How's the CPU efficiency in Live 10?

Discuss music production with Ableton Live.
Nokatus
Posts: 1068
Joined: Fri Jul 01, 2005 7:06 am

How's the CPU efficiency in Live 10?

Post by Nokatus » Sat Nov 04, 2017 9:07 pm

For the people testing the beta, how does this new version seem, CPU-wise? As in, are you getting comparable plugin counts and track counts as before, or is it maybe even somewhat more efficient? Not talking about the optimizations in the M4L devices, but the main application when hosting many plugins and running largeish projects.

Calagan
Posts: 268
Joined: Fri Jul 24, 2015 4:44 am

Re: How's the CPU efficiency in Live 10?

Post by Calagan » Sun Nov 05, 2017 9:07 am

I'm very interested in this subject too.

Actually, if I can gain some % on the cpu meter (specially while rewiring Reason), it worths the upgrade price...

On some projects I'm using on stage, I just need 5% less for being secure...

I don't have space on my hard disk for installing the beta, but you beta testers, please release the infos...

jlgrimes
Posts: 1770
Joined: Mon Jan 22, 2007 1:55 am
Location: Atlanta, Ga

Re: How's the CPU efficiency in Live 10?

Post by jlgrimes » Sun Nov 05, 2017 12:02 pm

Nokatus wrote:For the people testing the beta, how does this new version seem, CPU-wise? As in, are you getting comparable plugin counts and track counts as before, or is it maybe even somewhat more efficient? Not talking about the optimizations in the M4L devices, but the main application when hosting many plugins and running largeish projects.

Usually early betas can be high in CPU use as many beta versions contain logging code to track everything you do in program to help them debug.

At some point though they remove logging which gives you a better impression of performance.

Calagan
Posts: 268
Joined: Fri Jul 24, 2015 4:44 am

Re: How's the CPU efficiency in Live 10?

Post by Calagan » Sun Nov 05, 2017 1:04 pm

ok jlgrimes, thanks for the info...

But anyway, it means that if there is a change in CPU usage between Live 9 and Live 10, we can take this into consideration. And if the CU usage is lower with live 10 beta, it means it will be even lower with the official release.

So if any beta tester is willing to test some of his/her projects in Live 9 and Live 10, please feel free to post here your results...

Nokatus
Posts: 1068
Joined: Fri Jul 01, 2005 7:06 am

Re: How's the CPU efficiency in Live 10?

Post by Nokatus » Sun Nov 05, 2017 1:58 pm

jlgrimes wrote:Usually early betas can be high in CPU use as many beta versions contain logging code to track everything you do in program to help them debug.

At some point though they remove logging which gives you a better impression of performance.
Yep, thanks, should have added this as a disclaimer myself :). Having been here since v4 and tested some betas, there's a fair bit one can determine from how it runs, though.

Angstrom
Posts: 14921
Joined: Mon Oct 04, 2004 2:22 pm
Contact:

Re: How's the CPU efficiency in Live 10?

Post by Angstrom » Mon Nov 06, 2017 1:15 am

I've tested out a few things I'd say it's either the same or around 0.5% better, depending on case.

EG: Live 9: a big pad with stacked Operators that cost me a solid 16% for a 6 note chord and 22% for a 10 note chord
Live 10 that same pad vacillates between 16% and 15% and then between 22% and 21%

You might assume that some items like Max devices will run a little lighter, but I've not noticed it
This is an instance of Electric running through Four in-line Convolution Pro reverbs set on Nuclear Reactor with the time turned up to the maximum (*all on one track so no multicore)

You'll notice the performance is exactly the same.

Image

locojohn
Posts: 487
Joined: Mon Feb 19, 2007 8:42 am
Location: riga, latvia

Re: How's the CPU efficiency in Live 10?

Post by locojohn » Mon Nov 06, 2017 1:19 am

How about a difference in rendering time of a larger multi-track project between Live 9 and Live 10?

