have done 36 hours of testing on this one myself!

I created a 2min project using 8x crossfaded AVCHD 1080 50P clips from a Panasonic and Sony camera and encoded H.264.mts and HEVC .MP4 at the same res. and frame and bit rate as near as possible to Ray's tests. I replicated the Program settings Ray used in terms of the alternate GPU and monitor choice and in addition performed all encodes with Calculate video effects on GPU on and off.

I used Balanced encoding quality for all tests but also tried Good, Better and Best to see what effect that had. Balanced and Good gave approx the same time for export but Better and Best increased ET by 1min 15sec and I have yet to see any difference in quality between any of them. Exported file size was pretty constant.

CPU load was 20 - 29% across these tests with no video FX present.

The interesting thing is that - the HEVC encodes went through in virtually the same time as the H.264 encodes in every case, and the thing that varied between the codecs was the iGPU load HEVC 80-95% against H.264 20%.

The shortest times were with the Nvidia card selected in Program Settings> Display Settings - 1min 45sec as against 2.47 with the Intel GPU selected.

Ticking "Calculate VFX on GPU" was the only thing that brought the Nvidia load up from idle but pushed times out by approximately 1 min in nearly every test which surprised me a little.

More results to be continued tomorrow......

Peter

 

 

System Specs: Intel 6th Gen i7 6700K 4Ghz O.C.4.6GHz, Asus Z170 Pro Gaming MoBo, 16GB DDR4 2133Mhz RAM, Samsung 850 EVO 512GB SSD system disc WD Black 4TB HDD Video Storage, Nvidia GTX1060 OC 6GB, Win10 Pro 1903, MEP2016, 2020 Premium and prior, VPX7, VPX11. Microsoft Surface Pro3 i5 4300U 1.9GHz Max 2.6Ghz, HDGraphics 4400, 4GB Ram 128GB SSD + 64GB Strontium Micro SD card, Win 10Pro 1903, MEP2015 Premium.

CubeAce wrote on 3/31/2020, 7:19 AM

@Scenestealer

This is in VPX, yes?

At least this is something to do during lock-down if not a key worker😆.

Of course all of this testing gets thrown out with the bath water once various effects and more complex fades are applied but good to see base comparisons between systems. Where I see the most improvement between say Good and Best exports is when passing something like a repetitive pattern like a brick wall in the shadow areas or in sky detail of wispy clouds or in banding of the skyline when angles of view change. Also in complex areas with say lots of tree branches or faster paced movement when filming from the windscreen of my car. The bit rate of the original recording also has a lot to do with the outcome if using a small sensor. Not so much with my APS-C sensor of my DSLR.

Ray.

 

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

Scenestealer wrote on 4/1/2020, 6:58 AM

@CubeAce @johnebaker

All tests were done in MEP19.0.1.23.

I have had a close scrutiny of the picture after exporting (zooming the Preview monitor to 800x approx.), comparing Balanced against Best encoding quality in HEVC and H.264, and can reiterate that there is such a small difference between any of the tests that I could not say which looked better. As you have said Ray you would expect to see some difference in high motion high detail areas as the quality setting extends the encoders motion search (and hence the encoding time) to achieve this, but it does depend on how blurred the detail is in the original as to how successful the extended search is before the encoder gives up. In my tests Balanced took 1min 45 and Best took 3min with iGPU load 80% and 95% respectively.

So some more numbers from the same series of tests done in previous comment - all done with Nvidia GTX1060 then Intel HD530 selected in the "Display" Program setting Tab, with one of four effects - HDR, Colour correction with the colour wheel, Mercalli V4, and Sharpen All 30% - applied to random clips. Balanced encoding Quality.

Nvidia HEVC (w/o Calc. VFX on GPU) 2min 25, iGPU load 41- 56% GTX 0%, CPU 29- 56%.

Nvidia HEVC (with Calc. FX on GPU) 3min 34, iGPU load 22- 44% GTX 10- 18%, CPU 27- 47%.

Nvidia H.264 (w/o Calc. VFX on GPU) 2min 20, iGPU load 10- 25% GTX 0%, CPU 25- 57%.

Nvidia H.264 (with Calc. FX on GPU) 3min 34, iGPU load 10- 25% GTX 11- 19%, CPU 27- 46%.

