[linux-audio-dev] Buffer size settings - Mac/Windows

Benno Senoner sbenno at gardena.net
Sun Jun 13 00:01:05 UTC 2004


Michael Ost wrote:

>Well, no responses, no surprise... since this is a Windows/Mac issue.
>But if you are curious, we scoped a win2k machine with a audiophile card
>running vstack. 
>
>The latencies showed that the 128 sample setting matches an ALSA "semi
>buffer". That is, a minimum of 256 samples of latency. This despite the
>128-sample-like latency claims in the MAudio driver setup dialog.
>  
>

Yeah, these vague latency settings on windows apps/drivers are quite 
disturbing and confusing.
Even the german keyboards magazine (whose editors and reviewers are 
supposed to be experts in audio)
get it wrong each time.
They happily run their RME cards at 1.5msec latency while in reaility 
those cards can go as low
(in 44kHz mode) a 3msec ( 2 x 1.5msec buffers).
But it's a bit fault of RME too which never specifies if the latency is 
per fragment or total (it's per buffer).
Same can be said for most audio sequencers on windows/mac, and from what 
I understand most of the times
these numbers mean per-fragment latency. (so multiply with 2 or 3 to get 
the real midi-to-output (or PCM input-to-output) latency.

As said low latencies are cool so everyone tries to cheat and provide 
the per-fragment latencies in their settings/specs.

Btw what did you mean with
"We set the ALSA driver to 2x128 and we get results that jibe more with 
the 256 setting in Windows."

english is my 4th language and dictionary.com was not of too much help

----
jibe:
To make taunting, heckling, or jeering remarks.
To deride with taunting remarks.
---

So let me guess: you meant that 2x128 in ALSA provides lower latency 
than Windows at 256frames ?

cheers,
Benno
http://www.linuxsampler.org

>Anyway, more info for the curious... mo
>
>On Thu, 2004-06-10 at 20:52, Michael Ost wrote:
>  
>
>>Hi.
>>
>>Does anyone out there know what the audio buffer size settings in
>>Windows and MacOS really mean? If you say "128 samples" does that
>>translate to 2 buffers of 128 samples --- one buffer playing, one buffer
>>filling --- or 2 buffers of 64 samples? Is it 256 samples of latency or
>>128?
>>
>>I realize this isn't _exactly_ a Linux audio question, but perhaps
>>someone out there knows something about this. We're trying to get an
>>apples to apples comparison of our Linux/ALSA based system with a
>>Windows/MacOS system.
>>
>>We set the ALSA driver to 2x128 and we get results that jibe more with
>>the 256 setting in Windows. 
>>
>>But when we hooked a Windows system up to a scope it looked more like
>>the 128 sample setting was running 2x64 samples. So... we're confused.
>>
>>Any pearls of wisdom out there? ... mo
>>
>>===================================
>>Michael Ost, Software Architect
>>Muse Research, Inc.
>>most at museresearch.com 
>>
>>
>>    
>>
>
>
>
>  
>




More information about the Linux-audio-dev mailing list