On Fri, 2006-04-07 at 22:40 -0400, richard nienhuis wrote: > > > On 4/7/06, Rene Herman <[EMAIL PROTECTED]> wrote: > Sorry, you ofcourse meant a completely external card here. > Tired. Also > not very good though. Everything, and certainly latency, is > much more > straightforward to control with a PCI(e) card, coupled with an > extgernal > box with the analog parts. > > By the way, in the rest of this message I was talking about a > high-end > consumer card. I do not think there's much market for a > consumer card > over EUR250. No market whatsoever over EUR500. > > > This chart was enlightening: > http://www.evaluationengineering.com/archive/articles/0801pcbased2.gif > > If its true then 1000B-Tx or 10K ethernet will probably be as good or > superior to 1934. If it is really critical there is hardware > available for making low latency ethernet networks. The other option > would be using infiniband which is really low latency. But interface > cards for it are still pricey. I think ethernet is a good option > though. Given a sane network and decent hardware it should be able to > match 1394. Also I don't believe there is anything on the market in > the way of network attached audio devices. It would be great if you > could just drag one into a room with ethernet and be ready to go.
I am an audio engineer, or was until very recently, anyway. I am partially responsible for several of the 1394 audio reference designs out there. I have also worked on PLLs for regenerating the clocks in audio systems. Ethernet is generally considered useless for pro audio unless you are using CobraNet [1]. The standard Ethernet MAC is non-deterministic for accessing the medium. This causes jitter in reconstructed clocks and makes them useless. You can use CobraNet, if you don't mind paying incredibly high licensing fees (in the region of $10/channel of audio per device). 1394 has an incarnation that works over Ethernet cable - 1394b [2]. In my experience 1394 provides reliable isochronous data transport, but everything except the 1394 standard is screwed. The higher level protocols were designed by people that didn't understand that 1394 is a bus, not a network. Avoid it unless you do everything yourself. (The MOTU devices are an example of not conforming to any of the 1394 standards. Good for them, because the standards suck). Furthermore, unless you know what you are doing, it is unlikely that you will get reliable hardware designed and built. Most 1394 chips are horribly twitchy in subtle ways. MOCA [3] looks promising. They implemented a standard the right way - they made it work, then invited other people to make stuff using it. Finally, if OGP decides to do an audio card of some sort, they need to realise that generating and regenerating clocks is the most important thing to do. If you want to do pro audio, I recommend NONE of the above standards. It's possible that MOCA could do what you need; I don't know. I do know that 1394 requires fancy footwork to generate the kind of clock you need for pro-audio. I also know that CobraNet is just too damned expensive, and proprietary besides. If you want to do good pro audio today on a linux machine, do yourself a favour, and look at the sound card section at the RME web page [4], and buy yourself something there. Their boards are already based on a Xilinx FPGA, and OGP reinventing the wheel seems like a marginal idea to me. Cheers, Ray 1: http://www.cobranet.info/en/support/cobranet/ 2: http://www.lecroy.com/tm/Solutions/SerialData/IEEE/default.asp?menuid=29 3: http://www.mocalliance.org/en/index.asp 4: http://www.rme-audio.de/english/index.htm _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