Intel H.264 (w/o Calc. VFX on GPU) 3min 12, iGPU load 25- 32% GTX 0%, CPU 25- 48%.

Intel H.264 (with Calc. VFX on GPU) 4min 10, iGPU load 29- 50% GTX 0%, CPU 17- 37%.

Some notes on this:-

Video effects applied to clips can load the CPU more and reduce the iGPU load in some cases - ie Mercalli gave the highest CPU loads and the lowest iGPU loads.

Calculate Video Effects on GPU splits the load with the Nvidia card only if the Nvidia card is selected for Preview in the arranger in Program Settings but encoding times extend 25 - 35% - Why?

If the Intel HD530 iGPU is selected for Preview in the arranger in Program Settings instead of the Nvidia GTX1060 - exports with effects applied can take 30% longer, or without effects up to 60% longer. No idea why because the GPU and CPU loads appear quite similar!

Pondering!

Peter

 

System Specs: Intel 6th Gen i7 6700K 4Ghz O.C.4.6GHz, Asus Z170 Pro Gaming MoBo, 16GB DDR4 2133Mhz RAM, Samsung 850 EVO 512GB SSD system disc WD Black 4TB HDD Video Storage, Nvidia GTX1060 OC 6GB, Win10 Pro 1903, MEP2016, 2020 Premium and prior, VPX7, VPX11. Microsoft Surface Pro3 i5 4300U 1.9GHz Max 2.6Ghz, HDGraphics 4400, 4GB Ram 128GB SSD + 64GB Strontium Micro SD card, Win 10Pro 1903, MEP2015 Premium.

CubeAce wrote on 4/1/2020, 8:17 AM

@Scenestealer

Hi Peter.

I'm assuming that compression of video is similar to stills as each frame is calculated but there would be differences I assume between I,B,and P frames with only the I frames being the closest match to the original frame. Also don't discount the original file detail as a large sensor is going to record more accurately levels of light between photo-sensors. Although phones and small camcorders may use more photo-sites and combine the results to adjacent photo-sites the results can still be far less accurate than images captured by larger sensors which is why I mentioned the recording source in one of my earlier replies. When I combine motion with even not so fine detail the results are less than ideal. Fabric detail can also be lost with less than ideal light or poor camera settings that result from a lot of auto settings. Unfortunately for people who vlog or do action cam footage this is a problem as a preset setting is of no use in such situations. Loss of detail mainly occurs when adjacent photo-sites have similar values over given area sizes as you probably already know and these are the main areas that get re-compressed at lower settings. Large photo-sites collect more accurate data so values are less average between photo-sites and less compression occurs in those areas for a given compression setting.

You can see my problem at one minute sixteen seconds into my video beneath on the right hand wall with the brickwork. On the original file it was clean even though I used a small sensor. That is where I see differences and on clothing detail and why I tend to use the highest quality possible as YouTube will nearly but not always compress again on uploading.

 

When you watch on YouTube please check the quality setting is set to 1080p as at present YouTube is lowering viewing quality when set to auto. Presumably to save bandwidth at present.

 

When you have both NVIDIA and Intel working together does the Intel use dip as the NVIDIA workload rises?

I think we have to accept some effects due to the processing required is going to slow exports down quite a bit and may need rendering in the timeline to get smooth playback in the project.

"Calculate Video Effects on GPU splits the load with the Nvidia card only if the Nvidia card is selected for Preview in the arranger in Program Settings but encoding times extend 25 - 35% - Why?"

No Idea, but good question as my times dropped instead as was explained in this image in my earlier. post

Interesting comparison as your video card is much better than mine although my Intel GPU is newer.

Certainly food for further thought and thank you for sharing.

Ray.

 

 

Last changed by CubeAce on 4/1/2020, 8:29 AM, changed a total of 1 times.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

pmikep wrote on 4/1/2020, 4:45 PM

Perhaps forcing the GPU to do Effects is not always optimal? (Assumes that MEP would pick the optimal hardware (CPU vs GPU) if the box not checked.) I don't know if a "cross fade" is considered an effect, but I'll start encoding without this option checked to see if it makes a difference on my end too.

