Simple questions on synth tech basics...

[fluffy_little_something]

The fact that most plug-ins work in similar ways is what lends credibility to the application of benchmarks to compare processor performance.

You can find a number of benchmarks in any review of the processors you're interested in.

But that doesn't really answer my question, in fact it is a bit off-topic
Never mind, BertKoor has already answered it...

[aciddose]

No, it does answer your question. Select the processor with the higher performance across multiple benchmarks.

This will most likely provide higher performance across multiple plug-ins.

If you don't want to get into details, this is where you're stuck, so just deal with it.

You must realize that you're essentially asking "how can I get detailed without having to deal with details I'm uncomfortable with?" I'm certain you understand how contradictory (not to mention impossible to satisfy) such a request is.

[VicDiesel]

*) There was a simular benchmark done here at KVR Audio in the past, I think it was with a convolution reverb. Not sure though, anyone remember that? And what were the conclusions?

After digging through some rusty memory banks it came back to me: Ambience. Use some lush setting and see how many instances you can run.

Funny: I google for it and find a post by you.

viewtopic.php?p=2032594#p2032594

Clearly my memory chips are better than yours.

Victor.

[fluffy_little_something]

No, it does answer your question. Select the processor with the higher performance across multiple benchmarks.

This will most likely provide higher performance across multiple plug-ins.

If you don't want to get into details, this is where you're stuck, so just deal with it.

You must realize that you're essentially asking "how can I get detailed without having to deal with details I'm uncomfortable with?" I'm certain you understand how contradictory (not to mention impossible to satisfy) such a request is.

From what I have read so far, FP is more demanding and at the same time more important in mathematically complex calculations, which includes synth filters etc.
So, no, my question did make sense and I already know the answer, i.e. what to pay attention to when picking one processor over another. If I have to decide, I will go for the one that offers better FP performance rather than integer performance.

[Mayae]

From what I have read so far, FP is more demanding and at the same time more important in mathematically complex calculations, which includes synth filters etc.
So, no, my question did make sense and I already know the answer, i.e. what to pay attention to when picking one processor over another. If I have to decide, I will go for the one that offers better FP performance rather than integer performance.

Nearly everything in DSP boils down to something looking like this:

Code: Select all

y[n] += x[n] * a

This is otherwise known as FMA (fused multiply-add). Usually DSP plugins travels through a memory buffer, multiplies and adds some numbers and stores it back into another buffer. Another core element is delay lines; storing audio history into some large buffer.

A lot of heavy algorithms benefit greatly from enhanced instruction sets that allows to parallelize these operations, I would specifically ensure these instruction sets are supported: FMA3, AVX2 (possibly even AVX512).

Also, while single-core performance is great, multiple cores can give a higher return. If you're running multiple mixer channels, these can potentially run in parallel (and all DAWs out there I believe do multithreaded rendering of mixers). Since you specifically mentioned Diva, it actually is highly parallel and benefits greatly from a large number of cores.

To summarize parameters for an ideal DSP cpu, ordered after my opinion of importance: Cache performance (!!), supported instruction sets, SIMD floating point throughput, number of execution units, number of cores, memory performance, processor speed.

[BertKoor]

*) There was a simular benchmark done here at KVR Audio in the past, I think it was with a convolution reverb. Not sure though, anyone remember that? And what were the conclusions?

After digging through some rusty memory banks it came back to me: Ambience. Use some lush setting and see how many instances you can run.

Funny: I google for it and find a post by you.

viewtopic.php?p=2032594#p2032594

Clearly my memory chips are better than yours.

Victor.

That was it: the Ambience Test. Wow, ten years ago already... Am I excused from forgetting I posted that?

Btw, anyone knows how many instances a humbe i5 or so can run?

[BertKoor]

To steer this a bit back on-topic, personally I think the whole analog - digital thing leads nowhere. These have many simularities and differences as well. Bottom line though: is there such a thing as a typical "analog" sound? And if there is, is it worth persueing it like the Holy Grail?

