Latency Tolerance

thegummy

WiresDreamDisasters said:

A faster computer quite often only gives you marginal improvements to round-trip latency. Because the CPU isn't really the bottleneck when it comes to latency.

I don't agree with this statement at all.

The CPU determines how much processing can be done at each buffer size setting without audio breakup. And the buffer size setting is directly linked to how much latency there will be. I've never seen a single audio interface that doesn't give lower latency for a lower buffer size.

There are tons of benchmarks available that shows that the faster the CPU, the more processing can be done at a particular buffer setting.

WiresDreamDisasters

thegummy said:

WiresDreamDisasters said:

A faster computer quite often only gives you marginal improvements to round-trip latency. Because the CPU isn't really the bottleneck when it comes to latency.

I don't agree with this statement at all.

The CPU determines how much processing can be done at each buffer size setting without audio breakup. And the buffer size setting is directly linked to how much latency there will be. I've never seen a single audio interface that doesn't give lower latency for a lower buffer size.

There are tons of benchmarks available that shows that the faster the CPU, the more processing can be done at a particular buffer setting.

It's negligible versus having a better driver implementation. If you've got a piss poor driver, then even having a fast as fuck CPU you're still going to run into CPU spiking issues and audio-dropouts. You can have a fast as fuck CPU, but if your RAM sticks have dodgy timings, then you're not going to get good performance at low latencies, period.

A faster CPU does change your latency, but it's minimal, given all other factors. When I upgraded from my old quad core Q6600 processor to an i7, I saw less than a 2ms difference. It wasn't huge, and if I had bought the CPU solely for latency improvements, I would've been pissed off.

The CPU isn't the bottleneck, and it isn't the only factor to pay attention to. The reason I bring this up is because a lot of people think they can just plunge a load of money into a CPU improvement and see radical changes.

You can't.

Being more performant at similar latencies; ie - upgrading the CPU and seeing 40% CPU usage at a buffer size of 64 with a RTL of 8ms, instead of the previous 60% with the previous CPU - is not the same as *improving* latency... which is my point.

I'm not arguing against the fact that more processing can be done at a particular buffer setting. But that will just lower CPU usage. It wont drastically bring down your latency in milliseconds.

thegummy

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

A faster computer quite often only gives you marginal improvements to round-trip latency. Because the CPU isn't really the bottleneck when it comes to latency.

I don't agree with this statement at all.

The CPU determines how much processing can be done at each buffer size setting without audio breakup. And the buffer size setting is directly linked to how much latency there will be. I've never seen a single audio interface that doesn't give lower latency for a lower buffer size.

There are tons of benchmarks available that shows that the faster the CPU, the more processing can be done at a particular buffer setting.

It's negligible versus having a better driver implementation. If you've got a piss poor driver, then even having a fast as fuck CPU you're still going to run into CPU spiking issues and audio-dropouts. You can have a fast as fuck CPU, but if your RAM sticks have dodgy timings, then you're not going to get good performance at low latencies, period.

A faster CPU does change your latency, but it's minimal, given all other factors. When I upgraded from my old quad core Q6600 processor to an i7, I saw less than a 2ms difference. It wasn't huge, and if I had bought the CPU solely for latency improvements, I would've been pissed off.

The CPU isn't the bottleneck, and it isn't the only factor to pay attention to. The reason I bring this up is because a lot of people think they can just plunge a load of money into a CPU improvement and see radical changes.

You can't.

Being more performant at similar latencies; ie - upgrading the CPU and seeing 40% CPU usage at a buffer size of 64 with a RTL of 8ms, instead of the previous 60% with the previous CPU - is not the same as *improving* latency... which is my point.

I'm not arguing against the fact that more processing can be done at a particular buffer setting. But that will just lower CPU usage. It wont drastically bring down your latency in milliseconds.

Hold on, are you saying that moving from one processor to another changed the latency without changing the buffer size setting?

Or that the buffer size setting you could then move to with the faster CPU only have a 2ms improvement?

What interface and what settings is this?

WiresDreamDisasters

thegummy said:

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

A faster computer quite often only gives you marginal improvements to round-trip latency. Because the CPU isn't really the bottleneck when it comes to latency.

I don't agree with this statement at all.

