Hi

As OBVideos has highlighted, Magix has tailored the software to make use of Intel Quick Sync for accelerating playback and encoding of H264 &HEVC via the iGPU in Intel's 3rd gen and later "i" series processors.

A few experienced members of the forum run workstations without any Discreet video card, with virtually no loss of speed or functionality. Personally I prefer to select my Nvidia for timeline playback as I find it runs a little smoother than the iGPU in my i7 6700K.

The CPU does a lot of the "heavy lifting" in most playback and SW encoding and the iGPU kicks in when choosing Quick Sync HW encoding during export. A discreet Video card adds very little so spend the money you save on the best CPU that you can afford. The new i7 7700K would be my pick and I think 16GB of memory is plenty.

SS

1800X and RX 480 + 2X8GB 3200 CL 14 here and working very smooth even with 3rd party plugins such as Neatvideo v4, Titler Pro 4, NewBlue effects etc, I use VPX6 though but I tried VPX9 and it was even better but I'm not ready yet to upgrade from VPX6

I recently did some research into this and I found the following:

decoding and encoding is normally done using intel quicksync. This offloads the work to a special circuitry on the chip so the actuel cpu itself remains free to do other things. I recently bought a laptop with intel 7700HQ cpu (and nvidia gtx1050 and 16GB dual-channel memory which is indeed plenty) for this. Works well when i'm connected to AC power and enable the high-performance energy profile in windows. Playback of 4k in the timeline is smooth and HD editting is a breeze. CPU load of about 17% when playing back 4K video on the timeline. If your budget forces you to choose between a high number of cores or a high clock frequency it is better to go for the higher clock frequency. A high number of cores does not improve encoding or decoding speed much with quick sync, but a higher clock frequency does.

However, not everything depends on quicksync. When doing lots of image stabilisation, effects or using plug-ins, these typically put a load on the cpu. Having a powerfull ryzen cpu might work well in these cases and apearantly igniz-krizalid is having succes with it. A powerfull CPU appearantly can compensate for the lack of quicksync. Besides, a big cpu can make your whole system more responsive and can be used for all kinds of things. (But as there are no laptops with ryzen yet, I didn't choose that one).

Graphics card are used only for specific effects such as color correction and resizing. The manual states that "especially nvidia cards are recommended but others might work aswell". ryzen and amd cards play well together. I suspect they use opengl for these effects. No need to spend much money on a graphics card. It is better to verify you can connect all the monitors to it you want.

The problem with OBvideos setup is:

• the cpu doesn't have quicksync
• It has 6 cores 12 threads, but they run at a low max clock frequency (2.5Ghz)
• Oversized graphics card

vdx has its roots in a program which is targeted at the avarage Joe using the most common types of hardware in the market. I.e. an affordable intel cpu with an affordable nvidia card. The program appears to be optimized for just such a setup. Throw a fancy server cpu in and it breaks down.

Hi @mschagen

I agree with most of what you have (very well) said especially the last paragraph. It pays to keep this in mind when building a system.

If your budget forces you to choose between a high number of cores or a high clock frequency it is better to go for the higher clock frequency. A high number of cores does not improve encoding or decoding speed much with quick sync, but a higher clock frequency does.

Just wondering what information you based this on? It is easy to generalize in these discussions and every project is different depending on material and effects. If the CPU is underutilised during Quick sync HW encoding then the CPU is irrelevant but often I see the CPU over 100% utilisation (due to overclock) processing an effect like Mercalli Pro, and the GPU doing very little. In this instance I believe more cores or more clock cycles would help.

Also I should mention that my comment about spending more money on the CPU than the graphics card was not referring to very high end processors, but more the mid to higher i7 6700 or 7700 units.

There has been quite a lot of discussion about the Ryzen in the German forum and I think they will still be beat for export encoding performance by the Intel iGPU and Quicksync, but having said that I find I can get better than realtime encoding in a complex project using the MainConcept software encoder, not that much slower than the HW QS encode. SW encoding with the Intel encoder is a joke - like about 5 times slower.