Analog character is certainly not defined by a certain freq spectrum plot. Engineers try to make opams (the basic building block) as linear as possible. And anything digital will finally have to go through the very same opamps. Both worlds meet after the DAC.

There is however something very distinctive about lo-fi analog transfer media: wobbled cassette tape, AM/FM radio with bad reception, overdriven dinky amps & speakers. The digital world has its counterparts: hard clipping, bitcrushing, 48 kbps mp3....

So make music with whatever you can get your hands on. But analog synths did not sparkle your tracks with mojo and pixie dust. What you hear in older recordings is just talented people using whatever they had at full potential and having put great effort in whatever they did with it. Transpiration pays off...

[fmr]

DELETED

[stratum]

To steer this a bit back on-topic, personally I think the whole analog - digital thing leads nowhere. These have many simularities and differences as well. Bottom line though: is there such a thing as a typical "analog" sound? And if there is, is it worth persueing it like the Holy Grail?

"Analog sound" is defined by its contrast to marketting gimmick. I remember some crappy multi effect guitar processors of the 90's whose brand name is off topic to the discussion and I won't mention it. Their signal path was all-analog actually, but they included digital control circuitry. They were perceived as being "digital-crap" by the time. Not that it's really difficult to make a good sounding effect using either analog or digital hardware, the problem is often elsewhere, like packing a lot of gizmos into a product with a rather low price tag intended to have an appeal to teenagers. Not that there is a problem with that, as long as the price tag is low, the real problem begins when these things are imported into so called "developing countries" where they are no longer cheap for several reasons that are irrelevent to the subject matter (so I won't mention these, either).

[fluffy_little_something]

Analog character is certainly not defined by a certain freq spectrum plot. Engineers try to make opams (the basic building block) as linear as possible. And anything digital will finally have to go through the very same opamps. Both worlds meet after the DAC.

Still, people, including some developers, say their plugins sound authentically analog, unlike others, which are said to sound digital. I don't think that there are analog frequencies as such, but that we associate a certain sound with analog. And of course those typical frequencies of that analog sound can be reproduced in digital ways. As you rightly say, there is no mojo at work, it is all just physics.

And I continue to think that frequencies play a key role in it as well. After all, the specific frequency distribution is what characterizes any sound. That includes hum, hiss, distortion, transients, etc. At the end of the day it all boils down to frequencies and the amplitudes at them.

[stratum]

If you open a signal processing book and read it (will require some (perhaps a lot) patience) you'll see that it has some assumptions which are not meant to be violated. The assumptions are that the system is linear and time invariant (basically that means the theory in the book does not cover music-dsp applications but is still relevant for other reasons). Another problem is that sampling rates that are suitable for faithfully reproducing sound aren't necessarily adequate to process it. Engineers have invented ways around both of these problems. The discussion isn't very meaningful without being aware of the problems that were solved in this way.

[fluffy_little_something]

I think that would go too far here.
Basically, I was indeed referring to making the sound more "analog" by using an equalizer. And to a certain extent that seems to be possible. By analog I mean what people usually associate with that term, i.e. warm, full, solid bass, etc. (no idea whether those associations are correct to begin with, I only had one analog synth, but that was like 30 years ago and I don't remember the fine details of its sound; plus, I had it hooked up to my stereo, whose speakers did not reproduce any bass below 50Hz or so; my headphones back then were mediocre as well, the HiFi-type, so probably anything but neutral).

[stratum]

Old radio speakers that were in those old synths are still in use in guitar amps. They were intended to be full range speakers for their time but their upper frequency range is actually around 5khz and then it starts to roll off. That is probably one source of warmth associated with old analog gear. There are probably others, like a similarly band limited output transformer in a valve amp, harmonics added by the valve circuitry, their gentle overdrive character etc, but I have built a few such amps and withnessed first hand that this gentle overdrive character is not gentle at all without a properly band limited speaker cabinet.

