Horrible VI performance when recording in 2019.10!

[TNM]

Originally Posted by musicman691

TNM:
Is there any real, legitimate reason you work at such a low buffer?

Well...

that's a fair question...

I am old.. OK.. I am younger than you but I am sure we grew up with the same analog training and workflow..

I find it very hard to play well if the latency is too high for the VI.. This is why i love analog synths in an analog mixer cause they are instant.

My main issues are with bass and drums.. of course with chords I can get away with it and just nudge them, and in many cases, a decent quantize will help with the other stuff..
I know my analog synths going into UA console with the effects chain I use is about 4ms RTL (2.3 + the effects).
And at 128 buffer, the apollo is 3.9ms output, with the output latency being what matters when playing VI's.. so.. yeah.. I currently have my externals and my VI's playing at about the same latency when I choose 128 in pro tools.
But I can feel it.. I really can. So for me, lower is better.

That said, this is something I can live with, and have lived with using a 128 buffer in PT for 4 years..

But now things get more complicated which is why i am always doing testing.. bear with me but you wanted an explanation.

I HATE the apollo ecosystem and always have.. in an ideal world for me, i'd be using PT's mixer as a replacement for my analog and apollos. I'd be restricted to 32 inputs even with ultimate software and refuse to buy into HD hardware as it's more expensive than the apollos which is just silly...

Anyway... still.. considering I use 3 DAWs overall, with PT and Logic being the main two..and PT being number 1, Logic mainly used for collab and older projects...and the reason I keep testing it cause if I could run 32 channels into pt at 32 samples, and put all the monitoring fx I need, then I could get out of the UA ecosystem and it's ridiculously priced plugins and separate console app.
If I arm 32 audio tracks in PT at 32 samples, the cores are ready to go nuts.. the point i even add one monitoring reverb, the all spike to a very high load.. see in Logic I can add EQ and comp and up to 6 monitoring reverbs at 32 samples, and that was on my old 4 core macbook! I only need two monitoring verbs but you get my point cause in PT at 32 samples I can't really even do one.
And then the next issue happens.. I need to play VI's along side the external stuff playing back.. So if i have a buffer of 32 to use PT as my main mixer hub, then arm a VI... well, you get the drift.. it basically can't do it.

Yes, for VI's only, 128 is ok.

For a complete mixing desk handling all external inputs.. especially since I work usually at no higher than 48K (I am simply making music after all that ends up on streaming services so why go higher?), 128 is WAY too slow.. the total I/O roundtrip is around 9ms for a good interface at 128.. You have to work at 32 if you want a respectable roundtrip to use the DAW as your actual mixer for all sources, but then the DAW has to also be able to cope with things one might need to do, like, record a VI!
Look, I could stretch it and go to 64 buffer, which is between 5 and 7ms total RTL depending on the interface, but even then, arming a VI in pro tools has the same behaviour as 32.. so...having 32 channels being monitored into pro tools and then wanting to do other stuff.. forget it.. not at that low buffer.

In Logic, I can arm 50 channels, my complete apollo + adat i/o, at 32 buffer (incl digital), run as many monitoring effects as I need, and still arm any VI and play it.. and never change the buffer.. that's the difference.. but I don't want to use Logic, I want to use PT..
Can you believe in 2019, in Logic there is STILL no simple selection slot for midi input per channel? Just an example. I have different keyboards I want playing different externals and VI's, with Logic anything connected will trigger the highlighted track which drives me insane, and the environment work arounds are cumbersome and buggy.

I don't know if you have a large amount of external gear or not, but I like my template to have everything available at the click of a track, ready to play.. otherwise I can lose inspiration.. which is why I don't use patchbays.. I like everything connected "live" at all times..

Anyway, for now, the choice is clear, I have to stick with Apollo for external inputs, and use a 128 buffer to play VI's and just leave it at 128 permanently till something changes if it ever changes. Maybe the new mac pro would allow me massive external monitoring in PT at 32 buffer at all times, with 2 verb busses, 2 delay busses, chorus, and EQ/compression per track (good stuff like slate VMR). Ideally it would be 64 ins, but I guess Hell will freeze over before Avid do that LOL!

