Surf.Whammy's Technical Information on VST 2.4, VST 3, and Double-Precision Arithmetic relative to FSP 8

[Surf.Whammy]

TECHNICAL INFORMATION AND THOUGHTS ON VST 2.4, VST 3, AND DOUBLE-PRECISION ARITHMETIC FOR FSP 8

For reference, Reason 13 (Reason Studios) is 64-bit only and does not support 32-bit VST plug-ins; so if you have Reason 13, then it has scanned all your plug-ins and will scan any new plug-ins you add to your system.

I am testing Fender Studio One Pro 8 with 64-bit "Process Precision" (double-precision floating point arithmetic) for the "mix engine" after switching from single-precision to double-precision; and after adding Tonalic Essential (Celemony), both applications (Reason 13 and Fender Studio Pro 8) rescanned the plug-ins and reported no plug-ins were blocked.

Whether it did the plug-in rescan with focus on 32-bit or 64-bit is another matter, for which at present I have no information, although intuitively since Fender Studio Pro 8 is a 64-bit only application in macOS and Windows, it probably does the same plug-in scan. In other words, it appears there are not separate plug-in scans for 32-bit vs. 64-bit.

For reference, I suggest this because no matter how I set Fender Studio Pro 8 for "Process Precision", it does not flag any of my VST effects plug-ins. Additionally, it appears logical that there would not be separate and independent plug-in scanning algorithms depending on the setting of mix-engine "Process Precision".

Google AI reports "Fender Studio Pro 8 only supports 64-bit VST3 and AU plug-ins", but I am not certain what that suggests when Fender Studio Pro 8 has the option for single-precision or double-precision mix-engine "Process Precision".

This is what the Fender Studio Pro 8 "Reference Manual" says, "Fender Studio Pro features a cutting-edge high-precision mix engine. A mix engine is the “number cruncher” that does the mathematical summing required to mix multiple sources of digital audio. Fender Studio Pro employs a floating point, mixed-mode engine."

Google AI reports "Fender Studio Pro 8 is a 64-bit application."

From all this, I infer that running Fender Studio Pro 8 with 32-bit "Process Precision" requires doing what one might call algorithmic "downsizing" from 64-bit "Process Precision", whatever that entails.

The information on VST from Steinberg is that VST 2.4 is 32-bit and some VST 2.4 plug-ins are 64-bit; but VST 3 is 64-bit only and does not support VST 2.4. Steinberg moved to VST 3 in 2018, when it dropped VST 2.4 support.

Whether 32-bit VST is supported depends on the Digital Audio Workstation (DAW) application, although in macOS on M1+ machines, VST 2.4 is not supported, hence plug-ins need to be VST 3.

Windows has a separate directory where VST 2.4 plug-ins are stored, as does macOS.

Windows (Intel, AMD) supports VST 2.4 and VST 3, as do Intel-based Apple machines; but Apple M1+ machines support only VST 3.

SAMPLE RATE AND RECORD FORMAT CONSIDERATIONS

Two more things to consider the (a) sample-rate and (b) "record format" (a.k.a., "sample-rate resolution).

Based on the fact that both (a) standard CD (44.1-kHz) and (b) standard Audio for Video (48-kHz) are industry standards and are completely sufficient in every respect for working with audio that humans are able to hear, this tends to suggest that more emphasis on Digital Audio Workstation (DAW) applications and VST.2.4 and VST 3 plug-ins is provided or given to 44.1-kHz and 48-kHz sample-rates, since higher sample-rates mostly are marketing gimmicks and, according to acoustic physics and material reality, do absolutely nothing other than make technically confused people "happy", where the defining rule colloquially is called "Nyquist" and was discovered and defined at Bell Labs by Harry Nyquist in 1928, nearly 100 years ago.

What Fender Studio Pro 8 calls the "record format" also has an affect on the way everything works, although at the moment I have no concise idea about it, other than (a) 16-bits and (b) 24-bits appear in various bit-rate standards and refer to the number of bits transferred per unit time (usually seconds for audio), where for example the Red Book Standard CD format is 16-bit depth, which according to Google AI "Defines the resolution and dynamic range (difference between the softest and loudest sounds), providing a high-fidelity, 96 dB range".

The standard for audio in Video is 16-bits but 24-bits is preferred, and according to Googlel AI "provides a 144 dB dynamic range for high-fidelity, low-noise recordings".

Similarly, Google AI provided that "64-bit floating-point audio (64-bit float) provides a dynamic range of over 1,500 dB".

This tends to suggest that one of the techniques major recording labels and mastering engineers use to produce the best possible audio perhaps maps to using 64-bit floating-point audio (64-bit float), which is something I am going to explore with a few experiments--observing that I have been able to create "ballpark" audio with good frequency response compared for example to the remasters version of "Dazed and Confused" (Led Zeppelin), and the current official version of "Lux Æterna" (Metallica).

[NOTE: I included the recent experiment with what I call "Quantum Sonic Entanglement (QSE) Sparkles", and all three of these YouTube music videos are best enjoyed with listening with studio-quality headphones like SONY MDR-7506 headphones, where in particular I am very happy with the deep bass and sustained tail of my new Kick Drum sound. ]

QUESTIONS

(1) Are the reported problems with VST plug-ins a matter solely of outdated versions?