Also QS is only really fast on an encode with little in the way of effects applied as they kick a lot of work back to the CPU. Remember also that QS is only advantageous with H264 / 265 material during preview and export encode - Prores .MOV uses all CPU save for parallel GPU computing on a few effects. Actually I am still not sure that the Quick sync block is considered part of the GPU or part of the CPU and as such monitoring GPU loading in a utility like GPU-z is showing the whole picture.

The RX480 is an interesting one and I find it is quite favoured with the Vegas pro community. I wish I could have tried it when I bought my Nvidia card as I have long suspected it may work better with the Open GL accelerated effects used by New Blue, despite Magix saying they have optimised their effects for the Nvidia pieces.

When you say your rig plays 4k smoothly at 17% CPU do you mean 1 track with audio no FX? of what - H264 or H265? at what fps - 25 or 50P? on what size preview monitor 4K or Full HD?, Full screen? So many variables it is difficult to compare.

Peter

@Scenestealer

In the German forum is a thread https://www.magix.info/nl/forum/ein-allgemeines-posting-zum-thema-pc-kaufberatung-fr-vdl-pro-x--1189593/ where a user has done extensive research and some tests to see how vdx responds to different hardware configurations. He found that the number of cores has no significant impact on quicksync performance, but clock-frequency has.

Ofcourse anything other than quicksync decoding/encoding will indeed benefit from additional cores. This depends on the actual implementation of e.g. effects or image stabilisation. Sofar I find that effects and image stabilisation do use all cores I have (4) effectively. So you are right there. I just rarely use such effects.

The 17% is just playing one 4k 60p H264 clip with sound without any effects using the default small preview screen. As soon as I start using a simple collor correction, cpu usage jumps to 80%. Using a full screen preview leads to jerky playback.

That's indeed the problem with drawing conclusions from other peoples experiences. Performance can be very sensitive to all kind of settings, configurations, source material and even your style of editting. For evaluation it is therefore best to download the trialsoftware and see if you can make the program work for your usage. And the program itself can also behave oddly with respect to perfomance. E.g. playing back HD h264 hardly stresses the cpu, but I recently used some old SD dv-avi files (which to my knowledge are MP2) and they caused a cpu load of 100% upon playback which is ridiculous. With older versions of the software it was at most 20% on much weaker hardware. I suspect this is a bug where they broke something with respect to support for an old format. Playing back proxy files for a HD H264 file generates a higher cpu load (60%) than playing back the native H264 on my system. I suspect quicksync rendering is not supported for the proxy files, which is somewhat silly considering the whole point of proxy files. I also recently discovered that when removing an effect from a clip, the cpu load when playing back that clip remained 80%, as if the effect was still being calculated. The software is not free of bugs and never will be, that I have learned over the years. This might also cause perfomance to vary from version to version.

Hi @mschagen

Thanks for the link to the German forum but I have been following Akebono's sticky post for some time. He has done some good work there and states things well.

He found that the number of cores has no significant impact on quicksync performance, but clock-frequency has.

I would assume this is because the Quicksync hardware block gets over/underclocked as well, and of course because most of the work is being done on that separate hardware area on the CPU die, the number of cores would not be much of an advantage.

I am also not sure that just lowering the Quadcore Clock multiplier down to the level of the 2 core processor he is comparing to is entirely valid for this comparison. I have no reason to think that Magix does not utilise all the cores effectively, as it does so in my experience. Akebono refers to the next new range of processors that have gone to many more cores (24) and moderate clock speeds so someone thinks more cores is the way to go, but anyway if the dicussion is about QSync then there is only one block in the respective CPU's.

And the program itself can also behave oddly with respect to perfomance. E.g. playing back HD h264 hardly stresses the cpu, but I recently used some old SD dv-avi files (which to my knowledge are MP2) and they caused a cpu load of 100% upon playback which is ridiculous.

