gahabana wrote:
>
> Therefore it is either my specific config or squeezelite multithreaded
> soxr version causes much much higher CPU consumption on intel (as
> opposed to non-multithreaded version) ... i actually expected to see
> 'spike' when soxr kicks in spread on 2 or 4 cores and see 15-20% per
> core for a while but it seems its persistant load.
>
I think I already said this, make sure you are comparing apples to
apples. If using the ondemand governor, running single threaded will
cause the core that it is running on to run at a higher speed, and top
will show that core using less CPU%, relative to core speed. Spreading
the load across multiple cores, from what I have seen, results in the
cores running at a lower core speed, but top shoing them using more CPU,
because they are running at aa lower core speed. I suspect if you run
the test again, 1.6-dev and 1.6 with multi-thread, but fix the core
speed, by setting the performance governor, you will see what I mean.
If you think I've got it wrong, I can look again, but this is the
behaviour I have been seeing on both ARM CPU with multi-core and
x86_64.
The good thing is, from my point of view, and I'm just grabbing these
numbers, from my WB Dual right now....
Code:
--------------------
cpufreq stats: 996 MHz:7.62%, 792 MHz:0.99%, 396 MHz:91.39% (47688)
--------------------
You see that. >90% spent at the lowest core speed, while upsampling 44k1
-> 352k8, 2x threads, for the last 2 days. Sure, the CPU% use looks
higher, but if that had been using the single-threaded version, it would
be the other way round. I suspect >90% at 1G core speed, but showing
much less CPU% use.
I think Triode has taken the patch upstream that enables it by default.
I guess we could go back to not-enabling it by default, like I
originally had with the 'z' option, but I really see no need to do that.
There's plenty of smoke and mirrors here. Like, we know he load avg
numbers generated by the 3.0.35 kernel on the Wandboard are artificially
high..... You need to know what you are looking at, rather than just
taking a "headline" figure as being accurate.
------------------------------------------------------------------------
JackOfAll's Profile: http://forums.slimdevices.com/member.php?userid=3069
View this thread: http://forums.slimdevices.com/showthread.php?t=99395
_______________________________________________
unix mailing list
unix@lists.slimdevices.com
http://lists.slimdevices.com/mailman/listinfo/unix
