What's new

Software reverbs vs. hardware (continued from Bricasti thread)

I would love to hear a side by side comparison between the Seventh Heaven and Bricasti. While I don't expect them to be identical by any means... I am curious how much difference there is in an actual mix.

When you think about it - all the money I've spent on this reverb and that reverb and this plugin... I could've purchased a Bricasti outright.
 
So. A digital reverb in a box has dedicated DSP.

In the late '90s through around 2004, one of the great topics of discussion was dedicated DSP vs. native processing (usually meaning that everything runs on computers rather than using Digidesign Pro Tools TDM).

There were a few arguments. The most convincing one is that dedicated chips were optimized for the task so they could do a better job. I'll come back to that.

The others were that they could use more horsepower for the task, they didn't have to compete with everything else running on a general-purpose computer, nobody would be interested in (for example) a reverb that took over an entire computer, dedicated chips guaranteed you could run x amount of processing without your computer system soiling the bed... all variations of "processing is limited resource on general-purpose computer."

Processing isn't unlimited today, but it's so far ahead of what we were using 15 years ago that it isn't funny. I have a pile of Pentium 4s in my garage to prove it.

But what about the first argument? Can dedicated DSP chips programmed with low-level machine language sound better?

I remember asking one of the engineers at Universal Audio when they first started producing plug-ins, and he said I was confusing the algorithms themselves with what they were running on. The algorithms were the same!

If he was right, the only difference you'll hear between a digital hardware box and a plug-in is the box's converters (unless you go in and out the boxes digitally, as I do in the PCM91 I have the good fortune to have).

Analog emulations are a totally different thing.

***
Bill, I think the movement you're talking about with the Lexicon is the same thing they always had: the Spin parameter. It's part of their secret sauce. Convolution can't do that, although you can chorus the end of the tail.
 
So. A digital reverb in a box has dedicated DSP.

In the late '90s through around 2004, one of the great topics of discussion was dedicated DSP vs. native processing (usually meaning that everything runs on computers rather than using Digidesign Pro Tools TDM).

There were a few arguments. The most convincing one is that dedicated chips were optimized for the task so they could do a better job. I'll come back to that.

The others were that they could use more horsepower for the task, they didn't have to compete with everything else running on a general-purpose computer, nobody would be interested in (for example) a reverb that took over an entire computer, dedicated chips guaranteed you could run x amount of processing without your computer system soiling the bed... all variations of "processing is limited resource on general-purpose computer."

Processing isn't unlimited today, but it's so far ahead of what we were using 15 years ago that it isn't funny. I have a pile of Pentium 4s in my garage to prove it.

But what about the first argument? Can dedicated DSP chips programmed with low-level machine language sound better?

I remember asking one of the engineers at Universal Audio when they first started producing plug-ins, and he said I was confusing the algorithms themselves with what they were running on. The algorithms were the same!

If he was right, the only difference you'll hear between a digital hardware box and a plug-in is the box's converters (unless you go in and out the boxes digitally, as I do in the PCM91 I have the good fortune to have).

Analog emulations are a totally different thing.

***
Bill, I think the movement you're talking about with the Lexicon is the same thing they always had: the Spin parameter. It's part of their secret sauce. Convolution can't do that, although you can chorus the end of the tail.

Interesting topic Nick. Can the platform the algorithms run on make a difference? You would think, since it is all ones and zeros coming in at different delay times, it would not matter. I think that maybe it is possible that Apple and Microsoft do not include something in the operating system because Intel does not have it in the chipset that was in the original boxes' CPUs from these companies like Lexicon and Bricasti used so it could make the difference that you discuss.
 
I think that the only real difference with the dedicated DSP chips is the latency where those boxes typically need to be usable "live" and in the computer we can afford to have a delay (even if it's only 64 samples). I'm not sure if a general purpose processor will ever beat that as it's just not programmed to do so. Other than that, then there's no need.

Having said that, people on here have said that the UA chips being used are very outdated and nowadays there's much more processing power available natively so we could easily run those plugins natively at a fraction of the % usage on those cards. In comparing to those older chips, perhaps our modern processors can offer a better performance even "live."