Started with MEP 11, then 17, then MX, then MEP 2013, 2015, then 2016. Changed to the fast competitor after that which worked fine with my non-Intel hardware. Then bought a used Dell with an Intel GPU, just to play with MEP again. Installed MEP 2020 Plus in March 2020, even tho I don't like loosing patches if I have to reinstall after a year.

Testing on a Dell Vostro, i3-8100, UHD 630, 16 GB 2400 DDR4 (CL15), Win10 Home, heavily NTLite'd. Now with GTX-1650 Super OC'd. Added a WD Blue SATA SSD for OS, Apps, Temps and Video-In. 2 Monitors.

CubeAce wrote on 4/1/2020, 4:59 PM

Hi Mike.

I don't think so. Even the fastest motherboard ram is slower than sending it directly through the GPU. (Assuming that's the way it works).

I'm equally sure faster ram in general would help.

The next set of tests I'm planning will be putting the project on a slower drive and exporting to a faster one. and then If I can export back to the faster drive as it is read from and then export to the C: drive (I'm using SSDs) because I don't have another M.2 drive to export to.

While the speed of a drive does not matter too much while editing it may allow faster export rendering.

will see and report back.

Ray.

Last changed by CubeAce on 4/1/2020, 5:02 PM, changed a total of 2 times.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

pmikep wrote on 4/1/2020, 5:18 PM

I will be surprised if drives make any difference. Testing will tell.

In the meantime, I am still of the opinion that someone should upload a simple 2 min video in 4K and 1080p so that we can all have the same reference for benchmarking. (Kind of like Blender does with their BMW blend file.)

Or, I posted the link for the 4K Puppies video. Maybe start a thread of benchmark results?

Started with MEP 11, then 17, then MX, then MEP 2013, 2015, then 2016. Changed to the fast competitor after that which worked fine with my non-Intel hardware. Then bought a used Dell with an Intel GPU, just to play with MEP again. Installed MEP 2020 Plus in March 2020, even tho I don't like loosing patches if I have to reinstall after a year.

Testing on a Dell Vostro, i3-8100, UHD 630, 16 GB 2400 DDR4 (CL15), Win10 Home, heavily NTLite'd. Now with GTX-1650 Super OC'd. Added a WD Blue SATA SSD for OS, Apps, Temps and Video-In. 2 Monitors.

CubeAce wrote on 4/1/2020, 6:40 PM

@pmikep

Hi Mike.

Just tried at my best settings for a fast export. Only managed to equal but not beat previous attempts. Most export times were actually slower so abandoned the export .

I have found a UHD clip that is just over 2 minutes and under 2gig . It's taken in poor light I'm afraid but could send via Wetransfer.

Do I have your email address? can't remember. If I haven't just PM me.

I'll sort out uploads for those that want them tomorrow morning.

Off to bed now.

Ray.

 

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

Scenestealer wrote on 4/1/2020, 8:20 PM

@pmikep

Forcing the GPU to do the FX is well documented in the forum as slowing down powerful systems. I think the setting is a legacy from early days of slow processors and no GPU acceleration on a lot of effects.

I don't think Puppies is a good one because it is shot with a variable frame rate which MEP does not like.

@CubeAce

Likewise something shot in low light could create more variables due to the ability of different encoding settings and hardware to deal with noise.

And rather than one clip, maybe a few smaller ones to try different effects?

I have said many times that drive speed should make little difference to export times as the bottleneck (by far) is the actual processing speed by the CPU / GPU. Consider dividing the size of your output file in megabytes by the number of seconds it took to encode and look at the MBytes per second.....

Peter

System Specs: Intel 6th Gen i7 6700K 4Ghz O.C.4.6GHz, Asus Z170 Pro Gaming MoBo, 16GB DDR4 2133Mhz RAM, Samsung 850 EVO 512GB SSD system disc WD Black 4TB HDD Video Storage, Nvidia GTX1060 OC 6GB, Win10 Pro 1903, MEP2016, 2020 Premium and prior, VPX7, VPX11. Microsoft Surface Pro3 i5 4300U 1.9GHz Max 2.6Ghz, HDGraphics 4400, 4GB Ram 128GB SSD + 64GB Strontium Micro SD card, Win 10Pro 1903, MEP2015 Premium.

pmikep wrote on 4/1/2020, 8:57 PM

Good to know about FX on GPU. So I'm back to my suggestion that Magix should offer a stand alone test suite/sniffer that either sets optimum performance options for users, or recommends what would be best. (Maybe it did this on install and I was the one who enabled FX on GPU. If so, then is there a way to force a re-sniff and reset?)

