@simmiv

Hi and welcome to the forum.

Thanks for sharing your test results with the community. They are always interesting when carried out and reported comprehensively. A couple of my thoughts on your experiences:-

Test 1

Note: It ran fine, but produced no file. I gather this is a known bug with HW Acceleration and Main concept codec.

Correct when choosing the MC export template for basic H.264 .MP4, but if you choose an AVCHD Transport Stream .mts (Bluray) template and QuickSync HWA you will get a valid file. I have used this a lot over the years for Surround sound (or Stereo) exports and it gives good results fairly quickly.

Test 2

I also tried “Auto” and “OpenCL” in the advanced dialogue for the Main concept codec..  “Auto” defaulted to “OpenCL”, but when export is started it gives an error i.e. “MPEG-4 export moduel:Export was cancelled due to an unknown error.” So it's not possible to use those settings in my scenario..

You should not be adjusting those options if you are testing a SW only export as it is trying to enable HWA which will fail with either setting as you have found. It should be left set to "Software" and you should find this export quite fast compared to the Default Intel encoder set to Software (ie without HWA). I have seen near realtime exports with .mts files in the timeline.

The only options for those advanced settings which will work are Software, or QuickSync if you are using HWA in MC.

Test 4

Ran to completion in 0:54 minutes i.e. 74 FPS

I presume this was this with HWA selected? - and fits in with my experience that exports are faster with QS HWA with the discreet card (in my case a GTX1060) set in Program settings>Video Mode.

Test 5

Note: Here the load seems shift to the CPU and the GPU 4600 is hardly used. 

This is to be expected as the GPU is only used if QS HWA is selected, or, if you select "Calculate Video Effects on GPU" in the export template window. You may like to do a test with the latter as this can speed up or slowdown export depending on your project or computer Spec.

Test 6

Note: Here the load is balanced between the CPU and the GPU 4600, but no file is produced!  Also, using “Auto” or “OpenCL” in the advanced dialogue gave the same error as before in Test 2.

Same reason as in Test 1.

Conclusions I draw from this are:

The "Default" codec produces bigger files, but results in faster export in this combination.

Something is wrong here! There is too much disparity between the exported file sizes which can only mean the the export parameters for Format, Bitrate, etc were entirely different between the export templates / advanced settings used, for the Default and Main Concept encoders in the respective tests. For instance - If the MC encoder test was having to compress the original way more than the Default encoder then it could well take more time. You should use Media Info utility to compare the respective output files to see where the differences are and perhaps post the text view here.

Using the 5700 rather than the 4600 GPU in the "Program Setings/Display Settings/Video Mode:" with the default codec gives me the fastest export frame rate. I wonder what export frame rate others are getting for Mpeg-4 and with what settings..

See my comments in Test 4.

Also it is impossible to compare with others without sharing a lot of details about the material and effects, etc. in your test project as @CubeAce has suggested.

Peter

Thanks for the feedback. Given the many variations in setup and execution, I can well understand that it's difficult to replicate these tests.. However I've re-ran them and can now give you all the details I've used to do this test..

First, here is the GoPro file I used for the testing.. It's a short video and I didn't use any effects or fades.. Just the plain file in order to set a baseline.. Public link here.. https://drive.google.com/file/d/1wBcAN6fsAtXchRkFeeV2yFu2Y941tWvi/view?usp=sharing

Rather than clutter up this post, I dumped all the information into a word document detailing the program settings I used, plus export settings.. I also included the Media info for all the test exports.. The word document is here https://docs.google.com/a/valgeirsson.com/viewer?a=v&pid=sites&srcid=dmFsZ2VpcnNzb24uY29tfGVsZWN0cm9uaWNzfGd4OjZhNmMzMGNhZjU1YTRkMTE

In short, I ran six tests as you can see in the document..

First I disabled my internal Intel 4600 GPU and ran only with the AMD 5700 graphics card. I ran two export tests, one with the "Main concept" codec and the other with the "Default" codec..

