Hardware vs. Software aka why do VST sucks?

[bugs]

Emulations are never as good as the original, even if they are better. It is an aestethic principle, not technological. VST instrument/effects developers need to stop emulating and start creating.

Software has the potential to be eversomuch more than hardwired one trick pony hardware units.

There need to be more artists and visionaries involved with the development of VSTs and fewer marketing exeutives.

Idea: every software developer should replace one salesman with one composer during the next year.

[Urs]

Hmmm, most hardware VAs run on fixed point dsp processors. This is *way* less resolution than ordinary single precision floating point

The opposite is true. Say, the range of values to be represented is [-1,1[. Then with 32 bit floating point data, the distance between adjacent values is at worst 1/2^24, namely at the borders of the value range, whereas with fixed point data, the distance is everywhere the same and at worst 1/2^31. So when the signal is near maximum, floating point data loses 7 bits precision in comparison to fixed point data.

That's not completely true. You're missing that the exponent changes within the range.

For single precision floating point:

...
23 bit between 2.0 and 4.0
23 bit between 1.0 and 2.0
23 bit between 0.5 and 1.0
23 bit between 0.25 and 0.5
23 bit between 0.125 and 0.25
...
same on negative scale

The popular Motorola dsps... what do they use? 24 Bit?

[edit] In fixed point they even have to *truncate* to add small scale numbers. The rounding erros are thus higher than the ones in floating point.

Cheers,

Urs

[Torben]

could it bethe "it's not the wand - it's the wizard" thing again? Hardware is made by companys like korg roland moog etc, who, probably, have 30+ years of knowledge and experience designing synths and DSP effects. VST plugs are often made by young and fresh companys... n00bs compared to above mentioned.

In some cases I think you are right. But in other cases I think it is the thought of that company having 30 years experience that might make you think they are better. In many cases those who do the development are young engineers who do not have 30 years experience. You do not automatically adopt the 30 years of experience the day you are hired in a 30 year old company, and you probably have no use for 20-30 year old know-how anyway.

But anyway this was not about analog emulations or modelling but digital hardware vs native plugins.

Torben

[pekadan]

Maybe another reason for the VST scepticism is that you don't use these plugins live usually. I guess that a lot of performing artists feel uncomfortable with gear they wouldn't take with them on stage. Software plugins is just a different world like. So when respectable artists talk about their favourite gear it'd be mostly about hardware synhts and boxes. Easy then to label this gear "pro" and label instruments and effects in the world of software "toys". It's not a question of sound quality necessarily. That's my guess.

Ok ok, I'm speculating. So shoot me!

[aciddose]

the answer to that question lies in the question itself. if you are asking such a question and in such a context, you obviously have made up your own mind regarding the answer already, and without any doubt i can say that is the reason itself. vst is no better or worse than hardware, it is only your perspective which has any influence here, and your perspective, based upon the question you have asked, is obvious.

infact, it would be better for you to instead make a statement, "i believe vst is worse than hardware" and leave it at that.

[Torben]

Hmmm, most hardware VAs run on fixed point dsp processors. This is *way* less resolution than ordinary single precision floating point

The opposite is true. Say, the range of values to be represented is [-1,1[. Then with 32 bit floating point data, the distance between adjacent values is at worst 1/2^24, namely at the borders of the value range, whereas with fixed point data, the distance is everywhere the same and at worst 1/2^31. So when the signal is near maximum, floating point data loses 7 bits precision in comparison to fixed point data.

That's not completely true. You're missing that the exponent changes within the range.

For single precision floating point:

...
23 bit between 2.0 and 4.0
23 bit between 1.0 and 2.0
23 bit between 0.5 and 1.0
23 bit between 0.25 and 0.5
23 bit between 0.125 and 0.25
...
same on negative scale

The popular Motorola dsps... what do they use? 24 Bit?

[edit] In fixed point they even have to *truncate* to add small scale numbers. The rounding erros are thus higher than the ones in floating point.

Cheers,

Urs

On top of this you need to scale (attenuate) before IIR filter procedures to avoid overflow when dealing with fixed point so you looses allot of your dynamic range on that account.

