How accurate is Fractal's amp modeling?

York Audio

Fractal Fanatic
Vendor
All I gotta say is that there's magic in FW15. So far I've been able to match all of my high gain amps, but yesterday I wanted to see how Fractal's AC30 compared to a '94 Korg Era Vox AC30 I recently picked up for a gritty Voxy tone with a tube screamer in front of it. The test was more about determining accuracy rather than dialing in a perfect tone, so I just threw a 57 in front of a cab to see how comparable the amp and the model were.

For anyone questioning the accuracy of Fractal's amp modeling, rest easy... you'll save yourself a lot of time and money. Maybe some of you can hear a difference, but if someone played me either of these clips on their own, I'd have no idea if I was hearing a real amp or a modeler.

 
"How accurate is the Axe Fx?"

"Yes"

To me that's virtually the same clip twice
Same here. I made it yesterday and it sounded nearly identical. I listened to it today with fresh ears and felt the same way about it.

The first sounds a little thicker in the low-mids. Would not be surprised if it's the AF3.
This tone took me about five minutes to match. I'm sure I could get it even closer with more time, but even the quick results are pretty impressive.
 
Not being an overly technical player these days, still, thought I would pick-up a Satriani song the other day. There I was - messing around trying to find an amp for that. ..??? ...What does Satch play!?? ...looked it up! ..a JVM.. As luck would have it, the AXEIII comes loaded with those! ..Sounded good to me right off the bat. Feedback, Harmonics, all there.

I ain't no Satch', but that amp sounds like him. ..and that's as technical as I want to get with the whole "modelling" ..situation.
 
Our Vox models are based on a fairly new AC30HWHD head. The model matches that particular amp very well. Will it match an AC30 from a different era exactly? Probably not.

I don't nitpick minor deviations in frequency response because no two amps the same. People listen to clips and it's always the same types of comments: "Clip A has more lower mids" or "Clip B has less high treble". Yawn.

What I do nitpick are the things that make a model sound and feel like a "real" amp. Anyone can get the static frequency response the same. That's easy. Do an EQ match. That's what everyone else is doing. Take a crude algorithm and do an EQ match to "fix" the inaccuracies. Yawn.

The hard part is getting the dynamic frequency response and dynamic gain response accurate. What makes a tube amp "breathe" and sound "organic" is the constantly varying frequency response and transfer function. That is extremely difficult to model accurately. It requires intricate knowledge of exactly how a tube amp works. Even profiling and AI approaches can't do that. All they do is learn a static transfer function. The problem though is that the transfer function is dynamic. The frequency response is constantly changing and the transfer function is continually changing as well. Our algorithms model that stuff. The frequency response and the transfer function are dynamic. The virtual power tubes for the AC30 models even go into Class-B operation if you drive them hard, just like the real amp. No black-box approach can do that.

Are our algorithms perfect yet? Probably not but in my not-so-humble and biased opinion 15.xx firmware was a significant improvement and has the "mojo" of a real tube amp.
 
Our Vox models are based on a fairly new AC30HWHD head. The model matches that particular amp very well. Will it match an AC30 from a different era exactly? Probably not.

I don't nitpick minor deviations in frequency response because no two amps the same. People listen to clips and it's always the same types of comments: "Clip A has more lower mids" or "Clip B has less high treble". Yawn.

What I do nitpick are the things that make a model sound and feel like a "real" amp. Anyone can get the static frequency response the same. That's easy. Do an EQ match. That's what everyone else is doing. Take a crude algorithm and do an EQ match to "fix" the inaccuracies. Yawn.

The hard part is getting the dynamic frequency response and dynamic gain response accurate. What makes a tube amp "breathe" and sound "organic" is the constantly varying frequency response and transfer function. That is extremely difficult to model accurately. It requires intricate knowledge of exactly how a tube amp works. Even profiling and AI approaches can't do that. All they do is learn a static transfer function. The problem though is that the transfer function is dynamic. The frequency response is constantly changing and the transfer function is continually changing as well. Our algorithms model that stuff. The frequency response and the transfer function are dynamic. The virtual power tubes for the AC30 models even go into Class-B operation if you drive them hard, just like the real amp. No black-box approach can do that.

Are our algorithms perfect yet? Probably not but in my not-so-humble and biased opinion 15.xx firmware was a significant improvement and has the "mojo" of a real tube amp.
I agree on the 15.xx firmware update. It really made a leap forward in my humble opinion. If I may ask @FractalAudio Cliff, do you have any recommendations on resources to learn more about the inner workings of some of the more popular amplifiers? I am an electrical engineer working from home due to covid and have been trying to gather as much knowledge as I can.
 
I agree on the 15.xx firmware update. It really made a leap forward in my humble opinion. If I may ask @FractalAudio Cliff, do you have any recommendations on resources to learn more about the inner workings of some of the more popular amplifiers? I am an electrical engineer working from home due to covid and have been trying to gather as much knowledge as I can.
http://www.valvewizard.co.uk/index.html
 
Our Vox models are based on a fairly new AC30HWHD head. The model matches that particular amp very well. Will it match an AC30 from a different era exactly? Probably not.

I don't nitpick minor deviations in frequency response because no two amps the same. People listen to clips and it's always the same types of comments: "Clip A has more lower mids" or "Clip B has less high treble". Yawn.

What I do nitpick are the things that make a model sound and feel like a "real" amp. Anyone can get the static frequency response the same. That's easy. Do an EQ match. That's what everyone else is doing. Take a crude algorithm and do an EQ match to "fix" the inaccuracies. Yawn.

The hard part is getting the dynamic frequency response and dynamic gain response accurate. What makes a tube amp "breathe" and sound "organic" is the constantly varying frequency response and transfer function. That is extremely difficult to model accurately. It requires intricate knowledge of exactly how a tube amp works. Even profiling and AI approaches can't do that. All they do is learn a static transfer function. The problem though is that the transfer function is dynamic. The frequency response is constantly changing and the transfer function is continually changing as well. Our algorithms model that stuff. The frequency response and the transfer function are dynamic. The virtual power tubes for the AC30 models even go into Class-B operation if you drive them hard, just like the real amp. No black-box approach can do that.

Are our algorithms perfect yet? Probably not but in my not-so-humble and biased opinion 15.xx firmware was a significant improvement and has the "mojo" of a real tube amp.
You once said, "listen to the tails as the sound trails off" (paraphrased) as a way of hearing the dynamic changes the transfer function. That's been my goto ever since for critical listening of anything modeled. It's very much where the good from the truly great modelers separate.
 
Justin did you use the Class A 30W TB model?
Yup. Top Boost vs Top Boost. I didn’t use any post EQ, but I bet if I had a high pass at 80-100Hz it would be nearly impossible to tell the difference. I might do a test at some point with both of my AC30’s and the model to see how different the two physical amps are and compare that with the model. I have a feeling they’ll all be really close. At the end of the day, I don’t feel the NEED to plug into my real AC30.
 
You once said, "listen to the tails as the sound trails off" (paraphrased) as a way of hearing the dynamic changes the transfer function. That's been my goto ever since for critical listening of anything modeled. It's very much where the good from the truly great modelers separate.

You can hear functions?! But jokes aside, care to elaborate on that?
 
Back
Top Bottom