Can Bricasti make a plugin that's the same as their hardware? Certainly. But then people would think it's ridiculous for them to charge thousands of dollars for a single plugin so why not keep it in a black box to make it seem less ridiculous.
 
I'm pretty sure there's just one latency for all plug-ins. The audio is given to them, they process it, and it's given back.

It's not likely to be any different for external processors - never mind that latency with a reverb = predelay.
 
I agree. dsp in a box=plugin.
I also remember 2004 :)
I had the same argument back then.

The new slate digital video for the interface he really goes into this against UAD plugins and says dedicated dsp is very old school and that nowadays there is no reason for it technically. ( maybe needed for piracy issues)
but at least UAD pays licesing rights/split profit with the hardware emulations. but thats a whole other topic ;)

and of course, we all see this as operators of these "boxes" and not as the ones who program and build the hardware so our perception only goes as far as saying unit A sounds the same or similar as unit B and therefore the outcome is X.
 
I agree, nowadays dsp are dongles in a box and exactly the reason why i no longer support all those gigaenormous dongles inside the PC... the dsp cards are outdated inclusive for render farms.
 
FWIW, I read on GS that an M7 has about the processing power of a UAD Octo & quad combined. That's for just one stereo instance but apparently there are 4 (?) engins going on in the background. And this year with Version 3 updates there will be new algorithms and 3 times the processing power.
 
FWIW, I read on GS that an M7 has about the processing power of a UAD Octo & quad combined. That's for just one stereo instance but apparently there are 4 (?) engins going on in the background. And this year with Version 3 updates there will be new algorithms and 3 times the processing power.

i have the quad and i dont get a lot plugin power. it max's out very fast.

plus the ton of latency. which is why i dont use it a lot.
 
I'm pretty sure there's just one latency for all plug-ins.

I'm not sure what you mean. Different plugins have different latencies in order to accommodate the processing needed and balance that with what's reasonable to tax the cpu.

At a buffer 128 samples at 48k the latency is about 2ms which is probably what you'd typically expect from a hardware box. How much can you actually run with such a low buffer? A normal processor will struggle to keep up with that since it has to be doing a tonne of other things. The dedicated DSP chip will be optimized for that so it can easily handle it while very few people can probably run big sessions at only 128. Go up to 1024 and you can basically do whatever you want. You can have as much processing power as you need. It's just a matter of how long it takes to do which is where those chips excel. Depending on the latency you want from the Bricasti plugin, you'd need a computer dedicated to it while at 1024 you might be able to run 20 instances on a single computer.
 
I use an entire ADSP-21369 SHARC DSP chip for Phillips Bucket Brigade 12 tap delay that sounds as good as my MC5000 hardware algo.
I program it to be my Reverb even though it's not as realistic emulation of space as an M7 Bricasti.

The real benefit of FPGA and DSP Based FX is realtime parameter modulation.
Change the size of decay, gate the tails by sidechaining so regenerations don't muddy up a mix or solo.

But for static insert and forget FX there's plenty of Native plugs that work for recording.
Even if there's latency it's a Reverb.
Consider the latency a free pre delay.
 
I'm not sure what you mean. Different plugins have different latencies in order to accommodate the processing needed and balance that with what's reasonable to tax the cpu.

I remember hearing the word "frames," meaning plug-ins are handed audio by the host for x number of frames and then they have to give it back. Maybe a developer can straighten this out, but these days the latency is normally compensated so they're all in sync.
 
We are, I'm afraid, mixing all sorts of things, and while mixing is good in a musical sense...

Latency is inherent in EVERY processor - general purpose computer, DSP co-processor, or dedicated hardware. It takes a finite amount of time to convert from analog to digital, it takes a little more time to do the calculations, and then it takes more time to convert back to analog (I too use digital I/O for my PCM-90, so that saves a little bit of time???) The faster the processor the shorter the time to do the calculations... except that if you give a programmer more horsepower they'll use it.

The code itself - like it or not, there are advantages to DSP chips, they have instructions (the most famous being multiply and accumulate) that are not present in general purpose processors, and they make it easier (and more efficient) two write DSP code.