• Test One: Only AMD 5700 GPU and Main concept codec gave about 33 frames/sec for the export
• Test Two: Only AMD 5700 GPU and Default codec gave about 23 frames/sec for the export

Once the two runs were finished and I had collected all the data into the document, I re-enabled the internal Intel 4600 GPU.. I then ran another four tests..

Two using the Intel 4600 GPU and trying both codecs.

Two using the AMD 5600 GPU and trying both codecs.

• Test Three: Both GPU's, with Intel 4600 GPU selected in Video Mode and Main Concept codec, gave about 32 frames /second for the export
• Test Four: Both GPU's with Intel 4600 GPU selected in Video Mode and Default codec, gave about 25 frames/second for the export
• Test Five: Both GPU"s with AMD 5700 GPU selected in Video Mode and Main concept codec, gave about 32 frames/second for the export
• Test Six: Both GPU's with AMD 5700 GPU selected in Video Mode and Default codec, gave about 77 frames/second for the export.

I also did a screen shot of the CPU and GPU utilization during the export. Just note that only the Intel 4600 GPU shows up in the Task Manager, due to the fact that the display driver for the AMD 5700 GPU doesn't include the WMD 2.0 version of instrumentation, it only supports WMD 1.2..

Just wondering if I upgrade the AMD 5700 if it would do better!

Rgds,

Simmi

@simmiv

Hi Simmi.

I'm sorry but can't at present participate as your video file on google drive while visible does not give me the ability to download onto my system into a protected folder for scanning and Google drive is giving a warning that it can't scan the file.

Ray.

Hi CubeAce.. I've also put it on Onedrive.. Try this link.. https://1drv.ms/u/s!AnKtiksps-5zgcsz-S3H0hQYu1Y5og?e=wQo0M1

Rgds,

Simmi

@simmiv

Hi Simmi.

Much easier and secure. Thank you.

Downloaded video 1.

Will get back to you tomorrow.

Ray.

@simmiv

Hi Simmi.

The fastest export time I've had and will likely to get on my system using your export settings for the file but the program setup I use as described in the forum thread I referred to earlier is one minute thirty five seconds for a standard MP4 codec using MEPs codec rather than the main concept codec. This I fear is mainly down to the internal Intel GPU on my processor as my Graphics card is not that great, an nvidia GT1030. You can see my PCs spec in my signature. I'm using MEP Premium.

It makes no difference whether I use the Intel GPU on it's own or the Intel with the nvidia. I don't get the choice to use the nvidia on it's own but if disabled in the bios, the PC defaults to software only encoding and much slower.

The exported file is 223MBs compared to your source file which is 241MBs.

I will have to get my HEVC codec activated again before I try that but in the past HEVC exports have always been slightly slower on my machine.

[Edit]

Should have added so my test was similar to your Test 3.

Ray.

@simmiv

Hi

. . . . Test Six: Both GPU's with AMD 5700 GPU selected in Video Mode and Default codec, gave about 77 frames/second for the export. . . . .

That is a very respectable render speed - on my Intel UHD 630 render speed is 82.5 fps using HWA and the same export settings.

IMHO Test six nails the base settings you should use.

. . . . Just wondering if I upgrade the AMD 5700 if it would do better! . . . .

This is where you need to take into account the law of diminishing returns and, IMHO, is also the appropriate time for a 'reality check', ie

• does paying, for example, 50% more give you a 50% increase in speed - in my experience it does not - you may get a 10 - 15% increase however even this is not guaranteed.
   
• when exporting projects with editing, effects, transitions etc, then the performance is going to take a hit with any graphics card.
   
• which graphics card is going to be 'the one' - and is it compatible/usable with the program.
   
• how much time is this really going to save - are there deadlines to work to?

. . . . 4th gen i7 4770k cpu . . . .

My personal preference would be to hold off replacing the AMD 5700 and look to a future replacement of the PC with a 6 or 8 core Intel processor with integrated GPU unit and put the AMD5700 in it initially - the i7 4770K is probably approaching replacement time, I was forced to upgrade from mine due to motherboard failure after ~ 6 yrs of use.