The CPU determines how much processing can be done at each buffer size setting without audio breakup. And the buffer size setting is directly linked to how much latency there will be. I've never seen a single audio interface that doesn't give lower latency for a lower buffer size.

There are tons of benchmarks available that shows that the faster the CPU, the more processing can be done at a particular buffer setting.

It's negligible versus having a better driver implementation. If you've got a piss poor driver, then even having a fast as fuck CPU you're still going to run into CPU spiking issues and audio-dropouts. You can have a fast as fuck CPU, but if your RAM sticks have dodgy timings, then you're not going to get good performance at low latencies, period.

A faster CPU does change your latency, but it's minimal, given all other factors. When I upgraded from my old quad core Q6600 processor to an i7, I saw less than a 2ms difference. It wasn't huge, and if I had bought the CPU solely for latency improvements, I would've been pissed off.

The CPU isn't the bottleneck, and it isn't the only factor to pay attention to. The reason I bring this up is because a lot of people think they can just plunge a load of money into a CPU improvement and see radical changes.

You can't.

Being more performant at similar latencies; ie - upgrading the CPU and seeing 40% CPU usage at a buffer size of 64 with a RTL of 8ms, instead of the previous 60% with the previous CPU - is not the same as *improving* latency... which is my point.

I'm not arguing against the fact that more processing can be done at a particular buffer setting. But that will just lower CPU usage. It wont drastically bring down your latency in milliseconds.

Hold on, are you saying that moving from one processor to another changed the latency without changing the buffer size setting?

Or that the buffer size setting you could then move to with the faster CPU only have a 2ms improvement?

What interface and what settings is this?

Changing to a faster processor can potentially decrease round-trip latencies at the SAME BUFFER SIZE. But it's marginal. Nothing to write home about.

Yes, a faster processor will mean you can lower your buffer size. But then you're not comparing like for like. You can't say that on machine X with a buffer size of 128 you get 8ms of latency, but on the brand spanking new machine Z with a buffer size of 32 you get 2ms of latency, so OMG this cpu is amazing!!!

.. doesn't work that way. You have to compare like for like. At least when talking about latency improvements. CPU usage is a different thing.

The difference in latencies between machines at the same buffer size and sample-rate is down to a whole manner of things. Different thru-put rates in the PCI or USB bus, the bridge controllers on the motherboard, all sorts of things.

As I said - the CPU isn't really the bottleneck when it comes to latency.

thegummy

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

A faster computer quite often only gives you marginal improvements to round-trip latency. Because the CPU isn't really the bottleneck when it comes to latency.

I don't agree with this statement at all.

The CPU determines how much processing can be done at each buffer size setting without audio breakup. And the buffer size setting is directly linked to how much latency there will be. I've never seen a single audio interface that doesn't give lower latency for a lower buffer size.

There are tons of benchmarks available that shows that the faster the CPU, the more processing can be done at a particular buffer setting.

It's negligible versus having a better driver implementation. If you've got a piss poor driver, then even having a fast as fuck CPU you're still going to run into CPU spiking issues and audio-dropouts. You can have a fast as fuck CPU, but if your RAM sticks have dodgy timings, then you're not going to get good performance at low latencies, period.

A faster CPU does change your latency, but it's minimal, given all other factors. When I upgraded from my old quad core Q6600 processor to an i7, I saw less than a 2ms difference. It wasn't huge, and if I had bought the CPU solely for latency improvements, I would've been pissed off.

The CPU isn't the bottleneck, and it isn't the only factor to pay attention to. The reason I bring this up is because a lot of people think they can just plunge a load of money into a CPU improvement and see radical changes.

You can't.

Being more performant at similar latencies; ie - upgrading the CPU and seeing 40% CPU usage at a buffer size of 64 with a RTL of 8ms, instead of the previous 60% with the previous CPU - is not the same as *improving* latency... which is my point.

I'm not arguing against the fact that more processing can be done at a particular buffer setting. But that will just lower CPU usage. It wont drastically bring down your latency in milliseconds.

Hold on, are you saying that moving from one processor to another changed the latency without changing the buffer size setting?

Or that the buffer size setting you could then move to with the faster CPU only have a 2ms improvement?

