[Jack-Devel] intel_pstate vs. jack?

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

[Jack-Devel] intel_pstate vs. jack?

Fernando Lopez-Lezcano
Hi all,

New upgrades (Fedora 20), new kernels and suddenly I don't seem to get
the performance I used to from jack & friends (no hardware changes,
using a pretty beefy workstation with 6 core processor and 64G ram), I'm
seeing xruns when I used to have very solid performance[*].

To make the story short, I used to change the cpu frequency governor to
"performance" and set the frequencies to the maximum to get the best out
of jack. I have read many of you have no problems using jack with
frequency scaling active, but that has not been my experience. But I was
surprised to find the frequencies fluctuating even after telling my
computer to not change them... sigh...

I found that the method of controlling frequency on new cpus has changed
(apparently as of 3.10.x) to the intel_pstate driver. Disabling it (you
have to add intel_pstate=disable in the kernel command line and reboot)
seems to have returned the system to its usual behavior although I need
to test more.

Anyone our there also having fun with this??
Any recommendations?

-- Fernando


[*] running a LOT of software, several instances of jconvolver, two of
ambdec, a big ardour session, supercollider, etc, etc. All cores
fluctuate around 30-50% of capacity.
_______________________________________________
Jack-Devel mailing list
[hidden email]
http://lists.jackaudio.org/listinfo.cgi/jack-devel-jackaudio.org
Reply | Threaded
Open this post in threaded view
|

Re: intel_pstate vs. jack?

Holger Marzen
On Thu, 27 Feb 2014, Fernando Lopez-Lezcano wrote:

> New upgrades (Fedora 20), new kernels and suddenly I don't seem to get the
> performance I used to from jack & friends (no hardware changes, using a pretty
> beefy workstation with 6 core processor and 64G ram), I'm seeing xruns when I
> used to have very solid performance[*].
>
> To make the story short, I used to change the cpu frequency governor to
> "performance" and set the frequencies to the maximum to get the best out of
> jack. I have read many of you have no problems using jack with frequency
> scaling active, but that has not been my experience. But I was surprised to
> find the frequencies fluctuating even after telling my computer to not change
> them... sigh...
>
> I found that the method of controlling frequency on new cpus has changed
> (apparently as of 3.10.x) to the intel_pstate driver. Disabling it (you have
> to add intel_pstate=disable in the kernel command line and reboot) seems to
> have returned the system to its usual behavior although I need to test more.
>
> Anyone our there also having fun with this??
Since I have a new notebook I needed a recent kernel für my Xubuntu
12.04 LTS. I chose kernel 3.12 and noticed that the ondemand-governor
was missing. On my workstation at home with kernel 3.5.7 I use cpufreqd
to boost the frequency to the max when some of my audio programs are
running:

#-- Profiles ---------------------------------------

[Profile]
name=On_Demand_High
minfreq=100%
maxfreq=100%
policy=ondemand
[/Profile]

[Profile]
name=On_Demand_Standard
minfreq=0%
maxfreq=100%
policy=ondemand
[/Profile]

#-- Rules ------------------------------------------

[Rule]
name=Standard_Rule
cpu_interval=0-100
profile=On_Demand_Standard
[/Rule]
 
[Rule]
name=CPU_Too_Hot
acpi_temperature=60-100
cpu_interval=50-100
profile=On_Demand_Standard
[/Rule]

[Rule]
name=Music_Production
programs=rosegarden,jamin,qtractor
cpu_interval=0-100
profile=On_Demand_High
[/Rule]


But on my notebook (Core i5 like my workstation) I have no problems when
simply stopping cpufreqd and keep the defaults.

I don't know how the Fedora kernel is tweaked. On my workstation at home
I got the biggest improvements bei setting HZ1000 and the preemptive
model on my kernel 3.5.7:

make menuconfig
  Processor type and features
    Preemption Model: Low-Latency Desktop
    Timer Frequency: 1000 HZ

_______________________________________________
Jack-Devel mailing list
[hidden email]
http://lists.jackaudio.org/listinfo.cgi/jack-devel-jackaudio.org
Reply | Threaded
Open this post in threaded view
|

Re: intel_pstate vs. jack?

Christian Schoenebeck
In reply to this post by Fernando Lopez-Lezcano
On Friday 28 February 2014 01:47:02 Fernando Lopez-Lezcano wrote:
> To make the story short, I used to change the cpu frequency governor to
> "performance" and set the frequencies to the maximum to get the best out
> of jack. I have read many of you have no problems using jack with
> frequency scaling active, but that has not been my experience. But I was
> surprised to find the frequencies fluctuating even after telling my
> computer to not change them... sigh...

CPU scaling very often leads to xruns. However I wondered, has somebody
investigated this issue in depth already?

I mean, it seems as if Macbooks are also using CPU scaling, still providing
very solid low latency stability with its XNU kernel. So I wonder if CPU
scaling is not appropriately handled (regarding the aspect of low latency) in
the Linux kernel. Any thoughts?

Best regards,
Christian Schoenebeck
_______________________________________________
Jack-Devel mailing list
[hidden email]
http://lists.jackaudio.org/listinfo.cgi/jack-devel-jackaudio.org
Reply | Threaded
Open this post in threaded view
|

Re: intel_pstate vs. jack?

Jeremy Jongepier
On 02/28/2014 11:55 AM, Christian Schoenebeck wrote:

> On Friday 28 February 2014 01:47:02 Fernando Lopez-Lezcano wrote:
>> To make the story short, I used to change the cpu frequency governor to
>> "performance" and set the frequencies to the maximum to get the best out
>> of jack. I have read many of you have no problems using jack with
>> frequency scaling active, but that has not been my experience. But I was
>> surprised to find the frequencies fluctuating even after telling my
>> computer to not change them... sigh...
>
> CPU scaling very often leads to xruns. However I wondered, has somebody
> investigated this issue in depth already?
>
> I mean, it seems as if Macbooks are also using CPU scaling, still providing
> very solid low latency stability with its XNU kernel. So I wonder if CPU
> scaling is not appropriately handled (regarding the aspect of low latency) in
> the Linux kernel. Any thoughts?
>
> Best regards,
> Christian Schoenebeck
Hi Christian,

You might want to check out http://rg42.org/oss/jackfreqd/start

Bye,

Jeremy



_______________________________________________
Jack-Devel mailing list
[hidden email]
http://lists.jackaudio.org/listinfo.cgi/jack-devel-jackaudio.org

signature.asc (853 bytes) Download Attachment