[LAU] Discussion: Future CPU-technology vs. realtime audio?

[david]

On 1/27/21 8:17 AM, Will Godfrey wrote:
> On Wed, 27 Jan 2021 10:02:48 -0800 (PST)
> Len Ovens <len at ovenwerks.net> wrote:
>
>> On Wed, 27 Jan 2021, Michael Jarosch wrote:
>>
>>> As soon as frequency switching was introduced we LAU were told, not to
>>> use it to save xruns. And as far as I can tell, the rule still valid. Is
>>> there a chance in the future that we can stop thinking about it, because
>>> it just won't matter? Are we forced blowing loads of energy or do we
>>> spent too much time in sluggish UEFI menus?
>> It depends. Setting jack to frame size 64 or lower has in my experience
>> shown xruns with frequency switching. This includes Intel's "Boost"
>> setting which is not turned off by setting performance. There seems to be
>> no problem when the speed goes up but I often see an xrun when the speed
>> goes down (just at speed change).
>>
>> However, Intel has been doing speed switching in the CPU for a while now
>> and we are still able to set a steady speed with that on the fly. So with
>> the AMD it may be similar. It may still be possible to set an upper and
>> lower speed limit. What they do not say, is that the advertized speed may
>> not be usable in steady state. With the Intel, the advertized speed can be
>> set for all cores and run that way at 100% core use and run forever
>> without over heating. AMD tends to advertize a cpu speed based on some
>> cores running slower and the cpu managing heat by slowing some cores down.
>> In this case one will have to experiment to find out what speed can be
>> safely run on all cores without over heating and use that speed for audio.
>> Hopefully this can be set on the fly.
>>
>> Another comment of "blowing loads of energy" with performance mode. It was
>> actually found that the old "ondemand" governor actually used more power
>> than "performance" in many cases. Ondemand has to wake up every once in a
>> while to see what is happening, but in performance mode the core can go to
>> an idle state. The newer intel powersave mode does not have this problem
>> but AMD (although they started work on their own governor for linux) can
>> only use ondemand.
>>
>> The easiest way to see power use is to watch core temperature... all power
>> used ends up as heat.
>>
> FWIW I'm running an earlier Ryzen5 here at 32 frames, mostly with Rosegarden and
> Yoshimi, using quite complex patches, and averaging 8 active channels at once. I
> don't see any Xruns. I do see the occasional one if I do things like load a full
> patch set over a running program.
>
Ages ago, I forced my old i7 laptop (with USB2) to performance mode (via 
command lines on bootup, IIRC) and it stayed there, running at 2.4GHz 
all the time. Occasional xruns but nothing that interfered with anything.

I've not touched the performance modes on my current i9 laptop. It 
ranges from sub-1GHz to 4.5GHz clock, as needed. I think I've seen 
fewer  xruns than I had on the i7 with an RT kernel. I don't know if 
that's related to the fact that the i9 laptop uses USB-C/USB3.2 and I 
have almost everything (including the Ethernet network connection) going 
through an external Thunderbolt/USB-C dock. USB soundcard and USB MIDI 
connection also go through the dock.

I'm not running an RT kernel on the i9. I don't know how USB3.2/USB-C 
handles interrupts vs how USB2 handled them, or if that would have an 
impact on things.

Audio-wise, I don't do anything anywhere near as ambitious as Will does.

-- 
David W. Jones
gnome at hawaii.rr.com
authenticity, honesty, community
http://dancingtreefrog.com
"My password is the last 8 digits of π."