Andrejs
/*
  • the basic tone of life remains the same,
    and in it there are some happy melodies
    and some sad melodies
    - sekito kisen
*/

Angstrom
Posts: 14921
Joined: Mon Oct 04, 2004 2:22 pm
Contact:

Re: How's the CPU efficiency in Live 10?

Post by Angstrom » Mon Nov 06, 2017 1:22 am

I;ll try that out at some point, when I'm not making noises. But if that's where the CPU increase has hidden then they shouldn't have bothered spending their dev time there.
When I press render that's my time to have a piss and go and make a cup of tea. I'm not even in the room when the render happens. It's almost always less time than a cup of tea takes to make.

/england

locojohn
Posts: 487
Joined: Mon Feb 19, 2007 8:42 am
Location: riga, latvia

Re: How's the CPU efficiency in Live 10?

Post by locojohn » Mon Nov 06, 2017 1:34 am

Angstrom wrote:I;ll try that out at some point, when I'm not making noises. But if that's where the CPU increase has hidden then they shouldn't have bothered spending their dev time there.
When I press render that's my time to have a piss and go and make a cup of tea. I'm not even in the room when the render happens. It's almost always less time than a cup of tea takes to make.
It all depends. At times, I have to render multiple times on a single day. Reason being I must send multiple versions of a track out for audition. It happens more often when I work on a project as a part of a team (audio/visual installation etc.) Naturally, I want rendering process to be as quick as possible, but it often isn't the case in Live 9, so I am wondering if there are some improvements in this regard. I have an Intel Core i7 processor with 4 cores, and when rendering in Live 9 the CPU usage seems to never be at max.... It actually feels like Live is using a single core only!

Andrejs
/*
  • the basic tone of life remains the same,
    and in it there are some happy melodies
    and some sad melodies
    - sekito kisen
*/

Nokatus
Posts: 1068
Joined: Fri Jul 01, 2005 7:06 am

Re: How's the CPU efficiency in Live 10?

Post by Nokatus » Mon Nov 06, 2017 9:54 am

Thanks, Angstrom! Seems to be very nice.

Calagan
Posts: 268
Joined: Fri Jul 24, 2015 4:44 am

Re: How's the CPU efficiency in Live 10?

Post by Calagan » Mon Nov 06, 2017 10:08 am

Thanks a lot Angstrom !

So maybe Live 10 will be a bit better, if this beta version contains some logging codes that add to the cpu usage...
At least it's not worst, and it's supposed to be better with M4L.

Stromkraft
Posts: 7033
Joined: Wed Jun 25, 2014 11:34 am

Re: How's the CPU efficiency in Live 10?

Post by Stromkraft » Mon Nov 06, 2017 8:02 pm

locojohn wrote:I have an Intel Core i7 processor with 4 cores, and when rendering in Live 9 the CPU usage seems to never be at max.... It actually feels like Live is using a single core only!
There is no reason to assume Live can forego that a track is handled on one core and not two or three. This likely means tasks on cores with less work will have to wait for the core doing the most work as what's built is a 64bit sum of all tracks.

For natural resource reasons then you can' expect to fill every core to the max. Live can't reasonably start rendering a few seconds at 5:30 when another core is rendering at 0:45 and stitch everything together successfully at the end. That's not how software operates*, at least to currently. Also the computer do other things.

*I'm 98% sure of this anyway, so there's a chance I'm wrong. But I'm not.
Make some music!

locojohn
Posts: 487
Joined: Mon Feb 19, 2007 8:42 am
Location: riga, latvia

Re: How's the CPU efficiency in Live 10?

Post by locojohn » Mon Nov 06, 2017 8:41 pm

Stromkraft wrote:
locojohn wrote:I have an Intel Core i7 processor with 4 cores, and when rendering in Live 9 the CPU usage seems to never be at max.... It actually feels like Live is using a single core only!
There is no reason to assume Live can forego that a track is handled on one core and not two or three. This likely means tasks on cores with less work will have to wait for the core doing the most work as what's built is a 64bit sum of all tracks.

