[linux-audio-dev] Linux DSP Hardware?

[David Olofson]

On Tuesday 22 February 2005 05.41, jipi wrote:
> Hi All,
> 
> I was thinking,
> 
> we always have dedicated graphic cards for gaming/3D rendering etc..
> why don't we have some h/w optimised audio algos running on some 
> FPGA/DSP/ etc...

Because there isn't enough demand for it. Only musicians have any 
serious use for that kind of hardware, whereas half the planet needs 
accelerated graphics - especially with all the eyecandy that 
Microsoft, Apple, the KDE and Gnome people etc have decided we 
need. ;-)


> on linux...

Dito. The state of accelerated audio on Linux should realistically 
relate to the corresponding on Windows and Mac in the same way as 3D 
acceleration on Linux vs Windows and Mac - and it pretty much does, I 
think... Some popular hardware is supported by Free/Open Source 
drivers to some extent, some vendor has tried to support the Linux 
community to some extent, and that's about it.


> look at these.. are there simple/cheap/open versions of these?

Simple: Not possible, unless you'll settle for a toy synth...

Cheap: Cheap and good are basically mutually exclusive when it comes 
to hardware. The best you can do is to use high volume components. 
Currently, that means standard general purpose DSP chips for small 
devices, and probably AMD64 chips for heavier gear.

Open: Only if you're ready to pay a premium for the "lost revenue" 
resulting from opening up the "trade secrets". You don't want to 
design some cool hardware only to have some foreign company make a 
clone of it, sell it dirt cheap (*you* paid the development costs!) 
and put you out of business. That doesn't mean you have to close both 
the design *and* the programming info, but try to explain that to the 
management of your average company... Also try to explain why they 
should be taking the risk of making reverse engineering easier for no 
significant monetary gain.


> http://www.lyrtech.com/DSP-development/audio/index.php
> http://www.zpeng.com/Articles/Section1/digitalaudio.html
> http://www.altera.com/products/devkits/partners/kit-ate-dmck.html
> http://lts1pc19.epfl.ch/repository/Simeonov2004_737.pdf
> 
> what we require is a separate card which can take a few channels of 
> audio lines and process them in a cheap FPGA/DSP (put in your
> favourite  
> **HW LADSPA** algo).. and route them back to the main audio card..

If it's actually supposed to do something serious, it won't be cheap. 
Really powerful DSPs are nowhere near cheap.

If you're into dedicated hardware anyway, why not just hook up a PC 
(SBC, perhaps; though you still get more power/$ if you buy standard 
mainstream PC components) running RTLinux or RTAI? Cheaper and much 
easier to deal with...


> of course I am for the idea of using clusters of linux PCs while
> some of them do different types of audio processing,
> but not everyone has access to more than 1 PC.. : >

Well, if you can get any DSP hardware worth using, you might as well 
get another PC... As an extra bonus, you get a "universal machine", 
as opposed to some oddball DSP card that you're hard pushed to even 
get it to do what you purchased it for.


Now, if you really don't want to use another PC, and want lots of dirt 
cheap processing power, how about learning some EMU10k1 asm? :-) 
Seems to have more DSP power than your average studio sampler - and 
since it's a high volume product,  just like the 3D accelerators, it 
lets you see some of the "silicon is free" effect. That's the closest 
audio gets to the state of 3D acceleration currently, AFAIK.


//David Olofson - Programmer, Composer, Open Source Advocate

.- Audiality -----------------------------------------------.
|  Free/Open Source audio engine for games and multimedia.  |
| MIDI, modular synthesis, real time effects, scripting,... |
`-----------------------------------> http://audiality.org -'
   --- http://olofson.net --- http://www.reologica.se ---