Is software amp modeling *there* yet?

  • Thread starter Quitty
  • Start date
  • This site may earn a commission from merchant affiliate links like Ebay, Amazon, and others.

Rev2010

Contributor
Joined
Mar 28, 2005
Messages
6,330
Reaction score
1,498
Location
New York, NY
Using an analogy again, a single line reeeeaaaally fast magnetic railway won't do the job as well as a slower but 24 ways highway. Going fast isn't the only issue, it's doing a lot of the same at the same time reasonably fast that is.

Of course, but that is why we have processors with numerous cores, both in CPU's and GPU's. In my industrial electronic project to which I write all the music I use all software synths now, VST's, except for one hardware keyboard that I'll probably never part with (Yamaha AN1X which is also my controller). In one song we have I have about 24 VST synths loaded with several playing at a time throughout the song. In addition to that I have numerous VST effects running on many of those synth's mixer channels doing various things - FX, dynamics processing, EQ, etc and the CPU meter is just about at 50%.

If my machine can handle all that running at once there's no doubt in my mind it can handle a single, or even several, flawless guitar amp sims running at once. ;)

Oh, and I run at 2.5ms latency. I have an Intel Core i7-920 (the early original i7 model).


Rev.
 

This site may earn a commission from merchant links like Ebay, Amazon, and others.

Quitty

Hates 'mojo'
Joined
Feb 27, 2011
Messages
1,379
Reaction score
142
Location
Germany
This isn't about sounding like a production.

I'm hooking up something that's meant to emulate a preamp, or an amp, into a poweramp with a cab - and it doesn't feel or sound like an amp does.
I was under the impression that the AFx does that. The POD certainly claims to.
 

exclamation-mark

Well-Known Member
Joined
Apr 1, 2011
Messages
71
Reaction score
3
Location
Perth, Australia
My guess is that Fractal Audio isn't named Fractal Audio just because it sounded good. ^^ Cliff is likely a very high level mathematician.

I've heard rumours about Cliff using some sort of fractal algorithms. Apart from that fact that this doesn't actually make any sort of sense, it sounds pretty convincing!

But yeah, on those other points I mostly agree. Even with today's hardware we can't run a naive (no FFT's, that's cheating!!) convolution implementation (convolution is the process used for speaker impulses) at 96kHz that's more than 2 seconds long without running out of processing power or memory bandwidth. A naive implementation of discrete convolution is a really simple thing (it's literally just multiplication and addition). Unfortunately if you have a long convolution sum, you have several billion operations on your hand every second.
 

Spinedriver

Well-Known Member
Joined
Jan 22, 2007
Messages
3,074
Reaction score
885
Location
NB,Canuckistan
I wonder how one of these would compare to Line 6/Fractal/etc....

MuseBox_main.jpg


It's a rackmount processor but uses pc software (Peavey's ReValver software) but from what I understand you could also load in LePou or Amplitube or similar.
 

sevenstringj

Banned
Joined
Jul 19, 2008
Messages
3,054
Reaction score
231
Location
ny
I wonder how one of these would compare to Line 6/Fractal/etc....

MuseBox_main.jpg


It's a rackmount processor but uses pc software (Peavey's ReValver software) but from what I understand you could also load in LePou or Amplitube or similar.

It can only run 2 plugins at a time. Seems like a waste to me. Maybe a few more years from now they'll have something more powerful and cheaper.
 

Andromalia

Pardon my french
Joined
Dec 24, 2009
Messages
8,750
Reaction score
3,215
Location
Le Mans, France
If my machine can handle all that running at once there's no doubt in my mind it can handle a single, or even several, flawless guitar amp sims running at once. ;)

I wasn't talking about highways in the sense of using multiple VSTs, I was talking about modeling a single instrument. All the sound subtleties are the simple alogirythms but there's an awful lot of them, and calculating that follows the usual exponential progression the more details you're asking for.
The above detail about convolution is true: a "simple reverb" can crash your PC in seconds if you go overboard with it. So is modeling a cab. So you have limitations imposed by the hardware, as the NASA likely won't allow you to use their stuff for a guitar cab :D those limitations are what make modelers sound "fake", and the technology evolving, we're getting "less fake" as we go, to the point where our ears start to have problems picking up the difference.

