On Monday 12 January 2009, Giso Grimm wrote:

> Hi all,

>

> in several audio devices I observe a minimal block size setting of 1 ms:

>

> - In the Echo Layla 3G, it is possible to configure smaller block sizes,

> but the driver or device seems to use 3 fragments instead of 2.

>

> - With the RME HDSP9652, I hear artifacts when I configure a fragment

> size which is smaller than 1 ms.

>

> - With the RME HDSP9632 I am not able to configure a smaller fragment

> size at all.

>

> Does anyone know about the source of this mystic border? If it is in the

> ALSA abstraction or driver layer, what is the reason for this? Is it

> simply the assumption that nobody needs shorter fragment sizes, and

> shorter fragment sizes are likely to produce xruns? Or is this just

> caused by the hardware?

Either the hardware or the driver. I have used sub-millisecond latencies (roundtrip) on a delta66 card and a properly setup -rt kernel without XRUNS. E.g. 8 frames per period, 2 periods at 48khz (2*8/48000 is ca 0.3ms latency). ALSA itself (the framework, not each and every device-specific driver) and jackd can handle these latencies just fine.

Just as a sidenote: There is additional latency caused by the DA and AD converters themselfes which varies from device to device. I think Fons once released a program to measure these additional latencies.. I don't know the figures for the delta 66.

Regards,

Flo

--

Palimm Palimm!

http://tapas.affenbande.org