How much of "feel" can be fixed with better front end/converters?

[re-animator]

srsly?

Great algorithms on a great platform aren't going to work like an amp if the a/d conversion is bad, or if the front end is crappy. That's a given.

From my own experiences i would say that most modelers contain good algorithms (while some have great ones). The "feel" is where they fall flat, blah blah blah.

i'd have to assume in some ways lack of dynamics can be attributed to poor modeling within the algorithms, but i have an inkling that the majority of the problem in most systems lies in the hardware. After all, every digital rig contains the following:

a front end (where your guitar is plugged in)
a/d converters
a cpu
and the algorithms themselves.

naturally, we can't really quantify accurate modeling. we can quantify tech specs though. Most obvious is CPU power. there wasn't really much if any tweaking of the algorithms between pod xt and x3, but the higher available processing power accounts for a difference in sound (albeit, not a huge difference). The pod x3 uses a 266mhz sharc processor. IIRC, axefx has a dual core cpu with something like 10x as much processing power. That's gotta help with perceived latency/feel as well as the sum total of what a high end unit like axe can do in terms of amps, effects, routing, etc.

then there's the a/d converters and the actual input itself. I would guess that sub$500 units like the pod (and worse among older units) are just not well equipped with good converters, where as high end stuff like eleven and axe are better suited to the task. I would expect similar from the cheap front ends used in cheaper units versus expensive ones.

This perhaps is also why there are such varied receptions towards modelling software like amplitube, revalver, etc. You can maybe plug into an apogee with great conversion and love how the thing plays, or use an out of date $100 usb soundcard and wonder how anyone deals with modelling crap.


I don't have any concrete frame of reference for this theory, but i would bet that pod farm platinum sounds/feels much better with a great interface than a pod xt/x3 on its own. Conversely, i'm sure an Eleven Rack sounds/feels much better than using the protools eleven plugins with a crappy interface.


anyone confirm or deny? discussion?

[VTR]

I think you are on to something. Last weekend my son and I compared mic pre's, and various mics. It all started when I bought a Grace Design mic pre. He bought over his Steinberg unit and I have a Motu unit. We boiled a lot of what we heard between mics and pres to the converters in each unit. I ties into exactly what you are talking about.

[Axe-Man]

The DSP power differences are apparently not that huge (between the X3 and Axe-FX).

X3 DSP: SHARC ADSP-21369 at 266MHz, 1.6 GFLOPS
XT DSP: SHARC ADSP-21065 at 60 MHz, 180 MFLOPS
Axe-FX DSP: TigerSharc at 500MHz, 3.0GFLOPS

The XT lags a long way back though.

The X3 if it's clocked at 400mhz (the spec ranges from 266mhz - 400mhz for the SHARC) comes out with 2.4 GFLOPS.

[Brick_top]

Most importantly I think is how are those FLOPS used.

[Axe-Man]

Most importantly I think is how are those FLOPS used.

Definitely...300 HP makes a lightweight car accelerate pretty quickly but dump that into a Hummer and well...it's nap time.

[AdamCook]

A bad front end design aka "cheaper converters" could contribute to a decreased SNR ratio. That could make the modeler sound worse from the standpoint of lower fidelity...aka less high end clarity. The first-gen POD's for instance didn't sample at CD quality like today's modelers do. It wouldn't really have much impact on "feel" though.

The "feel" that you're talking about is based on things like the transient response of the model to the initial pick attack and how the amp transitions from linear to non-linear (aka "edge of breakup" type sounds). That transient response is a function of the overall latency of the system and the algorithm used to model the amp. A faster DSP means you have more CPU cycles per block of samples to work with so that helps greatly with latency and "feel". But of course you also need the right algorithms.

[stvnscott]

Much of the 'feel' comes from the preamp. For the best performance, you need an analog preamp that emulates the 1st gain stage in a tube guitar amp. Obviously, the quality of the gear and the though put into its design makes a great deal of difference.

[re-animator]

a lot of good info in this thread.


to me, the sounds in stuff like pods seem like they were designed for plugins. as if they were made for re-amping a dry signal. Switching pickups and playing with the volume knob doesn't work the way it works with an amp, and that's what the let down is (otherwise i think most modelers perform admirably, especially if you consider that they don't move much air).