What interface and what settings is this?

Changing to a faster processor can potentially decrease round-trip latencies at the SAME BUFFER SIZE. But it's marginal. Nothing to write home about.

Yes, a faster processor will mean you can lower your buffer size. But then you're not comparing like for like. You can't say that on machine X with a buffer size of 128 you get 8ms of latency, but on the brand spanking new machine Z with a buffer size of 32 you get 2ms of latency, so OMG this cpu is amazing!!!

.. doesn't work that way. You have to compare like for like.

The difference in latencies between machines at the same buffer size and sample-rate is down to a whole manner of things. Different thru-put rates in the PCI or USB bus, the bridge controllers on the motherboard, all sorts of things.

As I said - the CPU isn't really the bottleneck when it comes to latency.

Wow that's a very strange way to look at it, saying that the practical reality of it doesn't matter...

The reason someone gets unacceptable latency is that if they reduce the buffer size their PC can't keep up so the audio drops out.

So upgrading to a better PC allows them to lower the buffer size to get the latency they want without the audio dropping out.

Cirrus

@thegummy the CPU is only a latency bottleneck once you're trying to do DSP work in the amount of time the CPU has to work before that buffer needs to be sent to the output.

Drew's right that it doesn't make much difference what CPU it is if you're operating it well within its capabilities - as most are in a tracking situation unless you're piling on the plugins. If you're running your system within its capabilities, you could put a processor 10x faster in the machine and it's not going to significantly change your lowest possibly latency before dropouts start appearing - as Drew's said, the bottleneck is in the bus speed and driver implementation.

Throw on 50 instances of a modelling plugin and the CPU starts to matter.

thegummy

Cirrus said:

@thegummy the CPU is only a latency bottleneck once you're trying to do DSP work in the amount of time the CPU has to work before that buffer needs to be sent to the output.

Drew's right that it doesn't make much difference what CPU it is if you're operating it well within its capabilities - as most are in a tracking situation unless you're piling on the plugins. If you're running your system within its capabilities, you could put a processor 10x faster in the machine and it's not going to significantly change your lowest possibly latency before dropouts start appearing - as Drew's said, the bottleneck is in the bus speed and driver implementation.

Throw on 50 instances of a modelling plugin and the CPU starts to matter.

If you're talking about the audio interface's buffer already being at its lowest setting then there's no point in changing the CPU because latency isn't a problem.

It's only if you have a problem with latency that changing to a faster CPU fixes that because it allows you to lower the buffer of the interface.

I've never seen an interface that doesn't go lower than 6ms on its lowest buffer setting. If you know of any I'd be interested to know but if that was the case then the advice would just be to get rid of it and buy any of the interfaces currently on the market which definitely go below 6ms at their lowest setting.

WiresDreamDisasters

thegummy said:

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

A faster computer quite often only gives you marginal improvements to round-trip latency. Because the CPU isn't really the bottleneck when it comes to latency.

I don't agree with this statement at all.

The CPU determines how much processing can be done at each buffer size setting without audio breakup. And the buffer size setting is directly linked to how much latency there will be. I've never seen a single audio interface that doesn't give lower latency for a lower buffer size.

There are tons of benchmarks available that shows that the faster the CPU, the more processing can be done at a particular buffer setting.

It's negligible versus having a better driver implementation. If you've got a piss poor driver, then even having a fast as fuck CPU you're still going to run into CPU spiking issues and audio-dropouts. You can have a fast as fuck CPU, but if your RAM sticks have dodgy timings, then you're not going to get good performance at low latencies, period.

A faster CPU does change your latency, but it's minimal, given all other factors. When I upgraded from my old quad core Q6600 processor to an i7, I saw less than a 2ms difference. It wasn't huge, and if I had bought the CPU solely for latency improvements, I would've been pissed off.

The CPU isn't the bottleneck, and it isn't the only factor to pay attention to. The reason I bring this up is because a lot of people think they can just plunge a load of money into a CPU improvement and see radical changes.

You can't.

Being more performant at similar latencies; ie - upgrading the CPU and seeing 40% CPU usage at a buffer size of 64 with a RTL of 8ms, instead of the previous 60% with the previous CPU - is not the same as *improving* latency... which is my point.