Torben

[aciddose]

also, you guys dont know how to use interger units.

i prefer to use integer calculation rather that naive floating point in my software, and many would state my software sounds better than yours. i wont state the obvious possible conclusions.

most importantly though, arguing about it will not make your software any better. you should start working on it instead.

my opinions about using integer (and NOT fixed point, which is a very outdated and mostly useless technique) calculation are based upon the engineering aspects of this form of calculation, and not upon any basic differences.

for example, you can do many more useful binary operations on integer numbers than you can on float numbers and also do them more quickly. it makes for faster and easier to design systems, and it is also for me atleast much more fun to work with engineering problems in digital logic and binary calculations than doing naive floating point math with conditionals.

the x86 supports many modes of integer calculation and infact has two seperate and pipelined integer units as well as single instruction multiple data opcodes.

since the 386 a 32 x 32 : 64 multiplication has been supported making 'fixed point' (operation on integers where the fractional part is shifted away) mostly useless.

flame wars on the topic are not very useful. i'd much rather hear from you 'float guys' on some useful techniques for tweaking float registers to produce desireable effects than anything else.

is it possible to do any fancy engineering with floats? i'm not aware of the answer to the question, i have always assumed it is; no.

i'd really like to hear of some cool things you can do with floats, i would definitely use anything useful.

[D-Fusion]

Go to www.Clavia.se & Download the demo of G2x modular program and see what you think (Even the most basic patch can bring your cpu to it's knees and that is in mono mode)

Then you will also hear the difference between vst's & Virtual analogs

[Torben]

also, you guys dont know how to use interger units.

i prefer to use integer calculation rather that naive floating point in my software, and many would state my software sounds better than yours. i wont state the obvious possible conclusions.

most importantly though, arguing about it will not make your software any better. you should start working on it instead.

That was some statement, when you say "you guys" do you mean everyone dealing with VST?

FYI we use fixed point in many parts of our algorithms. One of the great things about native is that you are not locked to fixed or float, as with all DSP's I know; you can do whatever you want when you want it. But anyway, a card like UAD uses a floating point processor so I am not sure that fixed point is better than floating point.

Torben

[projectdan]

Go to www.Clavia.se & Download the demo of G2x modular program and see what you think (Even the most basic patch can bring your cpu to it's knees and that is in mono mode)

Then you will also hear the difference between vst's & Virtual analogs

They have very very good reason not to make it efficient on a PC, it is a demo for their hardware they want you to buy. Does sound good though.

[aallvor]

Go to www.Clavia.se & Download the demo of G2x modular program and see what you think (Even the most basic patch can bring your cpu to it's knees and that is in mono mode)

Then you will also hear the difference between vst's & Virtual analogs

More than anything I think people will hear the difference between the average bedroom musos soundcard and the dacs in a £1500-1700 synth..

[whyterabbyt]

Go to www.Clavia.se & Download the demo of G2x modular program and see what you think

Its not actually an emulation of the G2X, and its incomplete, missing some of the G2's modules. But I already have the demo. I also have two G2 Engines.

(Even the most basic patch can bring your cpu to it's knees and that is in mono mode)

I bet you dont actually know why, though, do you?

Then you will also hear the difference between vst's & Virtual analogs

Unlikely; the real G2's would have to pass through two sets of digital conversion, so no genuine comparison is possible.

[michu]

GIMP is just as powerful as photoshop!

I'll keep that in mind next time I've got to work with CMYK.

heh, i could try to make separations in some other software (we used to use a command line tool for that years ago ) but i hardly can ignore missing white and black point controls in editing dialogs...

[spacefox]

Simple enough - if it's not exclusive, you don't get to be a snob.

Urs' Zebra is how much? 200$? And it is one of the best synths around, sound-wise and otherwise. Now this is something almost anyone can get* (not me though, I'm skint). What's the fun in that?


* not to mention piracy, which make it even less exclusive. How much can most people respect what they can get for free

[pitchdrifter]

Most 400 buck soundcards have better dacs than what's commonly used in hardware VAs.