Synths, anything special with multiples of 8?

During the research for my blog about instruments commonly used in Prog I realised that certain features in keyboards / synths frequently showed up in multiples of 8 (or you could say of 4), particularly:

^ "3" was another common number in early synths.

Gerinski wrote: Ajay wrote: Powers of two are common in computing. The progression you mention, though, Gerinski, generally holds only from about the late '80s. Early polysynths (e.g. the Polymoog, the Korg PE-1000) were fully polyphonic - every key could sound simultaneously. The Prophets are 5-, 10-, and most recently, 12-voice polyphonic. Similarly, the Prophet 5 had about 40 patch memories, the Yamaha GX-1 had 70. Thanks for the info. According to this site, the GX-1 did not have any memory patches (specs table on the right: 'Memory = none'). http://www.vintagesynth.com/yamaha/gx1.php

The preset patches on the GX-1 were 70 "tone modules", each containing 26 resistor dividers for setting various parameters on the sound banks.

Gerinski wrote: b) the number of memory patches, again not a golden rule but it was also frequently 16, 32, 64, 128... None of these features seem to have any particular requirement towards this, I mean, a synth could perfectly have 50 voices or memory patches, or 100 or 125 or 150 or whatever, 64 or 128 seem like odd numbers to choose?

Some synths have 100 patches - the Korg M1 and T3 come to mind. So do a lot of older home keyboards. My Casio CTK-601 has 200 patches.

Gerinski wrote: [QUOTE=awaken77]  I mean, if I have a memory of say 2 Mb, which if I'm not wrong is 2048 Kb, and a memory patch may need (say for example, I have no idea how much) around 20 Kb, that would still give a capacity for 100 patches, the relationship to the binary series is lost by the magnification, isn't it?

I'd guess that the first polyphonic synth used on a lot of recordings was the Prophet 5? Or maybe the Oberheim 2 voice or 4 voice? Plus the 'string machines' which were fully polyphonic. So the eight voice machines didn't come until later.

Gerinski wrote:  The Prophet 5 came out in 1978, if I'm not wrong the first commercial polyphonic synth was the Polymoog which was fully polyphonic (1975) and the Yamaha GX-1 (8 voice polyphonic). The Oberheims came very soon afterwards, as well as the Korgs.

Gerinski wrote: I'm totally ignorant about electronics and computing and although I can follow these arguments to a certain point, I'm not sure that I get them because I don't think that things like a memory patch or a voice utilize single whole units of memory or computing power. I mean, if I have a memory of say 2 Mb, which if I'm not wrong is 2048 Kb, and a memory patch may need (say for example, I have no idea how much) around 20 Kb, that would still give a capacity for 100 patches, the relationship to the binary series is lost by the magnification, isn't it?

There's basically two factors that say otherwise. One is that it's faster for a computer to address things on power-of-two boundaries; if, say, a patch fits into 29 bytes, you could store them sequentially in memory, but then the program would have to find the memory address of a patch by multiplying its number by 29 - and in small CPUs of the late '70s-mid '80s, there weren't even necessarily any instructions for multiplication (you could still do it, but you had to code the process yourself, and either way it was slow.) Whereas if you sacrificed a miniscule amount of memory and padded patches out to 32 bytes (a power of two,) you could much more easily get the address by taking the patch number and doing a shift instruction (the binary equivalent of multiplying a number by 10/100/1000 by tacking on zeroes to the end,) which basically any CPU has and which is always faster than multiplication. (This actually still holds true today, it's just that even small CPUs are fast enough that it's not such a big deal. When the processor in your polysynth might run at a maximum of 3-4MHz, it made a lot bigger difference.) Sure, you're sacrificing three bytes, but you can write them off as "for future expansion" or something...

The other factor is that, as awaken77 points out, once you've got the capability, why waste it? There's pretty much no common decimal numbers where you won't have room for more up to a binary limit. If you've got 10 patches, you're using a 4-bit index anyway; why not make it 16? 20? 32. 50? 64. Sure, you could settle for less (and some actually did - my Oberheim Matrix-6 only has 100 patch slots,) but those extra patches are value added at generally no extra cost.