That is ridiculous and you cannot blame the program for that, as it only uses about 6% in task manager here. DV-avi is not a complex compression and not like MPEG formats which use Interframe GOP type compression, it uses intraframe which is more like jpeg compression on every frame and is consequently easier to decode / encode. It cannot however make use of the Quicksync block for decoding during playback so I would expect it to use similar resources to an unaltered 1080 H264 clip, which it does on my setup.

The 17% is just playing one 4k 60p H264 clip with sound without any effects using the default small preview screen. As soon as I start using a simple collor correction, cpu usage jumps to 80%.

I only get a few percent increase in CPU usage on my installation if I add a Magix colour correction ( which is GPU HW accelerated).

I suspect quicksync rendering is not supported for the proxy files, which is somewhat silly considering the whole point of proxy files.

No, because the proxy files are created in something other than H264 / 265 which are the only compressions that QS supports. I do agree that there is something not right with the default proxy mxv file format because it max's out my CPU on the timeline. This may not be the whole picture however because the whole idea of a less compressed format is to reduce processor load when effects and transitions are being processed during preview. Choosing another Custom format like MJPEG or an Intra frame codec in the Proxy settings (Alt+R) gives much lower CPU usage here.

I also recently discovered that when removing an effect from a clip, the cpu load when playing back that clip remained 80%, as if the effect was still being calculated.

I presume you are talking about a New Blue Effect and this does happen unless you reset the effect or Rt click the effect in the Key frame FX window and select Delete Effect. Unticking here will show the preview with the effect removed but the load will remain untill it is deleted.

It is very easy to blame Magix for the above faults but often the quirks are system related or in the case of a couple of the above, could be limitations of New Blue's Plug ins.

Peter

We already went well off-topic in this thread, but just to add a few remarks:

That is ridiculous and you cannot blame the program for that, as it only uses about 6% in task manager here.

Well, I just discovered that this problem occurs when I use specific resolution presets for the preview window e.g. PAL half 4:3 which equals 384x288. When I resize the preview screen to 385x289 the cpu usage drops to the 6% you indicated. Besides, the aliasing shown by the preview monitor also disapears. So I still think it is a bug but it only shows up under very specific conditions and is not related to the actual decoding but to the rendering of the preview monitor.

No, because the proxy files are created in something other than H264 / 265 which are the only compressions that QS supports.

quicksync should support decoding of a whole range of formats according to this information, including MP2. https://en.wikipedia.org/wiki/Intel_Quick_Sync_Video. This doesn't necesarily imply vdx will use qs for all these formats also and mxv might be something exotic which can't be pushed through the qs pipeline. However I recently found a post in the german forum suggesting that the poster found a performance degradation between the 2016 and 2017 versions of vdl with playing back mxv files (which is the default for proxy files).

https://www.magix.info/nl/forum/hohe-prozessorauslastung-beim-abspielen-von-filmen-vdl-2017--1192775/#ca1331855

The degradation apearantly is confirmed by support.

As the quality of the program improves, the conditions under which bugs appear become more rare or specific which explains why other users don't experience it or why support can't always reproduce it.

Well maybe we stray but I guess we are still within the realm of "Configuration"......

When I resize the preview screen to 385x289 the cpu usage drops to the 6% you indicated. Besides, the aliasing shown by the preview monitor also disapears.

I have always been cautious from what I have read, to keep the preview screen a multiple / divisor of the actual clip, resolution like 1/2 or 1/4 HD, and to keep the aspect ratio the same. This on the basis that the recalculation for the display is simpler and puts less load on the sub system.

After the last VPX9 update I was getting some strange artifacts (shimmering and jittering) in the preview monitor which I cured by updating the Nvidia drivers from Nvidia's website. I think I had the most recent one from Gigabytes site but it was a few versions old.

No, because the proxy files are created in something other than H264 / 265 which are the only compressions that QS supports.

quicksync should support decoding of a whole range of formats according to this information, including MP2.

