Reuben! What 64 channels? The card only has 16 input channels. Flo -- Machines can do the work, so people have time to think. public key DA43FEF4 x-hkp://wwwkeys.eu.pgp.net

On Wed, May 5, 2010 at 3:55 PM, Reuben Martin <reuben.m(a)gmail.com> wrote: RME uses FPGA's, which is why they're so expensive ( http://www.rme-audio.de/en_support_techinfo.php?page=content/support/en_sup… ). It's several bazillion of these http://en.wikipedia.org/wiki/File:FPGA_cell_example.png that you can connect up into any electronic circuit you want. I'd imagine they designed it so that each FPGA takes a certain number of channels and there's an addressing limit to how many FPGA's can be chained together in a multi-channel rig... thus the 64 channels of inputs even though each card has only 16 ins. It seems like after so many years of production, now that they've stabilzed on a particular architecture, they could have come up with custom VLSI to replace the FPGA and save everybody a bunch of money; and allow their cards to be used by a wider number of musicians whose budgets send them to "prosumer" cards with chips like http://www.via.com.tw/en/products/audio/controllers/envy24/ ; these implement their 20-channel 36-bit wide digital mixer in VLSI, and derive significant economies of scale from the fact that they can sell one of their chips in every PC made. I'm not complaining BTW, I'm just jealous of those that have RME cards. :-) For that price, I bet they don't stick any quasi-balanced outputs on their equipment and call them balanced like M-audio does for their Delta series .... ( which is actually "balanced" from an impedance and/or marketing perspective http://en.wikipedia.org/wiki/Balanced_audio#Differential_signalling ). -- Niels http://nielsmayer.com

Niels, The Spartan 3 doesn't cost that much. What's in the firmware is what makes them expensive. Not really. And it's only a few thousand. Well, there are constraints, like the number of dedicated multipliers, you have almost no storage capabilities and so on. But if you think of the 8192 mixers for example a MADI card has to calculate for each sample period, you can use a divide and conquer strategy to add up the data and be done with it in a matter of a few cycles. After all, 64 is a nice two's complement :) Flo -- Machines can do the work, so people have time to think. public key DA43FEF4 x-hkp://wwwkeys.eu.pgp.net

On Thu, May 06, 2010 at 10:46:33AM -0700, Niels Mayer wrote: If RME is not using a FPGA in the babyface, then they are probably using a 3rd party chipset. I don't know where they establish their prices from. If you look on Digikey, you will see that you can get some quite capable FPGAs for $15-20. FPGAs that can easily handle a 72x72 matrix AND a PCI bus (directly connected to PCI), and some amount of hardware mixing. Spartan3s can use a standard SPI PROM for configuration (a big improvement over the Spartan IIs), which probably costs about $3ish. You also likely need an oscillator, which might be about another $4. And these prices are for single unit quantities. If you order $1000 at once, you will start to see the price be lower still. Keep in mind, what might really drive up the size of the required FPGA would be the 8192 point mixer, which is something that no prosumer card does. A lot of cards with CME and Via VLSI chips also use normal off the shelf ADCs, DACs, and/or digital audio recievers and transmitters, so clearly these aren't the sole determiner of cost. Now, I've never seen the insides of a device with a large number of channels of IO that didn't use an off the shelf FPGA or DSP, so perhaps some higher end products from Apogee or Avid do put everything into one ASIC. I would venture to say that 5:1 might be a reasonable profit margin for a company RME's size. I know that is what a previous employer who made (among other things) multi-channel digital audio products aimed for. Sorry, but switching to a VLSI is probably a very large incremental cost. While the company I worked for was too small to make VLSIs, the rule of thumb I heard was a minimum order in the millions of dollars (usualy figure I see in EDN is something like $2 million). Going from HDL to chip masks is apparently not entirely automated, and once the masks are designed, actually making them and setting up a production line is extremely expensive. After that, each chip made is very cheap. Sure, but very low volume is probably anything under a quarter of a million units. Just for the heck of it, does anyone have any idea how many Protools Digi units are sold in a year?

Reuben, Than that's a bug in the snd-hdspm driver. It is an FPGA, so it's just a piece of firmware. If it really works to pass along 64 playback channels instead of the 'official' 16 (I don't know, I don't have an AES card), I really wouldn't recommend using this. If one day the firmware gets modified to add new features, some now officially 'unused' multipliers might be re-used and things will break. After all, there is just a limited number of multipliers available. Flo -- Machines can do the work, so people have time to think. public key DA43FEF4 x-hkp://wwwkeys.eu.pgp.net