I'm not saying it can't be done, but if you have only a handful of products that actually work, there's a reason, imho.

Note that the above and all isn't fact or truth, just my understanding of the thing and I do have some knowledge of how computing works.
 

PodHdBean

Well-Known Member
Joined
Jul 16, 2012
Messages
125
Reaction score
0
Location
NewYork
peavey revalver sounds pretty good to this day..ampitube just sound horrible for distortion ...guitar rig also just falls apart with hi gain.
 

Rev2010

Contributor
Joined
Mar 28, 2005
Messages
6,330
Reaction score
1,498
Location
New York, NY
I wasn't talking about highways in the sense of using multiple VSTs, I was talking about modeling a single instrument. All the sound subtleties are the simple alogirythms but there's an awful lot of them, and calculating that follows the usual exponential progression the more details you're asking for.
The above detail about convolution is true: a "simple reverb" can crash your PC in seconds if you go overboard with it. So is modeling a cab. So you have limitations imposed t by the hardware.

I know, I wasn't only talking about multiple Vst's specifically, but rather how a cpu can handle all those floating point calculations going on at once in near real time. The fact that we already have plugins that can emulate a guitar amp extremely well and still allow dozens of instances and still have cpu left to spare shows a PC can do it if it got more complex. Same goes for convolution, there are plenty of usable plugins that don't crash your machine and can even be run with a few instances in the project.


Rev.
 

Matt_D_

Well-Known Member
Joined
Jun 26, 2010
Messages
600
Reaction score
28
Location
Seattle, WA
I'm not so sure a Pc is up to doing what the axeFx does. If I got it well, the axe fx uses some chips dedicated to audio a PC is missing. the software isn't really the issue: you need hardware to process it and sound has its speciic requirements, most notably dealing with latency. A PC (or a mac, for that matter) is a multipurpose generalist hardware probably unsuited to the task. That's why we have sound cards. The axefx is basically just an extremely powerful and specialised sound card piloted by Fractal programs.
Software is important but you cant dissociate it from the hardware, else Fractal wouldn't bother to manufacture such a complicated unit if all they needed was a dongle.

the axe fx uses a couple of TigerSHARC cpus, which are designed to be dedicated DSP chips providing decent performance at a reasonable cost point. your PC's cpu will be an order(s) of magnitude faster, not to mention that you'll have much more working ram and higher sample rates available.

I'd be very confident that the boys at fractal have a working PC version of their software. They may choose not to release it due to piracy concerns, and/or latency issues with external soundcards (they get supremely low latency in their unit).

with exceedingly long instruction pipelines, and slow memory access (compared to internal cache access) linear bulk processing is where current modern CPU's work best, DSP/audio processing falls neatly into that use case.
 

Quitty

Hates 'mojo'
Joined
Feb 27, 2011
Messages
1,379
Reaction score
142
Location
Germany
The impulse response issue is a tad off topic here, but for the record -
Real time computation is indeed different from "real" convolution and not because of complexity but because the term 'convolution' relates to multiplying a function (your signal) by another function (your impulse response) in reverse - meaning your (A-Z) to the IR's (Z-A).
Obviously, this can't really be done without having the complete function at hand, so you can only do this once you've finished recording.

With that said, compared to true convolution, real time convolvers like Kefir have less than 1% error rate. that's <1% THD and that's lower than an actual cab will have.
By the way, while on the subject, Lepou's Lecab has an unusually high error rate. So high, in fact, that i suspect a built in HPF and LPF.

peavey revalver sounds pretty good to this day..ampitube just sound horrible for distortion ...guitar rig also just falls apart with hi gain.

Now, this is interesting.

For someone who calls himself 'PodHdBean', what's your take about software vs. POD?
 

larry

". . . . . . . ."
Joined
Apr 30, 2007
Messages
1,165
Reaction score
85
Location
npr fl
TL;DR

I'd like to see FAS pciex card and software bundles. As far as money, drm the shit out of it all like avid.
 
Top