Huh?

So why would anyone want to spend $3300 on a PCM 96 anymore, when as Michael Carnes (who had a part in coding it) says that the plugins are a direct port of the algorithms?

Analog workflow, offloading CPU processing, haptic UI (turning real knobs), studio reputation (lex on the website) or just a personal preference beyond rational reasons…

…or just if you find a used PCM96 for less than 1.5k and cant resist :-)

Fair enough - maybe the Lexicon PCM example I chose doesn't fly (it was just the first applicable thing that floated up from my distant memory).

I still stand by the sentiment of my point though, and it's not the only such promises of "this can't be done in a plugin" until the marketing people wanted to sell you a plugin that's happened...

I was definitely running plugin reverbs earlier than 2009 though. Early through mid 2004 (although I did also get a UAD1 for extra CPU grunt - the EMT140 Plate was an early emulation I got for that.) Logic's PlatinumVerb came in Logic 4.x-ish, and that would have been around 2000 - and that was a good example of the kind of reputation in the box reverbs had in the early 2000s - not really that great. Five years on though, and I was definitely using much better sounding in-the-box reverbs and the promise was there - at least from my recollection - eg Logic 6.3 (2003) introduced Space Designer, and that wasn't even the first convolution reverb plugin around at that time.

Ironically, I'm still on a 2011-vintage machine, I wonder what the CPU requirements of a Lexicon PCM plugin are compared to say Platinumverb..?

Edit: Interestingly, ten PVs in series gives some, ah, *interesting* sounds. More like an angle-grinder than a reverb ("How much more metal could it be? None. None more metal!"), but cool nevertheless!

Vibe. Mojo. The indescribable special something that happens when binary numbers are turned into millivolts and (after a short journey through some commodity electronics) back again.

Valhalla Plate is one of my go-to's, along with UAD's EMT250, Softube's TSAR, and even Soundtoy's Little Plate. Clearly I like the the EMT vibe; Plate as well. What's surprising is that the EMT250 is not at all new, and it uses little DSP. Yet on Masters and Remixes (just a touch of course,) it sounds awesome!

The EMT250 was the first commercial digital reverb, so the CPU usage shouldn't be that high. Honestly, getting ANY digital audio out of a device in 1976 was a pretty amazing achievement. Universal Audio had the help of Barry Blesser, the original author of the EMT250 algorithm, when making their plugin, so it makes sense that it sounds great!

Lots of other super sharp folks at UA, of course. The EMT140 from UAD is pretty darned old at this point - I think it dates from 2004. Yet it still sounds fantastic. I think Jonathan Abel and David Berners were the heavy lifters for that algorithm.

What was a conversation turned into a roundtable. Good for you. I hope your future products do well. You seem to be ahead of the curve. Cheers.

My. How technically concise. Blind test anyone? Sorry . . . I'm not denying you or others might hear something different, but - you know - confirmation bias, blah blah blah.

Either I was being too subtle, or your sarcasm meter might need attending to.

I'd say plugin reverbs are getting pretty close, especially when we talk about old hardware vs. plugins. 480, TC6000, EMT250, etc..

But what about Multi-FX hardware reincarnations? Do we have anything similar to PCM80/81, Korg A1, Ensoniq DP4/+, even Lexicon MPX1? Or my favorite Yamaha D5000, which is "just" a delay?

It seems like that algorithms on these units are way more complicated than just reverb parts of them. I'd really like to see those algos in plugin format. Eventide H3000 hardware presets, while not exactly the same in sound, sounds similar to their plugins. But Eventide approach is just to release a part of the complex hardware as plugin, not a whole device.

From the most recent plugins Tai Chi was a big step ahead, very modern, with a taste of vintage

If you need a 5.1 hardware verb. The plugins are stereo only.

The new Casseopia and Orion patches from Valhalla for Supermassive are pretty amazing, if you haven't downloaded them yet. Gorgeous.

I generally use analog reverbs to a large degree when they are good. However, there's still a place for convolution. Logic's Space Designer is fantastic; I own Live Standard, thus do not have their convolution reverb, but I really don't want anything to do with MAX. I too have little need for a third party convolution model. I'm contemplating Valhalla Rooms or Vintage, but there appears to be much overlap with TSAR, which I do happen to like a lot. I have BX_Rooms, but never use it. It's way too freaking complicated.

