96 khz sample rate i/o for better VSTi performance?

[zerocrossing]

So, hot off the heels of the Diva VS OB-8 thread, I have a question. In my experiments I can clearly hear some aliasing in high notes. I imagine it's also influencing even mid-range notes that have a lot of harmonic content. It's clear on most software synths, but I also heard it quite clearly on my old Nord Lead 2x and Virus TI Snow. It's more subtle but it's there on the KingKORG as well.

Someone mentioned running an audio rate of 96 ghz to help with such things, saying it made a huge difference. The thing is my Presonus 16.0.2 only does 48 khz which is what I've been running. My computer is a Dell i7 quad core running at 2.5 ghz.

The thing is, I need real time low latency performance. My workflow is to create audio loops in real time with synths and guitars. They're often synced to drum tracks coming from Live's sequencer or Maschine. Sometimes I use other software though. Tell me about running at high sample rates. If I get something that does 96 ghz i/o at 24 bit, will I notice a big audio difference when using software? Will this extra processing needed tax my CPU to a point where low latency will be impossible? If up against it, I'll always sacrifice some audio quality for my work flow. Sorry about the n00bish question... I've just always ran things at 48 khz, but if I can move more of my synths ITB, it'll be better for me. I'm going to cross post this to get a different perspective, so don't get put off.

[EvilDragon]

You mean kHz, not gHz with those sample rates

Well here's one thing - if you run your project at a higher sample rate, as a bonus you get a lower latency. Buffer size 128 at 44.1k is roughly the same as buffer size 256 at 88.2k. Something to have in mind.


As for plugins, it depends. Some plugins don't take this into account (they bandlimit stuff). Some do. Zebra for example can benefit from higher SR. This is best tested with a scope. If a plugin has stuff above 20k at higher project SR, it takes it into account. Quite obviously.

[zerocrossing]

You mean kHz, not gHz with those sample rates

Well here's one thing - if you run your project at a higher sample rate, as a bonus you get a lower latency. Buffer size 128 at 44.1k is roughly the same as buffer size 256 at 88.2k. Something to have in mind.


As for plugins, it depends. Some plugins don't take this into account (they bandlimit stuff). Some do. Zebra for example can benefit from higher SR. This is best tested with a scope. If a plugin has stuff above 20k at higher project SR, it takes it into account. Quite obviously.

Does the CPU hit double though, as it's being asked to creat double the information, or are we over sampling a lower bit rate signal?

[3ee]

Yes, there is a real benefit of running software @ higher SR!

Yes, the CPU hit seems proportional to the SR increase... A workaround is to work at lower SR and render @ higher SR.

Even if you convert to say: 44.1k for the final format, the benefit of rendering mixdown @ higher SR prior to that still remains... (you need to have at least 2 tracks/voices of audio playing at once though)

This is where usually many people just don't "get it" ... they say humans can't hear above 20k ... they doesn't understand that digital audio doesn't work like that...

The phase relations between multiple tracks/voices are more accurate and the "micro" inter-modulations are better preserved and more detailed resulting in a increase in clarity... ( OTOH, rendering a saw tooth on different SR and A/B-ing them won't get you anywhere... )

So, better phase relations and modulation is one benefit.... less aliasing and more accurate modulation like envelopes or LFO shapes with higher speeds etc is another, what else?

Don't forget to check out effects like flangers/choruses on higher SRs ... they're simply lovely sounding.

[UltimateOutsider]

Well here's one thing - if you run your project at a higher sample rate, as a bonus you get a lower latency. Buffer size 128 at 44.1k is roughly the same as buffer size 256 at 88.2k. Something to have in mind.

This is because the buffer is consumed at the speed of the current sample rate, so higher sample rate with the same buffer size = lower latency. But it ALSO means you're getting less glitch protection from that buffer.

For example, if the smallest sample buffer you're able to use with your computer and interface at 44.1kHz is 128 samples, if you bumped your sample rate up to 88.2k, you'd HAVE to at least double your sample buffer size to avoid glitching out, and you'd end up with the same latency that you were seeing at 44.1/128.

I personally haven't experimented with increasing sample rates to improve plugin sound. I understand why it can sometimes help. But I also imagine that: A) There are likely diminishing returns at some point and B) The degree of improvement you get at a given sample rate (and whether there's any improvement at all) varies by specific plugin. Am I off-base here?

(Also, are any potential gains lost when you ultimately render out to MP3 anyway?)

[zerocrossing]

Right, I get all that but my guess is it won't work for me as I'd want a handful of instruments and effects running in real time. Judging by my current CPU hit I bet I max out too early. I wish I could borrow a 96 kHz I/o box to test it

[Synthetic Wav]

96khz might lower aliasing buy few dbs not sure it worth it,it will depends the patchs, no idea about this.. but many softsynth upsample way more so 96khz, it won't make much diffference with these, better use well designed synth.
it s like with processors like eqs , comps, saturation... well designed stuff should not sound that much better if not at all at 96khz imo not to the point it worth using 96 khz.

resampling has to be really transparent it can affect a lot the sound in some plugins .

it make me think in the diva ob8 test, what i didn t like in A could be accentued buy a not transparent resampling, i don't mean it s the case, but it could be. it can affect the transients, the highs..

