G5 Xserve - Good Audio Potential? Options

[Tree Leopard]

G5 Xserve was recently released ...
http://www.apple.com/xserve/

1U, 19in rack mount unit, with multi-platform compatibility (mac / win / unix) with a lot of power. Bascially, a very attractive proposition, especially for hi-rez recording and DVD post.

From reports the unit is selling very well. (Sure, its price and features are aimed at medium to large enterprise.)
http://www.macworld.com/news/2005/01/04/xs...serve/index.php

For the realists: if you had one of these in yor studio, how would you - given your particular config. and work processes - make use of it? I'm very curious about how people would take advatange of its features.
http://www.apple.com/xserve/specs.html

For the dreamers: ...do you visualize something that is a like half-step between the Mac Mini and Xserve, an upgradable 1U media server / external processor hub that keeps older model G4s in action (i.e. rather than buying a whole new computer to get the power) and is mobile and robust enough for "out of house" production projects...?

What are your thoughts?

[lepetitmartien]

Well the Xserve is around for some time now, it's just an upgrade

now, on the use of Wserve in sudios for example we had a go a while back (in the early days of the xserves) on a Xgrid system for audio. there are issues, especially real time computing, and specific developments to make this things possible. also xserve are too expensive, big and build like servers for this use.

As I wrote, a node mac mini, maybe a little larger with a G5, RAM (enough and eventually soldered), a small fast HD, no CD/DVD player (everything thru network), gigabit Ethernet, stackable, rackable (but not necessarily in a rack form), would be a DSP farmhouse for logic nodes. Buying powerbooks for this even G5 is just buying in an expansive way something that can be produced for cheaper.

It doesn't even need a video card…

And the power unit can be serving a stack (you build along the system)

Making the Node software layer (which must be some Xgrid thingy in fact) available to other editors would only expand the market, in audio in video, in 3D, in DTP (photoshop). And it'd be a more viable road than a huge 4 or 6 CPU desktop computer too expensive to buy and too niche market.

well…

[Tree Leopard]

I've been thinking about your "node" idea and it makes perfect sense - your specifications are practical, too.

As for the HD, Solid State technology is (at last) coming down in price, and is getting a lot of interest from people in DVD post. I rather like the possibility of near-instant seek times and no moving parts.

Yes, we're still faced with the challenge of making realtime networking for audio (& a/v media data in general) really work, whether we use an existing product like Xserve or a "node".

I'm aware that IRCAM have been experimenting with these possibilities for quite a while, but what can happen in a specialized lab might not be easy to translate into a commercial proposition. The cost of component research and testing for one, comes at a high price..

... or ... Apple takes the lead and develops their own protocol that is fully functional across at least 2 generations of machines.

What do you know about the latest advancements in this area? I was looking here at info about Jumbo Frames...

http://www.small-tree.com/jumbo1.htm (an Apple dev partner)

What do you think? Could this kind of idea work for people like us?

[jeffca]

While a G5 Xserve Cluster Node would kick ass as a co-processer in a studio designed around Logic Pro, you'd have to have specific issues to make it worth the cost.

Logic Pro's distributed computing feature is the most powerful and flexible implementation of this concept yet commercially introduced (to my knowledge, anyway). If you have a need for numerous "live" virtual instruments or multiple Altiverbs, an Xserve Node would expand your capabilities very nicely at a much lower cost than a rack of synths and convolution reverbs.

In a live context, imagine a G5 with two 30" Cinema Displays as the head of a live mixing system. Several Xserve nodes could be hooked up via InfiniBand network cards (used in the Viginia Tech G5 super computer) to reduce a live system's latency to a few milliseconds.

Look at the potential savings over a standard live mixing rig:
$10,000 for the G5, two 30" displays and other peripherals (memory, drives, etc.)
$18,000 for six Xserve cluster nodes
$3,500? for seven Infiniband network cards (just a guess)
$1,000 for cables, etc.
$32,500 total

This isn't cheap, but when you consider the cost of a "real" live mixing desk (Midas, SSL) and all of the outboard gear that would be needed to do what this core system would, it's a pitance. Other advantages would be the systems scalability, automation capabilities, ease of use and the fact that the live mix could sound eerily close to the studio recording while still be performed totally live.

This, at present though, is a pipe dream. Logic can be used for live mixing and processing, but it certainly isn't optimized for it. And the last thing you need during a live show is a program crash or lock up.

It's only a matter of time, though, until distributed computing explodes into the studio and the arena. It will be really cool when it does.

Jeff

[swilder]

Distributed processing works well for the scientific community because there's no need to observe the results in real time.

Distributed processing in the recording studio is a different monster because of our need to listen in real time. The big bottleneck being the ethernet connection.

So nowadays we're mostly using 100 Megabits per second. Most of you probably know that on an ethernet, all computer have simultaneous access to the network. That means that the more computers you have on the network, the more "chatter" is going to occur. Ethernets typically are pretty stable up to about 25% of its capacity. Above that; the collisions in data packets start creating errors and drops in data.

Collisions and errors result in audio problems and latency.

I am very interested in seeing a review that publishes real numbers on performance and latency of the node technology.

Scott

[Francois Déchery]

I agree with swilder, distributed processing is not well suited for real time audio applications where latency is critical (ie recording). I would also be very interested to read reviews of nodes.

About the Xserve, if you had a chance to see (and hear) an xserve powered on, you would have noticed that this computer produces more noises than a washing machine.... Expect if sonically isolated in another room, an xserve could not be used in a studio.

Last note for jeffca, the systems that you describe already exists, costs less, is really more DSP powerful, latency free and scalable : it is called a Protools HD

[beyaRecords]

As for the latency issue regarding node work the usage of gigabit comms very much deals with this issue. The main shortfall at the moment is that the nodes can only process logic plugins, so if you have any ideas of farming off those altiverbs or reaktors forget it ;-)

[Tree Leopard]

OK.... thus far, some of the ideas we have are...

1) The Xserve - as it is - sounds like it would be great for a " Meet Your Local Italian Futurist" night. Off to the machine room with you, sir...

Sure, we can all see its potential for data back-up / sound library database use within a network. But it seems as if its just a more chic version of a number of similar products (e.g. SGI servers for HD TV editing)

2) "Distributed Audio Processing" (as featured in Logic 7) has turned out to be a real Problem Child. So much promise but still refusing to co-operate with the authorities...

3) The "Node" - the concept as defined by Lepetitmartien - is still a good one! Its a peep over the horizon. Its "The less-not-more solution that gives you even more than ever before. Not less."

4) Swindler is right - existing networking protocol is not for us. Again, any views / opinions on Jumbo Frames [ see link above ]. Has anyone actually tested this out? Or are we ultimately faced with same packet transfer conflicts...?

Speaking of ethernet, some of you guys would have seen this in recent trade news. Its really one of a number of new CAT5-based wiring options for audio but its a step closer...
http://www.ethersound.com/technology/overview.php

5) Um... fibre optics? Using a protocol based on MADI i/o... ? Any hope there?

[beyaRecords]

Question,
doesn't Nuendo 3 support this Node type technology?

[lepetitmartien]

To be more precise, "my" node is based on available things from Apple, just a little bit of development to bring the software behind to the other editors (and expand the potential market).

This said, it's aimed to non real time operations, so with latency workarounds you can do some jobs in "real time", but the main use would be in audio some treatments, in DTP image manipulation, and 3D. I'm sure it can be pushed in the direction of things like Mathematica

It would be the "low budget" solution on existing technologies already in the hands of Apple.