Do you own a real plate reverb (like an EMT140) and is your studio equiped with a real echo chamber (like the one from Capitol studio in LA)? I am super jealous because every other type of reverb would actually be a "digital" reverb.

Wunderverb 3

I suppose I could say digital emulations of analog sound as the descriptor. I wish I owned the actual hardware, but do not. Sean's stuff, UAD, PA, Softube, and others make it simpler to Produce in a more simplified format with their powerful emulations.

Probably means algorithmic instead of analog.

I have to push back on this a bit. Fundamentally Core was an improved version of Pentium 3 that could be clocked much higher and had superior SIMD capability. The cache plays a factor, but is still just an incremental improvement, not an order of magnitude game changer.

I'm going to make the claim that it would have been possible to make Lexicon quality algorithms in a plugin by 2001, if we define "Lexicon quality" as 480L HD cart (utilizing all four Lexichips).

By that time Pentium 3s running at close to 1 GHz were widely available. If we allocate 30% to the reverb (it is afterall "Lexicon quality"), that means 6250 cycles per sample and around 100 uncached memory accesses (assuming 70 ns latency for main memory). For comparison, we know from the Reverb Subculture thread that 480L was capable of 320 simple arithmetic operations and 320 memory accesses per sample.

By structuring the code that it processes the reverb in blocks of 64+ samples, the majority of accesses are going to be cached, bringing memory bandwidth to parity or better. We have about 20 cycles per each 480L instruction which is more than generous, considering all the code needs to do is perform one memory read / write, one integer multiply and one integer addition. Let's call that 15 cycles (bearing in mind that P3 is an out of order cpu capable of executing up to 3 instructions per cycle, 15 cycles allows for plenty of circular addressing overhead and for "slow" 4+1 cycle multiply + shift operations). That still leaves 25% free for calculating modulation (originally done by four Z80s IIRC) and other generic overhead. Emulating the saturation and quantization artifacts might be a bit too much, but then again, PCM92/96 didn't emulate those either (AFAIK).

After doing that back of the envelope math, it's fairly obvious that processing power itself was not the limitation, but expertise (for smaller developers) and market conditions (killing high margin hw sales, piracy).

Is piracy really an issue? It's reminiscent of something that occurred in the 90's. Perhaps that's what you're referring to in your long Post. Bullet Points might help.

It was in the 00s. Those were the days of Waves anti piracy stings etc.

Lexicon and other high end companies certainly asked themselves "Do we really want to risk our flagship products being copied by everyone and tanking the high margin high end sales for good?" That's largely why UAD still uses their DSP cards (which haven't been able to compete with a modern multicore cpu in processing power for a decade but act as fairly good copy protection dongles).

E: Another reason was avoiding making the brand appear too cheap (as confirmed to me in person by one of TC's head R&D guys in the mid 00s). There are a lot of people who think that if something is expensive, it must be good and conversely if something is cheap or widely available, it cannot be truly high end.

Any thoughts on the Pentium IV versus 3? Most developers I knew in the early to late 2000s were using Pentium IV desktops, with maybe a 3 in the laptop. My early VST work (2004) was on Pentium IVs, and the performance was not great. I always liked the Pentium 3 laptop I was using in the early 2000s, but that never moved past Windows ME.

As far as the Core processors being a game changer, I'm going by my own experience moving from Pentium IV running Windows XP to a Core 2 Duo MacBook Pro. The MBP was a hugely more capable machine than the Pentium IV, both as far as programming my own DSP, and running other DSP in both OS X and Windows via Boot Camp.

Not sure if anyone would have been comfortable releasing a reverb in the early 2000s that used 30% of the CPU. Even today, CPU usage is a big factor in plugin reverb design. There's a big difference between running a single reverb instance in a DAW, versus running a whole bunch without running out of DSP.

This works for certain reverbs, but not necessarily for 480L stuff. The 480L and earlier Lexicons processed things sample by sample, and some of the techniques used in 480L algorithms would be difficult to implement when using blocks of 64+ samples.