I'm not arguing against the fact that more processing can be done at a particular buffer setting. But that will just lower CPU usage. It wont drastically bring down your latency in milliseconds.

Hold on, are you saying that moving from one processor to another changed the latency without changing the buffer size setting?

Or that the buffer size setting you could then move to with the faster CPU only have a 2ms improvement?

What interface and what settings is this?

Changing to a faster processor can potentially decrease round-trip latencies at the SAME BUFFER SIZE. But it's marginal. Nothing to write home about.

Yes, a faster processor will mean you can lower your buffer size. But then you're not comparing like for like. You can't say that on machine X with a buffer size of 128 you get 8ms of latency, but on the brand spanking new machine Z with a buffer size of 32 you get 2ms of latency, so OMG this cpu is amazing!!!

.. doesn't work that way. You have to compare like for like.

The difference in latencies between machines at the same buffer size and sample-rate is down to a whole manner of things. Different thru-put rates in the PCI or USB bus, the bridge controllers on the motherboard, all sorts of things.

As I said - the CPU isn't really the bottleneck when it comes to latency.

Wow that's a very strange way to look at it, saying that the practical reality of it doesn't matter...

The reason someone gets unacceptable latency is that if they reduce the buffer size their PC can't keep up so the audio drops out.

So upgrading to a better PC allows them to lower the buffer size to get the latency they want without the audio dropping out.

Believe it or not, some interfaces won't give you particularly good latency figures even at lower buffer sizes. I remember the old 1st generation Digidesign Mbox units were pretty terrible for this. Even at low buffer sizes you'd see RTL's of 10+ ms. The computer would be handling the audio perfectly fine, but the driver just couldn't deliver low latencies.

If you're talking about comparing latencies across different machines, then you have to use the same metrics to get a scientific comparison. Comparing 128 on one machine versus 32 on another machine doesn't tell you anything about round trip latency efficiency. It only tells you about processor efficiency. Which isn't what we're talking about.

I don't dispute that if you've got the CPU horsepower then you can lower your buffer size. And I don't dispute the practical reality. I don't really know where you got that from tbh.

What I'm saying is.... plenty of people misunderstand the role of the CPU when it comes to audio, and they don't take other factors into account. Which leads them down a path of buying a crap audio interface, a crap motherboard, crap RAM sticks, and crap hard-drives, and then they wonder why they can't get good latency performance on their machine. Because they over-estimated the importance of the CPU.

With an Antelope Audio Discreet unit or a Presonus Quantum for example, running Thunderbolt, at 256 buffer size, you quite often get better RTL's than a lot of USB2 audio interfaces running at 128 buffer size.

So what does that mean? It means your CPU is running more efficiently because of the higher buffer size, meaning you can add more plugins and tracks to your projects, but still benefit from lower CPU. It also means that on a less than stellar machine, you can still get great RTL's by choosing the right audio interface.

The Presonus Quantum is particularly interesting, because they completely eschew the need for a built in mixer on the device. This brings down latency a LOT, which is why it's the fastest thunderbolt interface on the market, reporting RTL's of under 1ms.

I've been thinking of moving from my Universal Audio Apollo x8p to either the Antelope Discreet 8 or the Presonus Quantum, because I want lower latencies in my DAW.

The sentence you quoted by me was meant to be taken in context of the entire post. Not singled out as a factoid to tear apart.

thegummy

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

thegummy said:

WiresDreamDisasters said:

A faster computer quite often only gives you marginal improvements to round-trip latency. Because the CPU isn't really the bottleneck when it comes to latency.

I don't agree with this statement at all.

The CPU determines how much processing can be done at each buffer size setting without audio breakup. And the buffer size setting is directly linked to how much latency there will be. I've never seen a single audio interface that doesn't give lower latency for a lower buffer size.

There are tons of benchmarks available that shows that the faster the CPU, the more processing can be done at a particular buffer setting.

It's negligible versus having a better driver implementation. If you've got a piss poor driver, then even having a fast as fuck CPU you're still going to run into CPU spiking issues and audio-dropouts. You can have a fast as fuck CPU, but if your RAM sticks have dodgy timings, then you're not going to get good performance at low latencies, period.

