This is something that would have to be done purely from the NI side - we have no influence over how KK operates.

Originally Posted by: rockstarjazzcat

I wish you luck on any changes made moving forward.  Suffice it to say, I imagine that in any DAW, uniting midi routing out the send with graceful return of audio into a single track is going to make loads of people happier.

Exactly this is what is close to impossible to achieve today, and getting more and more difficult to achieve as time goes by. DAWs have kept moving in the direction of each plug-in being processed in its own little time- and buffer world, and synchronizing between plug-in instances has become incredibly difficult. Some hosts even completely pre-process certain tracks way ahead of time.

Consider this possible scenario:
- One plugin for MIDI events
- One plugin for audio return

The audio return plugin is buffered, and processed 30 seconds ahead of time. The midi plug-in is selected for recording, and is processed in realtime.

Unfortunately it isn't as easy as setting a "port per plug-in", since each plug-in (and instance) may live within its own processing chain - different threads, different blocksizes. 

In fact, the "Event input" plug-in already does what you want, but it also comes with a whole array of  drawbacks when used in a host that handles playback and recording tracks differently.

EDIT: The event input plug-in doesn't allow filtering channels, only ports.

When running into this issue again, please create a minidump from Vienna Ensemble Pro.exe using Process Explorer. Then please send the minidump to

https://technet.microsoft.com/en-us/sysinternals/processexplorer.aspx

(replace conhost.exe with Vienna Ensemble Pro.exe)

Originally Posted by: Daniel Stenning

Is it not possible to just add the for an "alias" name that us users can optionally just set ourselves?

I think that could be reasonable. I'm just wondering how many would actually use such a feature?

I've tried to make the presented parameter name as short as possible, while still keeping all information required to identify which plugin/parameter it controls.

If you have any ideas on how to improve this, I'd be happy to hear it!

The latest release of VEP should take care of this.

Thanks.

Hi,

thanks for your input on this matter. I believe you are referring to this Juce forum thread?

First of all, VEP is not written in JUCE. However, the issue you describe is very similar to what we're seeing - and the workaround that I have already implemented is just that - a delay (of 500ms) for loading projects upon startup. Perhaps this delay would need to be longer on other systems? Do you recall what delay value you ended up with?

In any case, this is really a bug with Kontakt itself - I think related to the modal dialog it pops when doing loading. We have contacted NI several times about this issue, but have yet to receive a reply.

Thanks,

Martin

Originally Posted by: cgernaey

I changed the Preload Size to 32768.  Right beside that, it tells me that the speed to both paths is around 250MB's.

What kind of drives are these? Usually NVMe drives should show a faster speed than this. Could you run a benchmark using https://www.atto.com/disk-benchmark/ - and paste your results?

Originally Posted by: cgernaey

I am confused at to why when I load Sychron Strings, every section is "sharing" the total voice count?  I can be sitting at the interface for Basses while Violins are playing and it shows me their current voice count usage?

So when you have each section, using all the microphone settings on a super fast NVME drive, because I hit the max voice count (using Vel X-Fade) I get dropouts long before the drive and CPU is saturated.

 

How do you get around this and actually use these strings with all microphones and velocity X-Fade?  What am I missing?  Is this just an illusion?

1. The voices displayed in the interface show the total number of voices used for all Synchron Player instances
2. The "Voice limit" setting on the "Perform" page is applied per instance.
3. You cannot deactivate streaming. You can only raise the streaming buffer up to 32768 samples, which should be plenty - unless you have some severe configuration issues with your drive(s).
4. Using ALL microphones and velocity x-fade is tremendously taxing on the system. If you really want this, you need to raise the instance voice limit.

This is unfortunately not possible. Individual plug-ins allocate memory from the system directly, and we don't know anything about it.

We cannot reproduce this here. Any chance you could make a screen capture video explaining what you are experiencing?

Originally Posted by: Thurisaz

Last night I've sent crash dump files to the Support e-mail. You can check them. :-)

Could you tell me the case number please?

Originally Posted by: Dewdman42

Just wanted to follow up to this.  After upgrading my system from Sierra to Mojave, performance with VEP7 is not only just as good as VEP6, its significantly better.  All the problems I mentioned above went away when I upgraded my Mac to Mojave.  

To clarify - the higher performance in Mojave will come with the next update of Vepro 7.