HTH

John EB

Hi Ray and John,

Thanks for trying this and giving me the assurance that my PC is the best it can be with the encoding.. I will live with it for a while, or at least until I decide to do an upgrade. Hopefully Magix do more in regards to using the power of GPU's in the future. I also note that the Main concept codec they supply is from 2012.. So no update since then, although I'm not sure if MainConcept have released an updated SDK.. If I do an upgrade I would do it on the motherboard and cpu first as my current motherboard Asrock Zextreme6 dates from about 2013. The new AMD Ryzen cpu's seem to be fast and cost effective..

Happy encoding!

Simmi

@simmiv

Hi

. . . . the new AMD Ryzen cpu's seem to be fast and cost effective. . . . .

That depends on what test software results you are looking at !

If you look at the current system requirements for MEP and Video Pro X (VPX) - they do not mention AMD or Intel processors specifically, however the important lines are:

MEP: Onboard, . . . recommended: Intel Graphics HD 520 for MPEG2/AVC/HEVC . . . .

VPX: Onboard, . . . recommended: Intel Graphics HD 520 for MPEG2/AVC/HEVC or NVIDIA GeForce 1050 for HEVC, or better *

and in the footnote there referred to by the asterisk

*An active, integrated graphics unit with installed, up-to-date drivers is required.

As you can see this points to an Intel CPU with integrated GPU.

John EB

 

@simmiv@simmiv@simmiv

Hi Simmi.

I have to agree with Johns summation.

Initially I was interested in your experiment because most people using MEP would be using an Intel CPU with integrated graphics. This would be true for a lot of people using a lot of the NLEs out there with perhaps the main exception being Adobe's offering that seems fairly happy with most modern processors but has it's own set of problems. As far as I can tell and I'm ready to be proven wrong at any time, the main use of a better nvidia card seems to be on the ability to get a smooth and responsive playback within the editor when working on a project. Some people here (Including John above) seem to be able to import what they want into MEP and edit with no need to drop the quality, or frame rates, or use proxy files to get a smooth and rapid playback. I, on the other hand, despite running an i9 processor do have problems but my graphics card is not as high end as theirs.

Ram speed may also play a part with export times but I won't know that until I've upgraded my graphics card.

Ray.

Ok, thanks for that reminder!.. As I see it, Magix are constantly upgrading it in order to make it do things faster. The focus is on two areas, during editing and playback and then exporting and coding.. Where they rely on Intel HD seems to be mainly for export and coding for Mpeg.. HEVC is now supported on Nvidia cards.. Also, the MainConcept drives provided by Magix haven't been updated since 2012, when Intel had Gen7 GPU's.. They are now up to Gen10 GPU cards and with every Gen refresh they tend to add more codec support in Quicksync.. We shall see how things pan out.. I'm not in a hurry as my rig seems to be fine after the latest tests and changes.. I mainly do 1k or 2k editing and coding.. I'm not doing 4k or 8k yet!

Rgds,

Simmi

@simmiv

Hi Simmi.

I don't pretend to know what the developers aims are but suspect the speed increase is down to them getting ready for more mainstream 8K output. Had the Olympics gone ahead this year I think I remember it was going to be recorded and broadcast at 8K for those that could use the option. It would have been a launch pad for 8K at least in their home territories. Another reason is that some may be recording at a higher quality to downsample to reduce noise and gain sharpness at lower resolutions. Either way it will demand higher specifications of the hardware for the throughput eventually. I've always been of the mind that upgrades should be for a reason and the main one is having new recording hardware. Whether it is for the newer codecs being used on such devices or their increase in resolution they are capable of. There is also the question of future proofing your current recording capability when a 4K or 8K screen becomes more normal for everyday use. How long ago was it we were using monitors at SVGA resolution and colour palettes?

We live an an age of constant and ever increasing capabilities. Who know what the next fifteen years will bring, but I do know what old footage looks like on YouTube. It makes me think.

Ray.

@simmiv

Hi Simmi

Thanks for another comprehensively presented summary of your tests.