(2) Does this happen on Apple M1+ machines?

(3) Was Fender Studio One Pro 8 tested primarily in macOS on M1+ machines and not on Apple Intel-based machines?

(4) Is anyone having the "plug-in" problem on the Mac?

(5) Is it even possible that extensive testing of Fender Studio Pro 8 was done at the higher, non-standard sampling rates?

(6) Are the reported problems with VST plug-ins occurring on machine set to use higher, non-standard sampling rates?

(7) What "Record Format" are folks using?

(8) Considering that Fender Studioi Pro 8 is a 64-bit application in macOS and Windows, does it make sense to do everything in 64-bit floating point (double-precision) exclusive with VST 3 plug-ins and corresponding Fender native format plug-ins (when they are 64-bit), as well as Reason Rack Plug-in VST, which of course by design is VST 3 only (and also AU on the Mac)?

[NOTE: I did an experiment where I changed everything to 64-bit floating point (double-precision), and the volume level of my recent song on "sparkles" nearly doubled when played back in Fender Studio Pro 8, which solves a long-standing puzzle here in the sound isolation studio.

For reference, the current YouTube music videos were not changed; so this only occurred when I played the song in Fender Studio Pro 8 and suggests another few experiments to determine how it affects YouTube audio, which for reference is processed by YouTube with its own internal audio processing technology, hence (a) you mix and master using your technology and then (b) YouTube masters your master (uploaded audio). ]

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies

 

[MisterE]

If only you were getting paid by the word

 

[reginaldStjohn]

My understanding is that 32 bit and 64 bit VSTs are how they are compiled, support 64 bit instructions, not that they do 64 bit audio operations.

 

[Vocalpoint]

"Precision" is simply a setting in Audio Setup for the S1/SP8 mix engine to use single (32 bit) or double precision (64 bit) floating point math.

It has nothing to do with the OS, how plugin code is complied or the "flavour" of a plugin either (VST2/VST3/AU etc).

Is there a difference/advantage? Unless you have super human capabilities to hear beyond the human spectrum - there are none. But if you are groovin in the realm about how floating point precision works (AKA truncation/rounding etc) - then 64 bit can be a bit more "accurate."

Have been rocking with Studio One since 2011 - not a single client of mine has ever mentioned that my stuff sounds "off" because I am using 32bit precision.

Doubtful that anyone around here can hear that "accuracy" but there it is.

VP

 

[SwitchBack]

Regarding architecture vs. process precision: A computer/OS with 64-bit architecture operates a 64-bit wide data bus, so a 64 bit value can be handled (moved, scaled, added/subtracted) in one step. In comparison 32-bit architecture has to handle that 64 bit value in two halves, taking two steps or more depending on the operation. Multiplying two values (an operation performed a horrendous number of times in DAW software) produces an outcome value double the length (in bits) of the input values. So a 64-bit app can multiply two 32-bit values and handle the outcome in one step, but handling the outcome of the product of two 64-bit values takes at least two steps. So there is a processing steps penalty for running at double precision even in 64-bit architecture. There is no penalty (steps-wise) for running 32-bit values through 64-bit architecture.

As for process precision: Nyquist states that the sample rate must be at least twice the highest frequency in the signal. But for correct inter-/extrapolation the theorem also assumes perfect sample precision. By sampling at ‘only’ twice the highest frequency the highest audible frequencies can have less than 3 samples per full cycle, so those samples better be good. Slightly off and you get incorrect extrapolations, every step of the way in your DAW.

A comparison: Think of foul shots in basketball. There’s more than one successful trajectory from hands to hoop. So you need a third reference point to exactly replicate the previous curve. If that reference point is even a fraction off you may still score but it won’t be the same trajectory. And what if all 3 reference points are a fraction off? So the effect of precision on the mix result can be more than just the smallest step (in dB) of the selected processing precision.

Having said that: Single precision is still very good

 

[Vocalpoint]

(1) VST 3 is 64-bits only.

(2) Windows 10 and Windows 11 generally support both VST 2.4 and VST 3.

(3) Some but not all VST 2.4 plug-ins are 64-bit.

(4) Intel-based Apple computers generally support VST 2.4 and VST 3, at least in the sense of allowing them; but whether a plug-in is supported by Fender Studio Pro 8 is another matter. Most are supported, but not all are supported.

(5) Apple M1+ computers do not support VST 2.4.

(6) Reason 13 (Reason Studios) and Fender Studio Pro 8 are 64-bit applications.

(7) Reason 13 only supports VST 3.

(8) VST 3 is not backward-compatible with VST 2.4.

(8) Fender Studio Pro 8 can run some VST 2.4 plug-ins that are 32-bit but not all VST 2.4 plug-ins, which also can be the case for VST 3 plug-ins, at least if they are not compatible or have errors.

(9) Fender Studio Pro 8 is a 64-bit application in Windows 10, Windows 11, and macOS (at least Sonoma through Tahoe [current]).