A faster CPU does change your latency, but it's minimal, given all other factors. When I upgraded from my old quad core Q6600 processor to an i7, I saw less than a 2ms difference. It wasn't huge, and if I had bought the CPU solely for latency improvements, I would've been pissed off.

The CPU isn't the bottleneck, and it isn't the only factor to pay attention to. The reason I bring this up is because a lot of people think they can just plunge a load of money into a CPU improvement and see radical changes.

You can't.

Being more performant at similar latencies; ie - upgrading the CPU and seeing 40% CPU usage at a buffer size of 64 with a RTL of 8ms, instead of the previous 60% with the previous CPU - is not the same as *improving* latency... which is my point.

I'm not arguing against the fact that more processing can be done at a particular buffer setting. But that will just lower CPU usage. It wont drastically bring down your latency in milliseconds.

Hold on, are you saying that moving from one processor to another changed the latency without changing the buffer size setting?

Or that the buffer size setting you could then move to with the faster CPU only have a 2ms improvement?

What interface and what settings is this?

Changing to a faster processor can potentially decrease round-trip latencies at the SAME BUFFER SIZE. But it's marginal. Nothing to write home about.

Yes, a faster processor will mean you can lower your buffer size. But then you're not comparing like for like. You can't say that on machine X with a buffer size of 128 you get 8ms of latency, but on the brand spanking new machine Z with a buffer size of 32 you get 2ms of latency, so OMG this cpu is amazing!!!

.. doesn't work that way. You have to compare like for like.

The difference in latencies between machines at the same buffer size and sample-rate is down to a whole manner of things. Different thru-put rates in the PCI or USB bus, the bridge controllers on the motherboard, all sorts of things.

As I said - the CPU isn't really the bottleneck when it comes to latency.

Wow that's a very strange way to look at it, saying that the practical reality of it doesn't matter...

The reason someone gets unacceptable latency is that if they reduce the buffer size their PC can't keep up so the audio drops out.

So upgrading to a better PC allows them to lower the buffer size to get the latency they want without the audio dropping out.

Believe it or not, some interfaces won't give you particularly good latency figures even at lower buffer sizes. I remember the old 1st generation Digidesign Mbox units were pretty terrible for this. Even at low buffer sizes you'd see RTL's of 10+ ms. The computer would be handling the audio perfectly fine, but the driver just couldn't deliver low latencies.

If you're talking about comparing latencies across different machines, then you have to use the same metrics to get a scientific comparison. Comparing 128 on one machine versus 32 on another machine doesn't tell you anything about round trip latency efficiency. It only tells you about processor efficiency. Which isn't what we're talking about.

I don't dispute that if you've got the CPU horsepower then you can lower your buffer size. And I don't dispute the practical reality. I don't really know where you got that from tbh.

What I'm saying is.... plenty of people misunderstand the role of the CPU when it comes to audio, and they don't take other factors into account. Which leads them down a path of buying a crap audio interface, a crap motherboard, crap RAM sticks, and crap hard-drives, and then they wonder why they can't get good latency performance on their machine. Because they over-estimated the importance of the CPU.

With an Antelope Audio Discreet unit or a Presonus Quantum for example, running Thunderbolt, at 256 buffer size, you quite often get better RTL's than a lot of USB2 audio interfaces running at 128 buffer size.

So what does that mean? It means your CPU is running more efficiently because of the higher buffer size, meaning you can add more plugins and tracks to your projects, but still benefit from lower CPU. It also means that on a less than stellar machine, you can still get great RTL's by choosing the right audio interface.

The Presonus Quantum is particularly interesting, because they completely eschew the need for a built in mixer on the device. This brings down latency a LOT, which is why it's the fastest thunderbolt interface on the market, reporting RTL's of under 1ms.

I've been thinking of moving from my Universal Audio Apollo x8p to either the Antelope Discreet 8 or the Presonus Quantum, because I want lower latencies in my DAW.

The sentence you quoted by me was meant to be taken in context of the entire post. Not singled out as a factoid to tear apart.

We're talking about what can be done to improve latency, not debating any kind of technical detail.

What interfaces don't have acceptable latencies at the lowest buffer setting? Not any that you can currently buy, even the cheapest.