[musicman691]

TNM:
Thanks for the explanation. I don't have all that much left in outboard analog gear. Mostly vi's. But yeah - I came up with the old tape-based stuff decades ago. But this old dog has learned to work with digital and adjust for latency. Being a drummer as just one of my instruments I can play ahead or behind the beat so latency doesn't bother me.

My brain has learned to automatically adjust. Especially when playing a pipe organ - you want to talk about latency? We're talking seconds here between hitting a key or keys and getting sound. So even 512 or 1024 samples in PT ain't no big deal.

Have no need to record arm 50 tracks at a time seeing as how I can only play one at a time.

[ejinbc]

TNM

I do a lot of live recording with midi triggered drums (Kontakt Abbey Road kits with internal effects - stereo out), and keys (Kontakt Pianos, Strings, Brass, and Omnisphere organs and synths - multi channel out). Multiple midi musicians with live guitars, perc and bass.

First, I am on a PC. Nonetheless, some observations could be relevant - if you have considered all this please ignore. I run a MOTU 1248 interface at 64 buffer 24bit/96khz.

I have found it is easier for a computer to handle shorter sample streams than smaller sized buffers - i.e. a 64 buffer at 96 khz will run better than a 32 buffer at 48 khz. Even though they have the same latency ! High sample rates are more for latency minimization than sound quality for me.

Further, on a PC with MOTU drivers, I found the processor affinity trick necessary to achieve really stable (3 hrs recording session) operation - assign PT 1 less cores than your system has - this seems to allow the operating system to run the MOTU driver on the free processor which significantly decreases processor spikes in the PT meters. Might be difficult on a Mac.

All of the common BIOS optimizations and a small overclock are also necessary on an Intel 8700K system running Win 10 32 G Ram.

Cheers

[Ben Jenssen]

Originally Posted by ejinbc

assign PT 1 less cores than your system has - this seems to allow the operating system to run the MOTU driver on the free processor which significantly decreases processor spikes in the PT meters. Might be difficult on a Mac.

Not difficult at all. At least not on PT10 on Mac, which is what I have. You can select the numbers of cores for PT in prefs. Using one core less than maximum is a well known optimization thing.

[TNM]

Originally Posted by musicman691

TNM:
Thanks for the explanation. I don't have all that much left in outboard analog gear. Mostly vi's. But yeah - I came up with the old tape-based stuff decades ago. But this old dog has learned to work with digital and adjust for latency. Being a drummer as just one of my instruments I can play ahead or behind the beat so latency doesn't bother me.

My brain has learned to automatically adjust. Especially when playing a pipe organ - you want to talk about latency? We're talking seconds here between hitting a key or keys and getting sound. So even 512 or 1024 samples in PT ain't no big deal.

Have no need to record arm 50 tracks at a time seeing as how I can only play one at a time.

1024 would be around 48ms roundtrip when you take driver latency into account, and that's for the best TBolt interface.. USB could be upwards of 60ms..

Unless you mean just to play VI's? Then at 512 and 1024 with a TBolt interface (since input usually has the most latency) you'd probably be looking at 12ms out for 512 and 24ms for 1024.. WOW you are tolerant.

There is one problem.. even if one adjusts for this for Vi's and even external synths, the moment you want a vocalist or guitarist to record you are screwed. I have never ever met one of either who would tolerate that sort of latency.. and if you have a busy project.. back to square one.. the moment you arm at low buffer, issues..
The workaround then would be to bounce your song so far to a stereo file and deactivate all other tracks till the new audio parts are recorded.. doable but that makes me feel like 2005 again.
Anyway, part of my point is, that reaper, Logic, and even Cubase to an extent, can arm multiple tracks at 32 buffer and rock away..Logic can even do it at 96K at 32 buffer, so it simply means PT could do it too with the right coding.. I wonder if they purposely hijack it to sway people towards HDX for that sort of latency? I mean anything's possible, but even I doubt that one LOL.

let's put this all aside though for a minute musicman..

