CPU Power of FV-1 compared with other reverberator hardware?

Hardware questions and issues with the FV-1

Moderator: frank

Post Reply
seancostello
Posts: 74
Joined: Mon Sep 11, 2006 10:04 pm

CPU Power of FV-1 compared with other reverberator hardware?

Post by seancostello »

Hi all:

I was reading an interesting article in a 1985 issue of the Computer Music Journal, by Anthony Agnello and Ed Hage, describing the architecture of the Eventide SP-2016. I didn't make a copy of the article, but from what I remember the "DSP" part of the system had a pretty sparse architecture by modern standards:

16x16 multiplier
16 general purpose registers
128 DSP instructions/sample at 40 KHz sampling rate
2 cycle access to delay RAM (but it seems that you could perform other operations while waiting for RAM access).

An example program, implementing a simple Schroeder reverb (4 combs w/o filtering, 2 allpasses), took 24 instructions. Apparently 5 of these reverbs could be run consecutively on the SP-2016.

In comparison, the FV-1 looks pretty darned good, especially for a $10 part. Allpass delays only take 2 cycles on the FV-1, compared to 4 cycles on the SP-2016.

Having said that, there are algorithms that were running on the SP-2016 (the Stereo Room, for example), that are eagerly sought today. For that matter, some of the simpler programs for the FV-1 still sound pretty good (like the reverb with 4 allpasses and 2 delays), while using only about 1/4 of the total instructions available per sample. This makes me think that many of the older reverberators were more dependent on good algorithm design than on particularly powerful hardware.

So, does anyone have any idea how the FV-1 stacks up against reverberation hardware of past years or present?

Thanks,

Sean Costello
frank
Posts: 1244
Joined: Wed Oct 19, 2005 12:26 pm
Contact:

Post by frank »

Well, I've designed a number of audio DSPs over the years for companies so I can donate my experience here.

The multipliers have gotten bigger, wider data paths. Memory fast enough to do an access every instruction cycle allows things like the two instruction all-pass. And with the chip geometries getting smaller, more can be done in the same silicon area, things like the LFOs for address modulation to do chorus, etc.

Now, this means that the instruction sets have become more complex to support the additional features. I remember designing chips with as few as 8 different instructions, basically they did reverb and some simple filters. Now, in an FV-1, the instruction decoder is far more complex. I designed the FV-1 instruction decoder by hand (really wish I had logic synthesizer that day!) and it is a very complex block. Making sure that all the operations happen at the right time (i.e. add LFO to address, route data into ALU, etc.), checking all the timing. running tests to make sure it decoded everything properly, checking and rechecking results to make sure it worked prior to going to fab.

So, how does it stack up against other dedicated effect hardware? I believe the FV-1 is more powerful than any of the other current offerings. As far as I know, all the other offerings require a micro if you want custom programs or real time parameter control.
Frank Thomson
Experimental Noize
Post Reply