Sorry if my statement was a little "loose" but I was referring to VPX / MEP and that is more based on the fact that I have only ever seen Magix refer to HW acceleration at the same time as mentioning H264 / H265. Also in my tests Export encoding an unaltered MPEG2 to MPEG2, whilst showing 10 - 20% activity on the iGPU, is always slightly slower with HW accn. selected. iGPU monitoring during preview of MPEG2 material shows very little load change with Preview HW acceleration selected or deselected in the Program settings.

the conditions under which bugs appear become more rare or specific which explains why other users don't experience it or why support can't always reproduce it.

Or if they can then they decide whether they see enough instances of it from consumers and push it down the priority list. I have, and others have experienced this with a problem with the HDR effect on a Win10 machine without a discreet graphics card.

Here is my neophyte comments/report:

two years ago I went into a local electronics store and put together a desktop system based on a great motherboard, Intel i7-4790K @ 4. GHz processor and a Nvidia GTX 960 graphics card. I was really happy, the computer did almost everything I asked it to do fast including Adobe Premier.

But it always ran slower with Video Pro, no matter which version. No hardware acceleration.

THEN, two days ago I bought a brand new MSI gaming laptop with i7-7700 and Nvidia 1070. A stock machine, SSD with 2 terabytes disk too.

I just ran a Magix Fastcut video with same 1080P drone video clips on both machines.

Desktop took 5 minutes and 48 seconds to create the 60 second movie.

My new laptop...28 seconds total!!! And it did use hardware acceleration! I I am not a computer expert. I use Video Pro because it is easier and faster to work with the few times a week I use it versus Adobe premier.

I am stunned and blown away.

So now I have to wonder if I should spend the money to buy a newer Nvidia 1080 card for the desktop or is it some different hardware configuration in the laptop that made it run faster?

I miss the old days of cutting and splicing 16mm film.

Hi

. . . . . I have to wonder if I should spend the money to buy a newer Nvidia 1080 card for the desktop . . . .

Why not try pulling the GTX 960 and test the i7 integrated graphics with HWA?

HTH

John EB

John E Baker:

Great suggestion. I pulled all my manuals out of a box last night and see I have jumpers on the motherboard, one of which disables the on-board video. I also now notice that my laptop has both video systems on.

I think you are correct in what you were suggesting and it could be by just turning on the on-board video, Fastcut and Video Pro will work with acceleration.

Thank you, Joe

y John Baker:

I updated the BIOS on my Sabertooth Z97 motherboard and then turned on integrated graphics.

Ran the movie creation again and it shaved a minute off of my processing time and also actually said using hardware acceleration.

The service department at the store said they may be willing to drop in a 1070 or 1080 into my system there in the store and see if it makes a difference.

And of course I have now updated every darn driver in my desktop computer, so there is some value in all I did today.

Thank you, John, for a nice piece of advice.

How would I confirm whether VPX is using Hardware Acceleration?

Also, how can I tell if it is using my integrated Intel graphics (i7 4790k) or my NVidia card? (Both are enabled).

Hi Craig

When the window opens after your encode starts you should see "Mixdown (Hardware Acceleration)" at the top. With a straight re-encode to the same specification Mpeg4 H264 movie file with basic effects using the "standard" Intel encoder you should see a speed of 4 or 5 times greater than "Software" only.

HWA only works with the integrated GPU (using both Intel and Mainconcept encoders) and you can see this if you open the sensors page of the free utility "GPU-Z" where there will be a marked increase in iGPU "load" compared to a SW encode.

If you tick "Calculate Effects on GPU" at the bottom of the VPX Export Template, you will see some additional activity on the Nvidia GPU when a clip with effects is being rendered, if you select the Nvidia card at the bottom of the GPU-Z sensors window. You will also see a concurrent reduction of GPU load on the iGPU if you open GPU-z twice and observe both GPU's with the above condition. Without Calculate effects on GPU selected you should see next to no activity on the Nvidia GPU.

Peter

Thank you, Peter.

So, just to be extra clear:

In my scenario where I have an Nvidia GTX 950 and the internal Intel 4600 built into my Intel 4790k cpu, I should set the Video Mode on the Display Options page to Intel HD Graphics 4600, rather than my discrete video card?