[LAU] Thinkpad R60 for Audio Update - Firewire Conflicts with Audio

Pieter Palmers pieterp at joow.be
Tue Apr 10 08:12:15 EDT 2007


Dmitry Baikov wrote:
> On 4/10/07, Robin Gareus <robin at gareus.org> wrote:
>> PCMCIA is ok, but not great for low latency. (adds one extra interface
>> between PCI bus & sound-card). - the cheap solution is to get a PCMCIA
>> card for your hard-disk - instead of using the built-in firewire for the
>> disk. - maybe you want to do that anyway. use the built-in firewire to
>> an external audio-device and the PCMCIA for storage.
> 
> Quite the opposite.
> PCMCIA is great for latency. That "extra" PCI bridge does NOT affect
> audio latency.
One thing is of course that you should check that your bridge's 
interrupt isn't shared either.

> Firewire latencies are bigger than PCI/PCMCIA and will always be. Ask
> Pieter Palmers for details.
Correct. The FireWire host controller is a PCI card, meaning that it has 
the same latency limitations as a PCI soundcard. The latency of the 
FireWire bus itself is added to this lower limit.

Let's say that if you can achieve a certain latency number on a certain 
configuration with a prefect FireWire setup, you will always be able to 
achieve a lower latency on the same system with a perfect PCI setup.

That being said, I don't think this is an issue. If we address the 
FireWire host controller in the same way as we address the PCI soundcard 
(i.e. DMA double buffering), it won't introduce any extra latency. The 
extra latency that occurs, is due to the buffering on the device. 
Therefore, with respect to latency, there is not really a difference 
between FireWire and ADAT or MADI: It depends on the buffering on the 
interface you hook up to it. Basically firewire can give the same 
performance as RME+MADI, Delta+ADAT, ...

There is one problem with this: currently we can't adress the FireWire 
host controller in this manner, meaning that we are limited in the 
performance. It is being solved in the new juju firewire stack.

> 
> So if you are for low latency, PCI/PCMCIA is unbeatable.
> If you want numbers, I can provide them for Echo Indigo IO and RME 
> MultifaceII.
> Ask somebody to jdelay their Firewire card.
Please don't :). The current system is far from 'ideal', and would 
certainly give a wrong impression on the capabilities of the FireWire 
technology.

Results for a first generation FireWire device: 
http://freebob.sourceforge.net/index.php/ESI_Quatafire_Latency

At 48k it comes down to the fact that the device + the windows 1394 
stack together limit the roundtrip latency to around 250 frames. FreeBoB 
has a similar latency lower limit. Note that the audio buffers should be 
added to this, making the actual number around 320 frames.

The newer FireWire devices (based upon DM1500 or certainly DICE) will 
exhibit much lower latencies. I don't have figures on that yet, I'll try 
and see if I can get some... I expect something like 50-100 frames of 
extra latency, for a total of around 150 frames. This is quite 
comparable to the PCI cards.


Pieter



More information about the Linux-audio-user mailing list