I downloaded your file and must say that the subject matter with a huge amount of similar detail combined with movement would be a tough task for any encoder - in fact the camera's encoder was obviously struggling to capture the detail faithfully even though it was using a pretty healthy 30Mbps bitrate for HD, causing blurring in clumps of low contrast foliage. I presume your file is the file straight from the camera?

I have tested all your export scenarios plus a couple more and note the following:-

My results

• Test 1: (No HWA) Nvidia GTX1060 GPU selected in Video mode(ie SW only) and Main concept codec gave 1min 50secs for the export and file size 155MB. (yours 1min 59secs + 150MB)

Note here: - this is with the .MP4 template that you used with MC which as I mentioned in my earlier post is a different bitrate 16Mbps compared to the Default codec template which is 28Mbps, so the MC codec is compressing the footage considerably more. So I ran a second test after increasing the MC .MP4 bitrate to the same 28Mbps used in the Default .MP4 template, as follows:-

• Test 1A: (No HWA) Nvidia GTX1060 GPU selected in Video mode (ie SW export only) and Main concept codec gave 1min 46secs for the export and file size 241MB - this explains the file size discrepancy you mentioned earlier.
• Test 1B: (No HWA) Intel HD530 GPU selected in Video mode (ie SW export only) and Main concept codec gave 3min 46secs for the export and file size 237MB.

• Test 2: (No HWA) Nvidia GTX1060 GPU selected in Video mode (ie SW export only) and Default codec gave 2min 14secs for the export and file size 224MB. (yours 2min 53secs + 225MB, 2min 18secs + 67MB?? in your first tests.)

• Test 3: Both GPUs (Intel HD530) with Intel HD530 GPU selected in Video Mode and Main concept codec using (Quick Sync) HWA ( export to .mts 28Mbps ), gave 1min 34secs for the export and file size 226MB. (yours 2min 4secs + 150MB - but you used lower bitrate .MP4 16Mbps template and no HWA possible.)

• Test 4: Both GPUs (Intel HD530) with Intel HD530 GPU selected in Video Mode and Default codec using (Quick Sync) HWA ( export to .MP4 28Mbps ), gave 1min 43secs for the export and file size 237MB. (yours 2min 37secs + 232MB)

• Test 5: Both GPUs (Intel HD530) with Nvidia GTX1060 GPU selected in Video Mode and Main concept codec using Quick Sync HWA (requiring export to .mts 28Mbps), gave 1min 34secs for the export and file size 226MB. (yours 2min 3secs + 150MB - but you used lower bitrate .MP4 16Mbps template and no HWA possible)

•  

  Test 6: Both GPUs (Intel HD530) with Nvidia GTX1060 GPU selected in Video Mode and Default codec using (Quick Sync) HWA ( export to .MP4 28Mbps ), gave 54secs for the export and file size 237MB.(yours 52secs + 233MB).

  So some conclusions:-

  For me it would be good to see how the MC encoder compared to the Default on your setup on the equivalent bitrate in the Advanced Encoder Settings with Quicksync HWA enabled by use of the AVCHD Transport stream template in MC. Your Intel 4600 did extremely well when being used for Quick Sync HWA with the Default codec but could not be enabled for QS using the .MP4 template.

  Having the discreet Nvidia GPU set in the Video mode always gives me better export times and your test sample scenario reflects that.

  I tested with "Calculate Video Effects on GPU" ticked and unticked where ticked usually gives slower exports, but in this case there was no difference as there were no FX applied, the tests being basically a re encode that did not require any rendering of FX.

  Peter

@simmiv

Hi Simmi

I also did a screen shot of the CPU and GPU utilization during the export. Just note that only the Intel 4600 GPU shows up in the Task Manager, due to the fact that the display driver for the AMD 5700 GPU doesn't include the WMD 2.0 version of instrumentation, it only supports WMD 1.2..

It would be interesting to see if your AMD 5770 is doing anything during these tests. I understand that you are probably using a legacy driver and that is why it is not showing up in Win10 TM but have you tried monitoring it with the free utility TechPowerUp GPU-Z sensors tab? If you install it you should be able to select either GPU from the drop down at the bottom and can in fact open 2 instances of it to monitor both GPUs at once.