The fact is that if someone has unacceptable latency, upgrading to a faster CPU definitely will solve that problem for them because it will let them lower the buffer - which definitely isn't already at its lowest setting if they're getting problematic latency when using a guitar sim.

Cirrus

On the other hand... if you're getting problems on a low buffer setting with, say, an Audient interface, you may well get better performance switching to an RME interface when using the same size buffer, even if you change nothing else about the system 

WiresDreamDisasters

thegummy said:

We're talking about what can be done to improve latency, not debating any kind of technical detail.

What interfaces don't have acceptable latencies at the lowest buffer setting? Not any that you can currently buy, even the cheapest.

The fact is that if someone has unacceptable latency, upgrading to a faster CPU definitely will solve that problem for them because it will let them lower the buffer - which definitely isn't already at its lowest setting if they're getting problematic latency when using a guitar sim.

I don't have an exhaustive list, but I'm sure there are going to be interfaces that give piss poor performance at low buffer settings regardless of the computer. I had to send a Roland Quad-Capture back to the shop because it was just absolute piss for audio-dropouts and poor latency performance. Don't know if they ever fixed it. Craig Anderton enjoyed his though!

I also had a MOTU 8-Pre which was very good at low buffer sizes. Very performant at 64. But I couldn't get it under 6ms RTL.

If someone has a card, and its running at 6ms RTL at a buffer size of 64, and the driver for that card only goes down to a buffer size of 64, then upgrading their computer is not going to do much to that 6ms figure. The Focusrite Saffire USB interfaces are known to be poor in this respect, and they only stopped making those within the last 6 years or so.

This situation is almost certainly going to happen with modern cards, because as I said.... there are only 3-4 off-the-shelf USB audio drivers around. Manufacturers customize them for their needs, but performance between driver implementations are fairly similar across devices.

Also an additional thought - some bright spark with some devices (Focusrite is the one I can think of) decided to allow their control panels to specify buffer sizes in ms figures, rather than block sizes that are power of 2. So if you've got a weird block size... like 100 for example... plugins aren't going to like that!!

What can be done to improve latency? Several things:

- Get an audio interface with faster drivers - RME, Antelope Audio, Avid HDX, Apogee
- Run your interface as thunderbolt over USB if it allows it (not many do)
- Run your projects at a higher sample-rate (doubling the sample-rate will roughly halve the amount of latency)
- Upgrade the CPU within the same socket series (minimal impact because no other components will change)
- Upgrade the CPU+motherboard+ram to a known good configuration
- Research which base driver your interface uses (not easy because they hide this knowledge)


There's a monster thread here all about this issue:
https://www.gearslutz.com/board/music-computers/618474-audio-interface-low-latency-performance-data-base.html

thegummy

Cirrus said:

On the other hand... if you're getting problems on a low buffer setting with, say, an Audient interface, you may well get better performance switching to an RME interface when using the same size buffer, even if you change nothing else about the system 

Absolutely - that's what I said originally, the two ways to get better latency would be either a better PC or an interface with better latency.

I did mention Zoom as being a more wallet friendly option though since even the cheapest RME interface costs as much as a new computer!

For the sake of transparency though, if my new PC doesn't get me the kind of sessions I want at the latency I want it will be an RME I go for rather than the Zoom.

WiresDreamDisasters

Cirrus said:

On the other hand... if you're getting problems on a low buffer setting with, say, an Audient interface, you may well get better performance switching to an RME interface when using the same size buffer, even if you change nothing else about the system 

*ahem*

Yes. Because of drivers.

I'm fairly sure the Audient devices use the Thesycon driver, which doesn't have a good rep.

stratman3142

thegummy said:

LooseMoose said:

Which Zoom interface do people recommend?

There are only really two - the UAC-2 with 2 inputs and the UAC-8 which I'd assume has 8 so it depends how many inputs you need.

I'm using a Zoom UAC-2. The reported specs for a Focusrite Scarlett gen 2 or 3 seem similar but I haven't got one to try. 

With my Zoom, I get 5.4ms at 44.1kHz at 64 spls without any issues.

thegummy

WiresDreamDisasters said:

thegummy said:

We're talking about what can be done to improve latency, not debating any kind of technical detail.