Kontakt doesn't share any memory between instances AFAIK, so the memory meter you see inside of Kontakt displays how much RAM is used for sample preload by that instance alone.

VSL's instruments can share preload memory between instances; thus, loading up a second instance with the same instruments will only add a tiny bit of overhead. The memory meter inside VSL instruments therefore display the total preload size used by that specific instrument, for the entire process. 

Thanks for the reports everyone. We have located and fixed a bug in the VST2 version of the VEPro plugin, that causes problems with tempo.

The next update (very soon!) will contain the fix.

A complete rescan should only happen if the cache folder has been deleted (perhaps a 3rd party uitility clearing system caches?) or if a manual rescan is triggered from preferences.

The plugin cache folder for VEP7 is:

/Users/YOURUSERNAME/Library/Caches/VSL/Vienna Ensemble Pro/cache

VEP7 does paint a few more things to screen, and yet it still performs better than VEP6 when using Native or HiDPI display modes. It might be, that the extra blitting stirs something up when using scaled modes, causing the CPU usage to tick slightly higher*.

* I have not yet tested anything myself regarding scaled modes.

Originally Posted by: Dewdman42

This is a longer discussion I don’t feel like having here but hidpi does not require your native resolution to be double what the hidpi resolution is. Even Apple branded retina monitors have modes where it’s getting closer and closer to the native resolution but still applying hidpi technology to improve the sharpness in corners and curves. In preferences the normal hidpi setting is 2:1 it you can say “smaller text” or whatever they call it in easy terms and then it will be using a larger internal canvas that is larger then the native resolution of the actual monitor to render and then it will be reduced back down to the monitor resolution by the video card. That’s is how hidpi and retina always works, you are not constrained to 1920 resolution in hidpi.

I'm sorry, but this is not correct.

The only modes where applications fully render to pixels are the 1:1 and 1:2 modes (native and HiDPI respectively). Using any scaled modes in between, applications will render at their native 1:1 resolution, then be scaled UP by the OS. The result is blurrier than a direct mapping, and will additionally put more strain on the system. 

Originally Posted by: Dewdman42

No you can run 4K monitor at larger hidpi settings then 1920. Mine is at 3200x1800 hidpi and looks great, but slow on vep7

Please be aware that scaling will occur at any resolution which is not 1:1 or 1:2. I have not tested anything at these scaled resolutions, since I really don't recommend using them, but I would believe that it performs worse.

Originally Posted by: Dewdman42

Try some larger hidpi settings. My system is performing much worse in hidpi mode non minimized, and vep7 is worse then vep6 in this regard by 6%. My tests were using my normal working resolution of 3200x1800 hidpi. You will need a 4K monitor and decent video card.

The fact that vep needs 15% of cpu over and above all the internal mixing engine to display meters and articulations on a Mac is disappointing. I hope vsl will look closer at that in years to come.

As I wrote in the PDF, I was using an AMD RX570 GPU and a 3840x2160 display for my tests. In HiDPI/Retina mode, this is called 1920x1080 HiDPI. I am not aware of a display that handles 3200x1800 HiDPI, which would be a native resolution of 6400x3600 pixels. Perhaps you mean 1600x900 HiDPI?

As I also wrote before, an application will always require processing power to perform updates of dynamic information on the screen. The more that's updating, the more processing power is required. I believe that VEPro is actually pretty efficient regarding this, many applications perform much worse when continuously updating so many things. We are slightly bound by the limits set by the GUI framework we're using here (Qt), but so are most other applications as well, except for Logic.

First of all, let me reassure you that we take any reports about CPU usage seriously, since we want VEPro to be as efficient as possible.

A busy GUI code, updating meters, peak displays, MIDI activity and other things, will always use some CPU power. The cost for updating such things tend to be a bit higher on MacOS than on Windows, due to blitting being a more expensive operation on MacOS. I don't find the CPU usage of VEPro as such to be in any way abnormal for the amount of work that is done.

In order to shed some further light on this topic, I created a performance comparison project, trying to isolate the impact of GUI updates on project performance. The results from this test, and the 16 possible test permutations, can be found in the linked PDF.

If anyone wishes to look further into the measurement data, or reproduce these findings - I am also enclosing the project files

Test Results (PDF)
Raw data and project files (ZIP)

Thanks,

Martin