[LAU] Kxstudio RT kernel vs low latency

Len Ovens len at ovenwerks.net
Fri Jul 25 00:20:23 UTC 2014


On Thu, 24 Jul 2014, Russell Hanaghan wrote:

> Wow! Pci(e)? USB? Fw? What IZ interface? That's a lot better than I'm 
> getting. What is your rig/desktop?

I was not long ago running an old P4 at 2.4G with 2.5Gram I could get 2/16 
on that too, but it didn't take much load to mess things up.

I just got myself an i5 (i5 has no hyperhtreading, i7 does... couldn't see 
any other differences) 3.2G and 8Gram.

I am using an older ice1712 based delta66 PCI audio card. I chose my new 
mother board for maximum number of PCI slots as I use an old ens1370 audio 
card as well just for MIDI. Own irqs, CPU to performance, I set my rtirq 
for my main sound card then any other snd (snd_ice1712 snd) I do the same 
on my laptop where I use a USB IF. I have found that USB3 is the only USB 
that is not shared so I do "usb3 snd usb" for order.

Boost is off in BIOS... Thats about it. With the old P4 it even mattered 
which order my PCI cards were in as the D66 seemed to work better if it 
had a higher irq than the ensoniq.

> I hear you... I look briefly thru those to look for snares such as irq 
piling. The bios in this old Vaio is about a useful as a chocolate 
teapot! But I set CPU to performance in user spc (I think...) nowhere to 
deal with hyper. I don't think this dual core has it actually.

dmesg or syslog should tell you near the begining of the boot logging if 
you have hyperthreading or not. If you do it will look like you have 4 
cores instead of two. Turn off cores 1 and 3 on the kernel command line in 
GRUB.

>> The second thing to remember is that multicore CPUs have changed the 
way lowlatency can ....
>
> Could I trouble you to elaborate on last paragraph a little??

With a single core, audio processing has to be able to have access to that 
core on time all the time. WIth multi-cores, The OS can be set to run most 
of it's stuff on one core and the audio RT stuff on the other(s). This is 
making true real time kernels less important than they were. The audio 
world has yet to really put this into practice, but it is possible and I 
think it is not long till it gets to be the normal way to do things. 
Servers already often have memory that is on a per CPU basis and I think 
this will work it's way into desktops as well. Effectively, it starts to 
become like having two (or more) closely coupled computers.

This is not a comprehensive description at all  :)  just enough to point 
you in the right direction if you want to learn enough to implement it on 
your own machine. (I am not there here either)

Even with no fancy set up, multicore CPUs seem to make a difference with 
audio stability. (maybe it is just that this is so much faster than what I 
had before...

--
Len Ovens
www.ovenwerks.net



More information about the Linux-audio-user mailing list