Plus, not everything gets more optimal with larger block sizes. Generally speaking, things do, but there tends to be a "knee" where you get the best tradeoff of block size versus efficiency. Often, this lies somewhere between 16 and 32 samples. Doesn't really disprove your point, and I'm sure you've found this in your own work, but I figured I'd mention it.

Expertise was certainly a factor. Smaller developers tended to come from more of a DIY background, or at least from an environment where they weren't learning the reverb secrets on the job. You can follow the forum posts of various individuals over the last few decades and see their growth - myself included. Market conditions were undoubtedly a factor as well, especially for established companies.

I would highly recommend looking up Michael Carnes' posts about the Lexicon PCM96 and PCM plugins here at Gearspace. Carnes was very upfront about the surprising results he found, when migrating from custom Lexicon DSP solutions, to the TigerSHARC in the PCM96, to native processing. He may have addressed the marketing issues he faced at Lexicon in his posts - it's been a while since I've read them, so I don't remember all the details.

From my own experience, there was a pretty big leap in the mid 2000s with native processing, compared to DSPs. In our division of Analog Devices, we had been working with native processing before we moved over to DSPs (our division was a small startup called Staccato Systems before it was acquired by ADI), so people felt comfortable programming both. The level of processing we could get out of a SHARC or Blackfin blew away anything we would get out of the Pentium machines of the time. Fast forward to 2007, when I was working on my own, and the results I was getting on the Core2 Duo MBP were far more impressive than what I was getting out of a SHARC just a few years before.

Anyway, this is just my perspective on things. I know that you had plugins released several years before I did, so it is good to hear your perspective! For anyone who doesn't know, Antti wrote some of the foundational papers for virtual analog synthesis, and had several plugins released as part of the Smartelectronix collective in the mid 2000s.

Pentium 4 was a failure and complete dead end. What Core really did was allow bringing the Pentium 3 architecture to near Pentium 4 clock speeds. Other major improvements of course were superior floating point performance, much better SIMD (although Intel really screwed up by not introducing gather reads at that time - those would have been a game changer for SIMD), having multiple cores as the standard (with Core 2) and various miscellaneous improvements. It just happens that for classic reverb algorithms in particular, the floating point performance and branch prediction penalties were not nearly as much of an issue as for many other algorithms.

Indeed. I'm not saying Lexicon or others should have released their flagship reverbs as plugins. Just that the since debunked (by you among other people) claims that "reverbs using custom processors can't be done on regular CPUs" were strictly speaking false even by that time. If you think of it from buyer point of view, "3 Lexicon 480Ls for the price of a normal desktop, 8 ch audio interface and a VST plugin" would have been a bargain compared to having to pay for 3 separate 480Ls. Of course Lexicon's existing customers might have been a bit annoyed if they'd just paid $12k for a 960L only for Lexicon to introduce (near) similar quality plugin for a fraction of that price. Likewise, trying to charge $1000 or more for a plugin would have been a flashing invitation for warez groups at the time.

TBH, that 64+ samples was just a very rough off the cuff estimate. The idea is simply to avoid too many reads / writes to separate parts of the delay ram within the same for loop so as to not completely kill the cache. I haven't reverse engineered any of the 480L algorithms, but unless they radically differ from the allpass & delay building blocks presented in Dattorro's "plate" reverb and 224 concert hall model, the only real limit is that any feedback loops shorter than 2x the block size need to be handled in one for loop (and having many output taps taken from within such loop might be painful). I've gotten the impression that by 480L Lexicon no longer used multiple recursive allpass filters so that would have actually made it easier. Not that it would have been very efficient compared to some other reverb structures, but just about efficient enough to be workable.

I saw this on the synthesis side in the mid 00s when I presented my work on circuit based modeling of analog filters. At that time more or less everyone else (barring perhaps UAD and one or two other devs) was winging it without any real understanding of how things work and what sounds good, which was rather ironic given that I was still (technically) a university student back then.

I can still see that same problem today with synth manufacturers where the specs are seemingly good, but the musical tuning of the circuitry / model is lacking. Conversely on the plugin side everyone seems to be trading large amounts of CPU for little incremental improvement for musical uses. It's not entirely different from how Lexicon and a few others were getting very musical results out of surprisingly little processing power. After enough time working on those things you begin to develop an intuition about what's important vs not and what works vs what doesn't.