My main issue now is that PT has lowered in it's efficiency performance since 12 and 2018.. Let's not think about even ONE armed track, and let's not think about low buffers at all.
When Pro tools is playing back tracks that are not monitor/record armed, it plays them back on it's higher hybrid buffer of 1024 samples. basically, that was the whole redesign of V11 for better performance (and aax to take over rtas). I am saying that projects that played back previously but were hammering the cpu can not play back on the same computer in 2019.. I have even set up a triple boot on the macbook now (i decided to keep the quad 2015 one as it's not worth much and i can keep installing junk to test on it).. Sierra, High Sierra, and Mojave (remember i can't test the 16" yet cause of Catalina).

PT 12 on sierra and 2018 on high sierra, gives me about 25 to 30% more plugins for PLAYBACK than 2019 on Mojave which is actually recommended for it. I also have a dual install of 2018 and 2019 on the sierra partition and once again 2018 is winning.
This IS a problem and has nothing to do with low latency.

I explained it all in my cpu post but i admit it was a super long read so i understand if one didn't..

So i'll TLDR it now:
in PT 12.8.4 and OS 12.12.6, out of a possible 800% cpu for an 8 logical core macbook, PT is able to utilise 700% of real cpu and NOT stop playback at ALL.. no glitches, nada. The OS and the interface get a bit slow during playback cause the CPU is hammered so hard.

in PT 2019, I can get a MAXIMUM 560% if I am lucky.. PT 2019 is not able to make use of all the power of the processor.. that's a huge drop..

I have replicated identical working conditions/plugins/projects.. I take this sort of thing seriously.

I can guarantee you, if you have a project in PT12 that was literally at the edge of not being able to play back, but DID, that 2019 would not be able to play it, not even for a second (on the same computer) unless you went to a more powerful mac. 2019 is overloading when the OS and PT interface are still at full speed and snappy, which also tells me there is so much cpu untapped.. not the pro tools meter, real cpu measurements.

I make sure the exact same version of plugins is installed on all OS and PT versions.. there are no variables other than the OS and PT version.. which is why I also installed 2019 on sierra (the minimum OS it can use) to prove that 2019 is the issue.
PT 2018 on High sierra also outperforms PT 2019 on high sierra.

That means you really are losing, say in a project with 16 VIs, 5 or more VI instances.. and for effects plugins.. well in the dozens..

If PT can just get back to the efficiency on *playback* tracks that it had 2 years ago, I would rather THAT, and live with the bad low buffer performance.

PT 12 beat Logic in VI performance for playback (this was in my VE PRO topic where i realised that PT on it's own could do more than any daw or ve pro combination on the same machine).. and now logic is 30% ahead.. 30 percent! I mean that drop is a LOT!

I think some people answering have confused my points, like Daryl for example.. I am simply talking about the ability of how many VI's and effects PT 2019 can play vs PT 12/2018 on the very same computer.. it's markedly lower.


Also

[TNM]

Originally Posted by Ben Jenssen

Not difficult at all. At least not on PT10 on Mac, which is what I have. You can select the numbers of cores for PT in prefs. Using one core less than maximum is a well known optimization thing.

That ability is long removed and you can't just simply do it in the OS like you can with windows. PT 11 removed that ability.

As I said, PT 12 and 2018 did not have the playback track issue, with performance that beat every DAW on mac.. ALL versions of PT since V11 have suffered from the 32 and 64 buffer track arm issue.. but now we have moved on to PT 2019 losing a lot of efficiency since previous versions.. I really wish I could change the topic title! Maybe I can and missed it, will try again.

PS I think if i explain to avid tech they can fix it.

[TNM]

Originally Posted by ejinbc

TNM

I do a lot of live recording with midi triggered drums (Kontakt Abbey Road kits with internal effects - stereo out), and keys (Kontakt Pianos, Strings, Brass, and Omnisphere organs and synths - multi channel out). Multiple midi musicians with live guitars, perc and bass.

First, I am on a PC. Nonetheless, some observations could be relevant - if you have considered all this please ignore. I run a MOTU 1248 interface at 64 buffer 24bit/96khz.