A Field Programmable Gate Array (FPGA) is even faster because it is, for all intents and purposes, a state machine, or at least it can be.

It is true that a typical general purpose computer can perform DSP tasks, it still won't be as efficient, which is one of the reasons some people like dedicated hardware, or even co-processors better.

I think it is probably true that any given signal process can be done on a general purpose computer, and if the programmer is really good it will sound the same, or maybe a better way to say that is any given process can be done on a general purpose computer, a DSP co-processor, or dedicated hardware with minimal if any audible difference. But you better be a pretty darned good programmer.

Now let's think about which projects will attract the best in the talent pool. I think some talented programmers will gravitate towards native programming because it is an interesting challenge. But I also think that more folks interested in DSP will gravitate towards DSP co-processors and dedicated hardware.

Another twist - all of my dedicated hardware has power supplies, analog stages, memory, clocks, and other sub-systems that make up the final package. And the sound is the sum of all those subsystems. Design a shoddy power supply and you will corrupt the audio, and the processing. Same goes for clocks (don't talk to me about clocks, they'll drive you crazy!). The analog stages are as important as any other section, and some companies still have a clue about how to design a proper analog stage... some do not.

There is a reason (beyond "what the market will support") why a Bricasti M7 commands $3000 when I can build a pretty fast computer and load it with UAD cards for less. Better parts, fantastic A/D and D/A, well designed analog stages, a well designed power supply, and however much processing power they needed to make it work. And that will continue to be worth the price to some and not to others.

There is nothing inherently wrong with native processing. I have lots of native plugins that I absolutely could not do without (well, I could, but you know what I mean!). I think the guys at SoundToys have some of the best processors available, regardless of platform. And they've clearly made the choice to pursue native processing, on multiple platforms no less.

There is nothing inherently wrong with a DSP co-processor, and I'd be just as lost without my UAD cards and Apollo interface.

There is nothing inherently wrong with dedicated hardware, I'm not giving up my PCM-90, and I'd buy an M7 in a second if I could justify the cost.

They are all different ways to get to (roughly) the same result. In April 2017 I believe the pecking order remains dedicated hardware, DSP co-processors, and general purpose computers. And we (the audio community) are not big enough for companies to add DSP co-processors to motherboards or (oh the beauty) the cpu.
 
Ah...you left out the natural sound of the space.

Which could be followed by the emulation of space as in hardware.

Which could be followed by DSP/FPGA emulations of the emulation.
Then Native which is an emulation of the emulation squared...

So a big 8 x 12 Sewer or a 12' circular sewer gets lots of attention.
 
I'd buy an M7 in a second if I could justify the cost.

Wouldn't you rather get (or have the option to) an M7 plugin for $500 and trade off a couple of ms of latency? Most of us are probably already working with buffer sizes large enough to accommodate the trade off of the less efficient processing (using for a live setting would be a completely different beast). Assuming that you were to use such a box with just digital then there's no need for the A/D D/A, clocking, or power supply so all of that money would be invested into your interface. I'd much rather spend $10,000 on the best converters, clocking etc. and use the plugin with additional latency than spend $3000 to have that hardware locked to that processing.
 
M7 is great but still sounds like a digital reverb. So many other great ones - don't see anything special about Bricasti to justify the cost. Plugin for $500? I would go for $199.

Once fashion changes and sales start to decline, it will be released as a plugin. Just a matter of time.
 
Ah...you left out the natural sound of the space.
You speak the truth!!!

I had spring reverb tanks on my guitar amps, of course, and I figured out how to patch into and out of them, of course, but I never really considered that 'reverb'.

I walked into my first recording studio in the mid 1970s, they had a couple of spring reverbs that were quite different than what I was used to (a BX something or other and a rack mount, maybe MicMix?). One of the first projects they put me to work on was mounting a plate reverb. Oh my... what a glorious sound!

A couple years later, in college, I became friends with a studio owner who had a chamber. To this day I've never heard anything quite like it. My next studio will have a chamber!

If we're going to compare any artificial reverb to a chamber I think we're all going to be really sad...
 
Top Bottom