[camsr]

Your CPU is kind of low to use 96khz. 2.5ghz will be put to the test on any large projects at 96khz.
You may be able to use 88.2khz instead. The benefits are almost the same, but the CPU time cost is only reduced marginally.
Synths are best oversampled, most decent synths have an oversampling option. I would not increase sample rate just to get a better sound from a synth, because everything else will double the CPU cost.
Equalizers, reverbs, delays, distortions, etc all sound better at 96khz. The true amount of aliasing reduction is 6dB, but if your sounds are bandlimited at around 16-20khz, you get less audible aliasing than -6dB. Aliasing is like a mirror, and with the distance from the mirror increased, the distance is doubled... there's more room for harmonics to exist between the area before the mirror and after it's reflected, above the bandlimit point. You can still have frequencies all the way to 48khz in the signal, but there's really no need.
One thing you must learn to do when working at 96khz is bandlimit. Use lowpass filters to cut away the ultrasonic treble before effects processing AND after.
The thing about plugins and 96khz is not all plugins support it correctly. Lots of plugins have only been designed to work up to 48khz. So take time to find out which ones work correctly, for example an equalizer should always work at the same frequencies at either sample rate. Delays should always delay the same amount of time at any sample rate. When a plugin does not take sample rate into consideration, uping the sample rate can result in effects like time "speeding up" and frequencies increasing, synths as well.
If you are sample heavy, resampling can cause a more unexpected slowdown than only double the CPU. The memory usage is also doubled in size, and this can result in a greater slowdown if the computer's memory subsystem cannot keep up. You can mitigate this somewhat by using faster, lower quality resampling processes.
You should notice the difference between 48 and 96khz for production immediately while using EQs and distortion effects.

[Synthetic Wav]

One thing you must learn to do when working at 96khz is bandlimit. Use lowpass filters to cut away the ultrasonic treble before effects processing AND after.

i have to try , just for testing , tdr made a filter for this.
someone should make a high quality resampling plugin (at least X 16 ) to host plugins in a daw no idea if something like this exist

[fluffy_little_something]

I also use 96kHz, it does consume clearly more CPU power, but then again, it does sound better and I use mostly Sylenth, which uses little CPU to begin with. I avoid all modern synths that use 0df filters as those seem to be CPU guzzlers, along with reverbs (long release times/tails generally seem to multiply CPU consumption).
But you also mention 24bit. That is something I don't use, 16bit seems to be enough for my needs. Both 96kHz and 24bit would be too much even for Sylenth on my AMD-based computer. Your Intel CPU is probably much more powerful than my AMD, though. Is your current system already struggling or is there still lots of unused capacity when making music?

[otristan]

Plugins do not take into account the 16/24 bit, all DAW run plugins using 32 bits float (or 64bits float in some version of Cakewalk and if the plugin supports it)

my 2 cents

[camsr]

One thing you must learn to do when working at 96khz is bandlimit. Use lowpass filters to cut away the ultrasonic treble before effects processing AND after.

i have to try , just for testing , tdr made a filter for this.
someone should make a high quality resampling plugin (at least X 16 ) to host plugins in a daw no idea if something like this exist

It's this plugin
https://vladgsound.wordpress.com/2014/1 ... a-version/

Although I can attest to results with the FL parametric EQ's lowpass. There is a truth to this "bandlimiting everywhere" thing, to different degrees I imagine.

[lfm]

After some tests I now run all synths/samplers in Metaplugin with 2xoversampling - and I don't need to run the full project in 96k. Works in realtime as well, you don't have to render to hear it.

I think improvement depends on material, especially on wavetable synths where every note is a resampled pitch. Start making chords and stuff and wow effect is there using oversampling, much less harmonic beating on harmonics as tempered scale will do, being not perfect thirds etc.

If you hear improvement clearly on a single synth track, imagine what it does to a full mix with a handful or dozen such instruments. All this adds up and is why all professional studios bought $2000 preamps and mikes for that little extra on each recorded part - so why not do it on VSTi's as well.

But Sonar added in june release automatic oversampling on freezing/rendering if you mark your synths with using that option. Really cool, and you don't even have to think about it once activating the option - freeze your synths and you mix with this improvement as well.

But http://ddmf.eu Metaplugin is a nobrainer investment either way.

[JJ_Jettflow]

You mean kHz, not gHz with those sample rates

As for plugins, it depends. Some plugins don't take this into account (they bandlimit stuff). Some do. Zebra for example can benefit from higher SR. This is best tested with a scope. If a plugin has stuff above 20k at higher project SR, it takes it into account. Quite obviously.

Does the project have to be in a higher sample rate or can stemming out the VSTs at a higher sample rate provide the same result?

[mridlen]

First watch https://www.xiph.org/video/vid2.shtml

96k will increase the upper frequencies that the audio can inhabit, which reduces the aliasing (aliasing is just distortion caused by frequencies higher than nyquist). Once you convert down to 44k PROPERLY in a way that does not alias (just basically removes frequencies above the nyquist) the aliasing will not be there. So you could basically record in 96k, convert back to 44k and put it back in your project as audio and the aliasing will be gone.