[torquil]

The DSP power differences are apparently not that huge (between the X3 and Axe-FX).

X3 DSP: SHARC ADSP-21369 at 266MHz, 1.6 GFLOPS
XT DSP: SHARC ADSP-21065 at 60 MHz, 180 MFLOPS
Axe-FX DSP: TigerSharc at 500MHz, 3.0GFLOPS

The XT lags a long way back though.

The X3 if it's clocked at 400mhz (the spec ranges from 266mhz - 400mhz for the SHARC) comes out with 2.4 GFLOPS.

Graphics card in my cheap new laptop: 140 GFLOPS :)

[Brick_top]

Graphics card in my cheap new laptop: 140 GFLOPS :)

As far as I know graphic cards are extremelly "multi-threaded" and real-time audio isn't.

[torquil]

As far as I know graphic cards are extremelly "multi-threaded" and real-time audio isn't.

Sure.

On the other hand, the CPU (dual-core T4400) can do approx. 16 GFLOPS.

I don't know if that is per core, or 2 x 8 GFLOPS

[journo]

Hi,

Had an interesting discussion with Paul Rivra a ew years back about lack of feel and dynamics in digital modelers. He said i was because there wasn't enough effort invested in the input stage of the products.

The two best feeling digital amp modellers I have tried have been the Hughes & Kettner ZenTera and the Digidesign 11R and both companies have stated that they put a lot of thought and effort into the input section. I haven't had a real chance to try the Axe-Fx since v 8.x so it's poiible that is equal to these other products. Recent sound clips tells me it may well be.

Cheers,

Mats N

[cliffc8488]

Sure.

On the other hand, the CPU (dual-core T4400) can do approx. 16 GFLOPS.

I don't know if that is per core, or 2 x 8 GFLOPS

Modern desktop CPU's boast all kinds of impressive GFLOP numbers, and they can achieve peak performance fairly close to those numbers under optimum conditions, i.e. "off-line" processing of vectored data.

Unfortunately, the actual performance is only a small fraction of the advertised value when processing real-time data (of which audio is inherently).

We've ported the Axe-Fx code to the Intel x86 architecture and on my 2.66 GHz Core i7 (quad core) machine the CPU usage is nearly identical to an Ultra. The i7 boasts something crazy like 50 GFLOPs but it's actual performance for real-time audio is only several GFLOPs.

Dedicated DSP chips always run close to their theoretical performance because that's their whole point of existence. They are specifically designed to process real-time data as opposed to a general purpose CPU which is designed for non-real-time multitasking.

[TowMater]

Modern desktop CPU's boast all kinds of impressive GFLOP numbers, and they can achieve peak performance fairly close to those numbers under optimum conditions, i.e. "off-line" processing of vectored data.

Unfortunately, the actual performance is only a small fraction of the advertised value when processing real-time data (of which audio is inherently).

We've ported the Axe-Fx code to the Intel x86 architecture and on my 2.66 GHz Core i7 (quad core) machine the CPU usage is nearly identical to an Ultra. The i7 boasts something crazy like 50 GFLOPs but it's actual performance for real-time audio is only several GFLOPs.

Dedicated DSP chips always run close to their theoretical performance because that's their whole point of existence. They are specifically designed to process real-time data as opposed to a general purpose CPU which is designed for non-real-time multitasking.

Well I think a lot of people would be willing to upgrade their computers to a 2.66 GHz Core i7 if you were to develop Axe-PC!!

[Guitar Vilain]

My Line6 TonePort UX2 sounds awful with GearBox (software plugin equivalent to a PODxT), but sounds (and feels) incredible with Revalver MkIII. So I'm guessing converters are not the defining factor for feel.

[torquil]

We've ported the Axe-Fx code to the Intel x86 architecture and on my 2.66 GHz Core i7 (quad core) machine the CPU usage is nearly identical to an Ultra. The i7 boasts something crazy like 50 GFLOPs but it's actual performance for real-time audio is only several GFLOPs.


That's some really interesting info.

So if I understand you correctly, this would not be significantly improved by moving to a realtime OS on the same CPU, e.g. a realtime Linux system or QNX (the only ones I remember the names of), since the CPU itself is not optimized for realtime DSP?

[cliffc8488]

Correct.

[torquil]

Correct.

Thanks!