I am a little interested in the fact that your Radeon is from back when VPX/MEP could use Open CL (and Cuda) to hardware accelerate exports and wondered if it could still be possible with an early driver? You could try clicking on the button in the MC Encoder Settings - adjacent to the HWA check box - called "Driver Information" and see what appears in the box at the bottom.

Peter

Hi Peter,

That was well done of you! Interesting to see that we are getting results along similar lines.. I re-ran the MC test and used the AVCHD template and it finished in 2:06.. The export file is now 238k. which is closer to what you got as well. Yours appears to be a bit faster.. I only have the 4th gen Intel HD 4600 and the older AMD 5770 card. I tried the TechPowerUp GPU-Z, but it also fails to show any GPU utilization. I susupect they also use the WMD 2.0 api. The driver I have for the AMD 5770 is dated 3/8/2015 version 15.200.1062.1004 I gather this is the last valid driver for this card. There is another one, Crimson Edition beta driver, but from what I have seen it doesn't include WMD 2.0 support so I haven't tried it.

@CubeAce

Hi Ray

. . . . what the developers aims are but suspect the speed increase is down to them getting ready for more mainstream 8K output . . . .

Both MEP and VPX already support 8K (7680 x 4320) in the project settings, import and export - there is a caveat for export, it can not be the h.264 codec which has a maximum 4K resolution support.

AVI, MOV, HEVC all support the 8K resolution.

MEP cannot use HWA, VPX does with a suitable NVidia GPU

John EB

@johnebaker

Hi John.

Yes, sometimes I find the information on the Magix site to be a bit confusing especially when it states neither program can import 8K files in the comparison guide.

Ray.

@simmiv

Hi Simmi

I re-ran the MC test and used the AVCHD template and it finished in 2:06.

But was it showing the Intel GPU was using Quicksync for Hardware acceleration in the Export "Mixdown" window?

I tried the TechPowerUp GPU-Z, but it also fails to show any GPU utilization. I susupect they also use the WMD 2.0 api.

Do you mean it didn't show sensors for the card at all? If you mean that it did not show any GPU load then that is probably correct as there was nothing for the Radeon to render, as no effects had been applied to the clip, and it is probably not using Open CL for encoding either.

The driver I have for the AMD 5770 is dated 3/8/2015 version 15.200.1062.1004 I gather this is the last valid driver for this card.

To test if the Radeon can still perform Open CL HWA I imagine you might need to use a driver from 2012 or earlier from when the MC encoder that Magix used was current.

Peter

@CubeAce

Hi Ray

Yes they have been keeping quite about it recently.

OT:

8K really benefits from big screens ~ 80 inch or more - on a 90 inch TV you can see the difference between 4K and 8K when you are at a viewing distance that is comfortable to watch. On smaller screens, approx 60 inch or less, you have to be very close, which makes for uncomfortable viewing, to see the difference between 4K and 8K.

Commercial/industry environments, where big screens are the norm, is where I see 8K coming to the fore.

John EB

Hi Peter,

But was it showing the Intel GPU was using Quicksync for Hardware acceleration in the Export "Mixdown" window?

It defaulted to Software.. I have to go into advance dialogue for the Mc coder and I can change it to Quicksync. Doing that it finishes in 1:30.. Thus it is using the Intel card and gains in performance.. The sensor tab on the TechPowerUp GPU-Z always shows 0% for the AMD 5770. I'm thinking it's due to the fact that the driver for that card only supports WDDM version 1.3. The driver has to support WDDM version 2.0 or higher in order to show GPU utilization. Thus I can't see if the 5770 is being used. The Intel GPU goes from 4% to 17% compared with the original test, so I think using quicksync rather than software offloads some more to the Intel GPU..

If I choose OpenCL it gives the usual error (AVC-Exportmodule: The GPU-Encoder could not be initialized), so it doesn't proceed with any encoding..

Simmi