[JCJR]

A couple of wild guesses regarding a good dsp computer, if the use is plugins working inside a DAW/Sequencer or whatever-- Not necessarily ideas that would help in shopping for a computer--

SSD drives show vast performance improvement over hard disks in my experience. Just faster read/write of "big chunks of data".

If a program has the luxury of keeping most or all of the audio data in memory it is easier/simpler to program and can run faster. So lots of memory ought to generally benefit, assuming a program is written to exploit lots of memory. All the audio data gets loaded into memory before playback starts. The speed of SSD to memory determines the "wait time" getting ready to play.

CPU chips seem to run a lot faster than the memory can be accessed. So a computer that can transfer RAM to CPU the fastest ought to do audio and DSP faster, even if the actual math performance once the data is in cache is about the same as some other computer with slower memory. So a fast RAM system ought to be important.

The intermediate cache memory between slow RAM and fast CPU-- Ideally the CPU could access cache "instantly" with no waits, and cache-RAM transfers real fast, and real big cache memory. Maybe the ultimate ideal would be if cache is unnecessary and the computer has giant amounts of RAM that is just as fast as the CPU.

Last time I looked, one of the tricks to fast code was to minimize cache swapping. So if cache could be bigger, or RAM faster, then maybe things would work a lot faster even if the CPU math performance doesn't get a lot faster?

[JCJR]

Old radio speakers that were in those old synths are still in use in guitar amps. They were intended to be full range speakers for their time but their upper frequency range is actually around 5khz and then it starts to roll off. That is probably one source of warmth associated with old analog gear. There are probably others, like a similarly band limited output transformer in a valve amp, harmonics added by the valve circuitry, their gentle overdrive character etc, but I have built a few such amps and withnessed first hand that this gentle overdrive character is not gentle at all without a properly band limited speaker cabinet.

I generally go for "clean as possible" but a lot of folks like the sound of synth run thru a guitar amp with the cheap output transformer and band-limited high-distortion speakers. Though so far as I can tell, "low distortion speakers" are vanishingly rare. So far as I know, speakers below 2 percent distortion over the audio spectrum are quite rare, and perhaps speakers below 1 percent distortion over the audio spectrum are nonexistent. For instance--
http://www.neumann-kh-line.com/neumann- ... term=THD+N

Ideally the lower the harmonic distortion, the cleaner, or more transparent, the loudspeaker will sound. Less than -30 dB (3%) at low frequency and less than -40 dB (1%) at mid-high frequencies is normally considered to be good, lower values than these is of course better.

So if that is a "high fidelity studio monitor" distortion level, then PA speakers or guitar speakers probably lots worse on average.

Some of the more popular guitar speakers have presence peaks in the upper mids before they start rolling off in the highs. The upper-mid boost can be "ear candy" like with celestion greenback or old stiff-cone "classic" speakers. A combination of the high frequency rolloff, presence boost, and several percent of distortion.

Maybe running computer synths thru a guitar amp set for "clean" and re-miking back into the song would make them sound "fatter" or "older" or whatever.

Low-level coupling transformers are another thing that adds distortion while still sounding "fairly clean". Maybe a box with a few coupling transformers could be used in an interface loopback to "warm up" computer synths. I looked into it awhile ago, but can't recall details of which ones to look for.

The "highest quality" coupling transformers are rather clean and wideband, so looping back thru premium high-fidelity transformers probably wouldn't make much audible difference. On the other hand, a 2 dollar telcom 1:1 600 ohm coupling transformer might just sound crappy rather than subtly sweetening the signal. Or maybe it would be nice. Dunno.

Some of the coupling transformers of yesteryear, in expensive studio gear were prized for the sound, but they were not "near ideal" low distortion devices. I think there are new-manufacture transformers also designed to be "hifi" but with warts like that.

So far as I know, a good quality saturating transformer would tend to add a little distortion / saturation to bass frequencies while remaining clean at mids and highs. Also I believe heavy saturation of the bass ought to intermodulate the mids and highs, though the "hifi sweetening" probably is with "just a little bit" of bass saturation and everything else clean.