Yo, back on Wednesday 05 May 2010 Pau Arumí was all like: I built a box for a friend a few years back using the PCI version of the AES. Works fine, but the part that sucks is there is no GUI tool to set the matrix send levels or show the dsp meters. There is only a small CLI demo tool for setting and retrieving matrix send levels. Kind of a pain in the ass to deal with. On the upside, ALSA exposes all 64 matrix inputs that are available for the matrix. I believe (not for sure) that on windows systems, only the first 16 are exposed. So instead of the advertized 16x16 matrix, you have a 64x16 matrix. -Reuben

Yo, back on Wednesday 05 May 2010 Florian Faber was all like: Inputs to the card's matrix, not inputs to the card itself. If I remember correctly the matrix has 128 inputs, the first 64 are reserved for hardware inputs (on this unit, only the first 16 are actually used by the card inputs) The last 64 are reserved for OS audio sink inputs There are 64 outputs from the matrix for the card outputs (again, on this card only the first 16 of them go anywhere) Alsa exposes all 64 matrix inputs as alsa ouputs, and I don't think the windows driver does that. (Never played with the windows driver, so I may be wrong about that) IIRC, they put such a huge matrix on it because they designed one chip to slap on all their cards in that family. Saves money to just design and fab one chip instead of a separate chip for each unit. -Reuben

Niels Mayer wrote: <snip> I wouldn't agree that RME should switch to VLSI. I also wouldn't say it is entirely fair to blame the cost on them using FPGAs. FPGAs can be quite cheap. Ignoring the mixer for a moment, the FPGA for a 64 channel matrix + numerous input and output channels + high speed link to other cards + PCI interface would be about about $15-20 in quantities of one. Expect discounts for 1000 units, and more discounts for 10,000 units. A VLSI would cost millions to make. They might have to move 100,000 units using the exact same chip at the exact same card price to break even, and if RME was that high volume of a vendor, then their FPGA prices would be much lower than the list prices one can find online.

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?

Yo, back on Thursday 06 May 2010 Florian Faber was all like: I remember being a bit surprised when I first saw that there were 64 playback channels, but at the time I tried them all and I could route any of the 64 playback channels to any of the 16 outputs. Didn't think to test if there were limits to how many I could use at one time. But that was two years ago. Haven't had a chance to play with it since then. Have been thinking of getting one myself (or something similar) to replace the old RME 96/8 PAD I've been using for quite a while now. -Reuben