[Capn Spanky]

So I'm guessing converters are not the defining factor for feel.


I think you're right. In my limited experience, the difference between high end convertors and recent "pro-sumer" level convertors is extremely subtle and hard to hear unless you have great ears and a pristine listening environment.

[octatonic]

Until I went Axe FX (which is awesome) I was using a POD Pro XT.

Using the digital IO on the Pod Pro XT is a much better option than using the analog converters within the unit if you have good quality converters.
The overall realism of the models was very much compromised using the internal converters in the POD.

[re-animator]

Until I went Axe FX (which is awesome) I was using a POD Pro XT.

Using the digital on the Pod Pro XT is a much better option than using the analog converters within the unit if you have good quality converters.
The overall realism of the models was very much compromised suing the internal converters in the POD.

interesting.

this recent software modeler post also confirms that:
http://thegearpage.net/board/showthread.php?t=734553

i'd really like to do a test of my own soon. unfortunately i'm away from the bulk of my gear in new york. I do have my pod x3 on me... anybody in san diego with a nice audio interface want to have a little shootout?

[cliffc8488]

The X3 has good converters. The other components driving the converters are average though.

[stvnscott]

Also keep in mind that converters are part of a system. If the implementation is not done well, designers can make the best converters sound like crap. The converse is also true. There are a few dozen other components involved in the AD conversion process. The quality of these components is as important as the converters themselves. And making the input stage guitar-centric is incredibly important.

[spentron]

Most of these boxes can be set to do nothing but digitize and then you can listen to that. With the RP155, it's about like a somewhat poor bypass on a single pedal... at least as good as some older rack units that were far more expensive. Since it's a multi, you can add as many FX as you want and not increase the basic loss.

[Pietro]

How much difference does this make in the mix with the band? Not trying to be argumentative (well, maybe a little...), but is that one more box you want on stage to worry about breaking down? The beauty of the Axe, and the X3Live for that matter, is the ability to do so much without worrying about all those cables and cords and power supplies and such.

So you can rock out without worry.

[lv]

My biggest concern when getting the Axe-FX (standard) was that the feel and response to subtle changes in pick attack wasn't going to feel/respond right. Happy to report it not only feels like a tube amp, but the touch response is great, and better than many tube amps I've owned (not all).

I'd love to hear Glen Kuykendall play his 59 LP through the Wrecker amp.....
http://www.youtube.com/watch?v=Y2_12Ler9B8

[tonesville]

My biggest concern when getting the Axe-FX (standard) was that the feel and response to subtle changes in pick attack wasn't going to feel/respond right. Happy to report it not only feels like a tube amp, but the touch response is great, and better than many tube amps I've owned (not all).

I'd love to hear Glen Kuykendall play his 59 LP through the Wrecker amp.....
http://www.youtube.com/watch?v=Y2_12Ler9B8



seconded.....in the feel dept. you can really do some things that are "exaggerated" like silly amounts of sag on some models or turn a perfectly great blackface into a goofy sounding supro (tonestack frequency parameter all the way left)complete w/ oval speaker!
I've found that I can get more overtones/harmonics on cleaner setting than is "normal" something that I was always looking for.

[MarkF786]

So if I understand you correctly, this would not be significantly improved by moving to a realtime OS on the same CPU, e.g. a realtime Linux system or QNX (the only ones I remember the names of), since the CPU itself is not optimized for realtime DSP?Correct.

That's really surprising. The Korg OASYS, which I used to own, only contained a 2.8 GHz Pentium 4 yet it could do some of the most complex synthesis with high polyphony (up to 172 voices), 12 insert effects per voice (along with master effects, EQ, etc) and 16 tracks of recording and/or playback - and all of this simultaneously. They were using a realtime Linux OS, and the OASYS software was highly optimized to work with the specific CPU.

I'm not doubting you since you're the expert, but it's surprising that the Axe-FX would use more CPU cycles than a Korg OASYS running a full-throttle - and the OASYS was using an old Pentium 4, much less powerful that a quad core i7.

I know you can't speak directly to the Korg OASYS, but do you suspect the Axe-FX uses more CPU resources than it? When you ported the Axe-FX code to an Intel architecture, was it optimized for it or was the existing code just compiled as-is?

Thanks for sharing the info!

Mark

[cliffc8488]

Rewritten using the IPP library.