For natural resource reasons then you can' expect to fill every core to the max. Live can't reasonably start rendering a few seconds at 5:30 when another core is rendering at 0:45 and stitch everything together successfully at the end. That's not how software operates*, at least to currently. Also the computer do other things.

*I'm 98% sure of this anyway, so there's a chance I'm wrong. But I'm not.
Suppose there's a heavy effect on a particular track. Why can't Live utilise ALL cores to process this effect if it detects there's less work for other cores at that moment? Eg why should all other 3 cores be stuck while a single core is trying to complete a CPU-intensive operation?

I simply assumed Live 10 could handle this differently, this is why I asked about rendering performance.

Andrejs
/*
  • the basic tone of life remains the same,
    and in it there are some happy melodies
    and some sad melodies
    - sekito kisen
*/

Nokatus
Posts: 1068
Joined: Fri Jul 01, 2005 7:06 am

Re: How's the CPU efficiency in Live 10?

Post by Nokatus » Tue Nov 07, 2017 4:12 am

locojohn wrote:Suppose there's a heavy effect on a particular track. Why can't Live utilise ALL cores to process this effect
Because this isn't how audio computation works. Seriously :). There is no way to arbitrarily share the load of a single effect on multiple cores if the effect doesn't do that by itself. Graphics computation can be parallelized very efficiently, but when you're dealing with audio, you often run into situations where you need to have the result of a previous operation (upstream in the signal chain) in order to compute the next step (downstream in the signal chain). Note that a single effect also contains its own internal signal chain, in a heavy effect potentially a complex one; many sequential operations that need to know the preceding result, and so on. Of course there are plenty of opportunities for parallel processing in audio, too, as we are used to dealing with many parallel signal chains when using a DAW -- or a hardware mixing environment -- but the total load isn't quite as readily and arbitrarily distributable as you presume.

(It is, however, more distributable than using just one single core, heh, because of the inherent parallel operations that are present in a multitrack DAW signal flow, as said above. And yeah, historically Live has gone through a looong period of literally rendering using only one single CPU core :P, that's true.)

Stromkraft
Posts: 7033
Joined: Wed Jun 25, 2014 11:34 am

Re: How's the CPU efficiency in Live 10?

Post by Stromkraft » Tue Nov 07, 2017 8:23 am

locojohn wrote:
Suppose there's a heavy effect on a particular track. Why can't Live utilise ALL cores to process this effect if it detects there's less work for other cores at that moment? Eg why should all other 3 cores be stuck while a single core is trying to complete a CPU-intensive operation?

I simply assumed Live 10 could handle this differently, this is why I asked about rendering performance.
I think for this to be possible there would need to be a breakthrough in how software interfaces with modern CPUs, so that cores can be presented to the software as one core and used that way effectively, also when a code thread has started to work on something. I'm outta my depth on how exactly these things work now so my take here is conceptual.

The OS can distribute work to some extent, but this functionality may not be up to the needs of the music software in this case, because a render are made by so many inter-dependent calculations and handing over work to threads running on another core does not come free, especially as likely results need to come back when ready. So there's a resource cost involved.

Think about, at every single sample step Live needs to calculate every little action on every track, on every send, every routing and sum up the result.

This is normal operation of course, not just when exporting. So what Live does now when rendering, I believe anyway, is to speed up this normal process to the max of what it has allotted on one core (which the OS negotiates). In practice this may be different. I haven't investigated.

The code in a plug-in sometimes can distribute some of its operations on multiple cores and it's perhaps possible that on certain CPUs, even if this plug-in works better with multi-core toggled off when monitoring, that having this toggled on for exporting could speed things up. It also could just end up being a fight for resources (with Live) and negate any gains made.

Granted, there are ideas out there how to route around problems like these involving cores, and new approaches may be better than older, but there's some expected inertia in changing an entire industry, of which music software is only a smaller part.

Talking about cores, do you realize that iPhones (some anyway) have at least six cores these days?
Make some music!

Post Reply