Started with MEP 11, then 17, then MX, then MEP 2013, 2015, then 2016. Changed to the fast competitor after that which worked fine with my non-Intel hardware. Then bought a used Dell with an Intel GPU, just to play with MEP again. Installed MEP 2020 Plus in March 2020, even tho I don't like loosing patches if I have to reinstall after a year.

Testing on a Dell Vostro, i3-8100, UHD 630, 16 GB 2400 DDR4 (CL15), Win10 Home, heavily NTLite'd. Now with GTX-1650 Super OC'd. Added a WD Blue SATA SSD for OS, Apps, Temps and Video-In. 2 Monitors.

CubeAce wrote on 4/2/2020, 1:27 AM

@pmikep

@Scenestealer

@johnebaker

Good morning all.

I did one last disk test when I got up. This time putting the project file on my SSD C: drive that is striped 0, (I know, no redundancy which on reflection of current crisis was a bad move) and exported to the C: drive which I think would be the default setting in MEP.

The resulting export was one full minute longer than than my previous fastest export time. Those using laptops maybe should think.

The worst export time I got was putting both project and export file to my M2.1 drive which started off really quick at 85 of disk activity but rapidly slowed to a crawl at less than 3% activity. I stopped the full export when it became obvious it was going past a half hour to export.

Hi Peter. I don't have multiple files of exactly the same length and varying resolutions to test out but willing to give out different file lengths of different resolutions and frame rates to check.

Ray.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58

johnebaker wrote on 4/2/2020, 2:09 AM

@CubeAce

Hi Ray

. . . . The resulting export was one full minute longer than than my previous fastest export time. Those using laptops maybe should think. . . . .

That is not surprising because the drive is having to switch between read and write mode much more frequently with write mode taking more cycles then a read.

My drives are arranged so the OS and programs are on C:, project files and all resources are on W: and export to the M: drive which is a small SSD drive.

The principle being that the project drive and export drive are not switching read/write modes during export.

I letter my drives in reverse alphabetical order, the OS drive by convention is C: and the missing letters from z to n are external drives or burners.

John EB

 

 

 

Lateral thinking can get things done!

VPX, MEP Premium, MEP 2016, and earlier versions, Music Maker Premium, Music Maker 2016, 2015.

Running Windows 10 64bit on Intel i7-8700K 3.2 GHz, 16Gb RAM, 1Tb + 2 x 2Tb internal HDD + 60Gb internal SSD, + 6 x 2Tb ext HDDs, Sony FDR-AX53 Video camera, Contour HD 1080 and Sony HDR-AS30V Sports cams.

CubeAce wrote on 4/2/2020, 2:43 AM

@johnebaker

Hi John.

I understand and have always adhered to normal wisdom regarding good practice from very early on 😄.

I was just interested if anything had changed. I was surprised at the speed the M.2 drive slowed down by doing both read and write.

Ray.

Windows 10 Enterprise. version 1909.OS Build 18363.720. Latest Bios update as well as latest hardware updates for Western digital hard drives.

Asus ROG STRIX Z390-F Gaming motherboard with Supreme FX inboard audio using the S1220A code. Intel i9900K Coffee Lake 3.6 to 5.1GHz CPU with 32GB of 2133MHz Corsair DDR4 ram. 1000 watt EVGA modular power supply. 2 x 320GB SSD drives striped for faster R/W times are my C: drive. 1 320gig Toshiba M2.1 drive. + x2 WD BLACK 2TB internal SATA 7,200rpm hard drives. Total 4TB. Three external WD drives for backup. NVIDIA G Force GT 1030 Graphics clock 1252Mhz Memory data rate 6008Mhz. 384 CUDA cores. Memory interface 64bit Memory bandwidth 48.06 GB/s 2GB of dedicated video memory, shared system memory 9967MB PCi Express x4 Gen3. Running MEP Premium 19.0.2.58