[5184] L8 uses 30% more CPU than L7

[Dalibor Loncar]

i compared the cpu performance with tarekith´s test, no more than 7 % increase of cpu power consumption

I suspect you are using a quad core? 7-8% per core => ~30%

indeed i have a xeon quad core, but i´m betatesting on a laptop with an intel core2duo mobile processor T8100 (2.1GHz).

i´m on win xp, the cpu power consumption complaints obviously come from mac users.

maybe the actual beta code of the os x port has some design flaws ...

[Friedemann]

Hi,

thank you all for your input.
The problem is probably related to simpler and we are investigating it.

Let me add some comments about *expected* differences regarding performance between
beta builds and the final build:

On PC there is some extra code that allows us to get more information after a crash. This *slightly* degrades the
overall performance (both GUI and audio), but typically not more than a few percent.

On both, Mac and PC there is the EventRecorder running in beta builds that records all user events (like mouse-clicks)
for the sake of reproducing bugs (this is what ends up in the bug reporting .alp files). This should not have any
impact on audio performance but adds a little extra cpu load from the GUI. The EventRecorder can be turned off by putting the line '-EventRecorder=Off' in the Options.txt file (in the preferences folder). If it is off, Live's title-bar
will not show 'recording events' after the name of the current set. Please turn this off *only* for performance tests because otherwise we loose very valuable information in the case of a crash.

If you compare Live 7 and 8 please make always sure that you use the same settings in the preferences, like
sample rate, audio buffer size etc.

Differences in performance between (Intel-)Mac and a comparable PC can arise from the fact that the code is compiled with different compilers which might perform different optimizations even when the target cpu is the same.

[Dalibor Loncar]

The EventRecorder can be turned off by putting the line '-EventRecorder=Off' in the Options.txt file (in the preferences folder). If it is off, Live's title-bar will not show 'recording events' after the name of the current set. Please turn this off *only* for performance tests because otherwise we loose very valuable information in the case of a crash.

guys and gals, don´t forget to switch the eventrecorder on after performance testing

Differences in performance between (Intel-)Mac and a comparable PC can arise from the fact that the code is compiled with different compilers which might perform different optimizations even when the target cpu is the same.

many thanks, that´s very enlightening!

[slirak]

The problem is probably related to simpler and we are investigating it.

OK, that would explain why I experience similar differences with other Live sets: I use Session Drums in basically every set. Will test without them after work.

Let me add some comments about *expected* differences regarding performance between beta builds and the final build...

Brilliant info, thank you very much!

/Jonas

[glitchrock-buddha]

The problem is probably related to simpler and we are investigating it.

I just thought I would mention too that I tried replacing all the simplers in that test set with samplers and got exactly the same performance discrepancies between 7 and 8 as with simpler. So the problem must be both with simpler and sampler. But I guess that's probably assumed since they're the same under the hood.

[Friedemann]

yes, that's correct.

Best, Friedemann

[glitchrock-buddha]

Should this maybe be moved to the bugs section then?

[slirak]

The problem is probably related to simpler and we are investigating it.

I just thought I would mention too that I tried replacing all the simplers in that test set with samplers and got exactly the same performance discrepancies between 7 and 8 as with simpler. So the problem must be both with simpler and sampler. But I guess that's probably assumed since they're the same under the hood.

I just did the same but used Tension instead. Still the same result.

[Friedemann]

My first assumption was wrong, Reverb is causing the problem not Simpler. We are working on a fix.

Best, Friedemann

(I move the thread to the bugs forum)

[bulo]

My first assumption was wrong, Reverb is causing the problem not Simpler. We are working on a fix.

Best, Friedemann

I've just made tests with some of my sets which contains Reverb (two or three instances), and the results were identical from 7 to 8.
Is the problem you've just pointed out with Reverb caused by the fact there are plenty in the Tarekith's test?

Whatever, glad to see that you've find the problem one more time.
You're making such an intensive and great job those last months.
Respect.

[Alex]

It depends on the settings of the Reverb.
However, it should be fixed since Live 8.0b25.

Regards
/Alex

[glitchrock-buddha]

I ran the test again with b24, and yes, it is much better now. What used to take 70% in Live 7, but took 90% in the previous Live 8 betas, now takes 74%, which seems pretty normal. I don't think a few extra percent is anything to worry about, given that it's beta.

I'd say this thread is ready for the --done-- folder!

[mr.adl]

I ran the test again with b24, and yes, it is much better now. What used to take 70% in Live 7, but took 90% in the previous Live 8 betas, now takes 74%, which seems pretty normal. I don't think a few extra percent is anything to worry about, given that it's beta.

I'd say this thread is ready for the --done-- folder!

Cool, thanx for checking!

[Friedemann]

Thanks for verifying, I move the thread to the 'done' section.

Best, Friedemann