What interfaces don't have acceptable latencies at the lowest buffer setting? Not any that you can currently buy, even the cheapest.

The fact is that if someone has unacceptable latency, upgrading to a faster CPU definitely will solve that problem for them because it will let them lower the buffer - which definitely isn't already at its lowest setting if they're getting problematic latency when using a guitar sim.

I don't have an exhaustive list, but I'm sure there are going to be interfaces that give piss poor performance at low buffer settings regardless of the computer. I had to send a Roland Quad-Capture back to the shop because it was just absolute piss for audio-dropouts and poor latency performance. Don't know if they ever fixed it. Craig Anderton enjoyed his though!

I also had a MOTU 8-Pre which was very good at low buffer sizes. Very performant at 64. But I couldn't get it under 6ms RTL.

If someone has a card, and its running at 6ms RTL at a buffer size of 64, and the driver for that card only goes down to a buffer size of 64, then upgrading their computer is not going to do much to that 6ms figure. The Focusrite Saffire USB interfaces are known to be poor in this respect, and they only stopped making those within the last 6 years or so.

This situation is almost certainly going to happen with modern cards, because as I said.... there are only 3-4 off-the-shelf USB audio drivers around. Manufacturers customize them for their needs, but performance between driver implementations are fairly similar across devices.

Also an additional thought - some bright spark with some devices (Focusrite is the one I can think of) decided to allow their control panels to specify buffer sizes in ms figures, rather than block sizes that are power of 2. So if you've got a weird block size... like 100 for example... plugins aren't going to like that!!

What can be done to improve latency? Several things:

- Get an audio interface with faster drivers - RME, Antelope Audio, Avid HDX, Apogee
- Run your interface as thunderbolt over USB if it allows it (not many do)
- Run your projects at a higher sample-rate (doubling the sample-rate will roughly halve the amount of latency)
- Upgrade the CPU within the same socket series (minimal impact because no other components will change)
- Upgrade the CPU+motherboard+ram to a known good configuration
- Research which base driver your interface uses (not easy because they hide this knowledge)


There's a monster thread here all about this issue:
https://www.gearslutz.com/board/music-computers/618474-audio-interface-low-latency-performance-data-base.html

Bizarrely the very link I was going to give you you've posted yourself at the end.

That has so many RTL results of interfaces, if you go through you'll see how rare it is to see an interface that has noticeable latency at its lowest buffer setting.

Even if it was common though, it wouldn't be that much of a problem because you can buy one for under 100 quid that does go down to low latency.

thegummy

stratman3142 said:

thegummy said:

LooseMoose said:

Which Zoom interface do people recommend?

There are only really two - the UAC-2 with 2 inputs and the UAC-8 which I'd assume has 8 so it depends how many inputs you need.

I'm using a Zoom UAC-2. The reported specs for a Focusrite Scarlett gen 2 or 3 seem similar but I haven't got one to try. 

With my Zoom, I get 5.4ms at 44.1kHz at 64 spls without any issues.

It's a Scarlett I'm using and at 64 buffer is where I get the 8.4ms so the Zoom is literally the difference between unnoticeable and noticeable.

It's a shame it's only got 2 inputs though, that's the main reason I'm going for an RME rather than a Zoom if I need to get a new interface.

WiresDreamDisasters

thegummy said:

WiresDreamDisasters said:

thegummy said:

We're talking about what can be done to improve latency, not debating any kind of technical detail.

What interfaces don't have acceptable latencies at the lowest buffer setting? Not any that you can currently buy, even the cheapest.

The fact is that if someone has unacceptable latency, upgrading to a faster CPU definitely will solve that problem for them because it will let them lower the buffer - which definitely isn't already at its lowest setting if they're getting problematic latency when using a guitar sim.

I don't have an exhaustive list, but I'm sure there are going to be interfaces that give piss poor performance at low buffer settings regardless of the computer. I had to send a Roland Quad-Capture back to the shop because it was just absolute piss for audio-dropouts and poor latency performance. Don't know if they ever fixed it. Craig Anderton enjoyed his though!

I also had a MOTU 8-Pre which was very good at low buffer sizes. Very performant at 64. But I couldn't get it under 6ms RTL.