I suggest that when a 64-bit application is told to run its "mix engine" as 32-bits (single-precision floating point), this requires using single-precision arithmetic, which is not the same as 64-bit (double-precision) arithmetic.

Suggesting that nothing matters is like suggesting a 4-cylinder engine is the same as an 8-cylinder engine or that there is no difference in a diesel engine and a gasoline engine.

It's easy to write and to say, but writing it or saying it does not make it factual.

Surf

Respectfully - still not sure what your long writeup is intended to accomplish. And - writing it does not make it factual.

If it is facts we need - here they are:

1. Some plugs are acting weird in SP8 for some users.
2. As suggested by Lukas - switching from 64bit precision to 32bit precision makes things calm down.
3. Action 1: If there is an issue here with SP8 - let's have Fender DEV tend to it.
4. Action 2: If a plugin is poorly coded and is now being exposed by SP8's increased VST SDK scrutiny - that is on the vendor to fix.

That is really it.

VP

 

[MisterE]

Things have taken a decisively academic turn. I think I took a wrong turn. I was just looking for a forum to discuss The Stooges' "I Just Wanna Be Your Dog."

 

[Vocalpoint]

Me too.

 

[SwitchBack]

Things have taken a decisively academic turn. I think I took a wrong turn. I was just looking for a forum to discuss The Stooges' "I Just Wanna Be Your Dog."

Academic? One of the propositions was that a 64-bit system is happier running at double precision than at single precision. That's a very practical thought. My suggestion was that double precision is taxing your system more than single precision. So when experiencing problems it may indeed be a good idea to switch to single precision. You can do with that what you want.

 

[MisterE]

Academic? One of the propositions was that a 64-bit system is happier running at double precision than at single precision. That's a very practical thought. My suggestion was that double precision is taxing your system more than single precision. So when experiencing problems it may indeed be a good idea to switch to single precision. You can do with that what you want.

Don't mind me. Academic all you want!

 

[SwitchBack]

@Surf.Whammy On precision vs. dynamic range: With fixed point formats (16-bit, 24-bit) precision and dynamic range are two sides of the same coin. But with floating point formats (32-bit FP/single, 64-bit FP/double precision) some of the precision is traded in for more dynamic range (the exponent bits).

The biggest advantage of floating point isn’t the extended dynamic range but it's the floating of the point, always making optimal use of the precision available. With fixed point formats the actual precision of a signal diminishes with the signal level: A 16-bit -18dBfs signal uses only 13 of the 16 bits available, which makes the precision of that signal 13 bits too (assuming zero noise floor). A floating point format would simply float the point and maintain (near) maximum precision all the time.

Conclusion: Fixed point formats are perfectly fine for final results (mastered to best use of bits available) and a necessary evil for recording (AD converters only do fixed point, so mind your levels!). And floating point is the perfect format for everything in between, so you don’t have to worry about diminishing precision during the mixing process.

 

[PTUG]

Sometimes I wonder if someone is using us to teach AI and or is testing us to see if AI can join a forum/community and go un-noticed. So in the end we are the experiment. Just saying, it could be happening.

 

[John Vere]

Only AI could type that much stuff out! So I think you have hit the digital nail on the head!

On topic it’s been very common for people to get the two different 64 bits totally mixed up.

Of note is that a few audio interfaces are now using 32 bit as an option. Zoom is one of those.
Scary part is they just released a new mixer/ ASIO interface that doesn’t have a gain control!! They say because of 32 bit you no longer need to worry about distortion??? Hmmm.

 

[sintil8]

A comparison: Think of foul shots in basketball. There’s more than one successful trajectory from hands to hoop. So you need a third reference point to exactly replicate the previous curve. If that reference point is even a fraction off you may still score but it won’t be the same trajectory. And what if all 3 reference points are a fraction off? So the effect of precision on the mix result can be more than just the smallest step (in dB) of the selected processing precision.

Having said that: Single precision is still very good

Just asking...
I suppose this is where "Dithering" jumps in and takes a hand when reducing the bit depth.

Regards to all

 

[SwitchBack]

Dithering is a trick to lower the noise floor. All errors due to quantisation etc. translate into noise, but this noise is very peaky (spiking in some frequencies and very low in others). Dithering adds a little noise to shake the ground under the sand sculpture so to speak, bringing down the peaks and with that the noise floor. But it does not miraculously always push the reference points (and with that the curves) in the right direction. Half of the time 'wrong is still just as wrong, but with less noise’.

 

[sintil8]

Dithering is a trick to lower the noise floor. All errors due to quantisation etc. translate into noise, but this noise is very peaky (spiking in some frequencies and very low in others). Dithering adds a little noise to shake the ground under the sand sculpture so to speak, bringing down the peaks and with that the noise floor. But it does not miraculously always push the reference points (and with that the curves) in the right direction. Half of the time 'wrong is still just as wrong, but with less noise’.

Thank you, and yes, I like assimilation to the sand sculptures.

Kindest regards

 

[PTUG]

I guess AI bots is going to be the new thing that forums are going to have to contend with.