I have found it is easier for a computer to handle shorter sample streams than smaller sized buffers - i.e. a 64 buffer at 96 khz will run better than a 32 buffer at 48 khz. Even though they have the same latency ! High sample rates are more for latency minimization than sound quality for me.

Further, on a PC with MOTU drivers, I found the processor affinity trick necessary to achieve really stable (3 hrs recording session) operation - assign PT 1 less cores than your system has - this seems to allow the operating system to run the MOTU driver on the free processor which significantly decreases processor spikes in the PT meters. Might be difficult on a Mac.

All of the common BIOS optimizations and a small overclock are also necessary on an Intel 8700K system running Win 10 32 G Ram.

Cheers

Windows (aka bootcamp for me) works great with PT on the very same mac.. it's an OSX/PT issue.
NO processor affinity trick here and even if the os can be hacked via terminal, i'm not the one to do it.

Now, it's also a total opposite to your situation with mac and buffers/Sample rates.

It scales linearly on mac.

If you work at 44 or 48K., 32 or 64 buffer give the issue..
If you work at 88K or 96K, 64 or 128 buffer give the issue.. and since 256 buffer at 96 has similar latency to 128 at 44 there is just no point for stero music that is intended for ipods and the internet..
If you work at 192K, only 512 buffer or above is usable for large track count monitoring and playing VI's in realtime.

But as I said, since that just won't be fixed.. let's talk about the massive performance loss for overall plugin instance capability in 2019 vs previous builds.

[musicman691]

Originally Posted by Ben Jenssen

Not difficult at all. At least not on PT10 on Mac, which is what I have. You can select the numbers of cores for PT in prefs. Using one core less than maximum is a well known optimization thing.

But not for PT 11 and above as that's when being able to set the number of cores manually went away.

[musicman691]

Originally Posted by TNM

I can guarantee you, if you have a project in PT12 that was literally at the edge of not being able to play back, but DID, that 2019 would not be able to play it, not even for a second (on the same computer) unless you went to a more powerful mac. 2019 is overloading when the OS and PT interface are still at full speed and snappy, which also tells me there is so much cpu untapped.. not the pro tools meter, real cpu measurements.

much snippage

I think some people answering have confused my points, like Daryl for example.. I am simply talking about the ability of how many VI's and effects PT 2019 can play vs PT 12/2018 on the very same computer.. it's markedly lower.

Here's one of those things where I'm going to dispute you.When I jumped from PT 11.3.2 on the 2012 cheesegrater in my profile to PT 2019.6 and then 2019.10 I found I could use WAY more plugins at the same buffer setting as I could in PT 11.3.2 and lower cpu usage as well.

An example:
I did my rendition of Tubular Bells with about 100 tracks of mixed vi's (sample based and computational) with some compressor on busses. Before everything was rendered to audio I could barely keep things running in PT 11.3.2 (at 1024 buffer). When all was mixed down and vi's disabled it ran at about three quarters cpu load. Pull up that same session in PT 2019.6 I could run with all vi's playing before rendering to audio and it dropped to less than half cpu load. Render that to audio and barely one quarter cpu load. Could even drop to 256 buffer before things got sketchy.

[TNM]

Originally Posted by musicman691

Here's one of those things where I'm going to dispute you.When I jumped from PT 11.3.2 on the 2012 cheesegrater in my profile to PT 2019.6 and then 2019.10 I found I could use WAY more plugins at the same buffer setting as I could in PT 11.3.2 and lower cpu usage as well.

An example:
I did my rendition of Tubular Bells with about 100 tracks of mixed vi's (sample based and computational) with some compressor on busses. Before everything was rendered to audio I could barely keep things running in PT 11.3.2 (at 1024 buffer). When all was mixed down and vi's disabled it ran at about three quarters cpu load. Pull up that same session in PT 2019.6 I could run with all vi's playing before rendering to audio and it dropped to less than half cpu load. Render that to audio and barely one quarter cpu load. Could even drop to 256 buffer before things got sketchy.

That's fine but I didn't once talk about PT 11 in my testing, other than one brief mention that that was when Avid added a hybrid 1024 playback buffer.

I specifically said PT 12.8.4 on sierra and 2018 on sierra/high sierra.