If someone has a card, and its running at 6ms RTL at a buffer size of 64, and the driver for that card only goes down to a buffer size of 64, then upgrading their computer is not going to do much to that 6ms figure. The Focusrite Saffire USB interfaces are known to be poor in this respect, and they only stopped making those within the last 6 years or so.

This situation is almost certainly going to happen with modern cards, because as I said.... there are only 3-4 off-the-shelf USB audio drivers around. Manufacturers customize them for their needs, but performance between driver implementations are fairly similar across devices.

Also an additional thought - some bright spark with some devices (Focusrite is the one I can think of) decided to allow their control panels to specify buffer sizes in ms figures, rather than block sizes that are power of 2. So if you've got a weird block size... like 100 for example... plugins aren't going to like that!!

What can be done to improve latency? Several things:

- Get an audio interface with faster drivers - RME, Antelope Audio, Avid HDX, Apogee
- Run your interface as thunderbolt over USB if it allows it (not many do)
- Run your projects at a higher sample-rate (doubling the sample-rate will roughly halve the amount of latency)
- Upgrade the CPU within the same socket series (minimal impact because no other components will change)
- Upgrade the CPU+motherboard+ram to a known good configuration
- Research which base driver your interface uses (not easy because they hide this knowledge)


There's a monster thread here all about this issue:
https://www.gearslutz.com/board/music-computers/618474-audio-interface-low-latency-performance-data-base.html

Bizarrely the very link I was going to give you you've posted yourself at the end.

That has so many RTL results of interfaces, if you go through you'll see how rare it is to see an interface that has noticeable latency at its lowest buffer setting.

Even if it was common though, it wouldn't be that much of a problem because you can buy one for under 100 quid that does go down to low latency.

It depends what you think is noticeable I suppose, which brings us back to your original post. The Helix at 13ms RTL at 64 buffer size is not quite good enough for me. I can "feel" something... don't know what it is, but it's not as instantaneous as it is when it's just used as a multi-fx.

My Apollo is great when the latency compensation is turned off. But then my CPU usage goes through the roof. So in that scenario, you're exactly right - upgrade the CPU to as hefty as I can afford, so that I can run as many plugins as I want (hope) to.

When it's at medium setting, at 64 buffer size, the CPU usage is less. But for a 100 track project with tons of plugins, I have to set my buffer to 1024. Which gives me a ridiculous latency figure.

I am gonna upgrade my machine at some point, but I'm also gonna get another audio interface, so swings n roundaboobies!

Ideally a RTL of 6ms or less is what I'd like. Anything higher than that, I start to feel it.

thegummy

WiresDreamDisasters said:

It depends what you think is noticeable I suppose, which brings us back to your original post. The Helix at 13ms RTL at 64 buffer size is not quite good enough for me. I can "feel" something... don't know what it is, but it's not as instantaneous as it is when it's just used as a multi-fx.

My Apollo is great when the latency compensation is turned off. But then my CPU usage goes through the roof. So in that scenario, you're exactly right - upgrade the CPU to as hefty as I can afford, so that I can run as many plugins as I want (hope) to.

When it's at medium setting, at 64 buffer size, the CPU usage is less. But for a 100 track project with tons of plugins, I have to set my buffer to 1024. Which gives me a ridiculous latency figure.

I am gonna upgrade my machine at some point, but I'm also gonna get another audio interface, so swings n roundaboobies!

Ideally a RTL of 6ms or less is what I'd like. Anything higher than that, I start to feel it.

6ms is the figure where no one can really feel it, that's the ideal to shoot for really.

I think some people who have been playing real amps from across the room for years would possibly be used to higher latencies but I've always used amp sims and mostly through headphones so I'm used to no noticeable latency.

WiresDreamDisasters

First time I played on a big stage was an odd experience. Coz I definitely felt some latency with the cabs being 4 meters away from me. I always bring them forward a bit now. Well.. I used too... when gigs were a thing....

I prefer playing smaller places really. Better vibe. 

octatonic

The advantage of HDX is I just never think about latency.

The cost of entry is vast- but I can track a full band through it, manage the tracking and cue mix from the same location.
It is really seamless compared to all other solutions and I know this because I've had most of them here, either as purchases or as reviews.