Those threads were / are a gold mine, including your own contributions! I myself was able to reverse engineer the delayline lengths and allpass filter coefficients of the 224 concert hall algo from a few IR recordings based on those threads sometime around 2009 - 2010.

Ps. Any chance you might revisit some more 224(X) algorithms in the future? I managed to use VVV Dirty Hall to get results remarkably close to the grainy and awesomely faux-lofi reverb on Alphaville’s debut album. I wouldn’t say no to Rich Plate or others… (”Shut up and take my money! Again.”)

Space Designer was released as a separate purchase to Logic users - it cost about £450 originally, in Logic 6.3, Mac/PPC only.

I moved to Mac from a Pentium4/Logic 5.x system to Logic 6.4 Pro in about 2005, when all the previously extra plugins in Logic were all bundled in, and I bought a top of the line Mac Powerbook (G4) for £2.2K.

That machine was less powerful than the Pentium machine of course, but it was still a top of the line Mac laptop, and I'd say Space Designer would easily eat 30% or more of that. I remember that I could only run a handful of voices of Minimonsta on that machine, for a benchmark - 8-voice unison mode would kill the CPU.

Desktop G4 machines were a bit more powerful, and the G5's would come along shortly on the desktop to increase power again.

But that was a fairly-expensive commercial reverb plugin released in 2003, costing around low/mid-price reverb rack hardware, that was certainly in that range of heaviness on still fairly expensive machines. SO some marketing people were deciding there was commercial mileage in that product. Just as an example of what was around in the reverb plugin space at the time - at least one that springs to mind, there are others I'm sure.

I agree with you that good reverb in plugin form was still regarded as a heavy CPU task, and probably best outboarded to good hardware in full mixes unless you were running a fairly turbo-charged system, but it was definitely in that transition process. The really good algorithms weren't yet available in plugin form, but convolution was certainly making good reverbs happen in the box, until the algorithmic stuff would catch up in quality terms.

5+ years later, the landscape had changed a lot, Macs had switched to Intel Core2Duo's, and had, as I previously stated, probably have been capable of running those algorithms for some time - like Antti re-iterated, probably the bigger factor were commercial considerations, not whether those reverbs were viable in pure CPU terms...

Great thread - thanks for the significant contributions of people with the actual practical implementation knowledge of this stuff, which is somewhat beyond me...

Wow, I had no idea Space Designer was that old... and I definitely had no idea how expensive it was when it was released! I've honestly never liked it too much, and like many, Valhalla has completely satisfied my reverb needs for the decade, roughly. Although I do like Logic's newer ChromaVerb, that plug has some very cool capabilities and flexibility. The UI is reasonably intuitive, too.

Thanks indeed to Sean and Antti for contributing such significant knowledge and history to this thread! Developments in digital audio technology over the past thirty years are fascinating to me, often more than the analog circuitry audiophiles worship.

This is so cool! It goes a long way toward explaining why I love the Lexicon sound. As a pure rank amateur who wouldn't know an all-pass from a hall-pass, I think we are so lucky to have some heavy hitters weighing in on this topic. Reverb is an obsession of mine, and this thread is gold.

I remember when the Lexicon verbs came out on VST they priced it quite high, somewhere around $1800 or so, which was a fortune for a plugin.

I also remember some guy who paid that much getting generally pissy about it when the price dropped like a rock a few years later. Obviously there was more competition in the market and the price needed to reflect that, but boy was he angry he paid so much.

Any opinions on the Eventide sp2016 Reverb from people who have experience on it ?

Lexicon hardware itself was nothing special by the early to mid 90s. Lexicon firmware and the knowledge that ended up in it OTOH very much was so (and arguably still is, given how well their classic reverbs hold up)

I'd also like to thank Sean for the constructive disagreement which is by no means given on forums such as this (as I was just reminded two days ago elsewhere). Whether it really is "blue-gray" or "gray-blue", everyone benefits from the shared knowledge and perspective.

Understandable, for sure. I didn't pay that for it, but I paid probably $700 or $800 something for the suite. I look at it as "value added" for those 3-5 years (don't remember exactly) I had use of the suite versus if I hadn't, especially relative to those who didn't! For those of us who couldn't afford the hardware (well, I had Lex hardware, just not the high-end stuff) it was a god-aux-send.