The chance of zero latency

[jestermgee]

While the comments invoking the laws of nature / physics are all correct, ditto the statements explaining that any processing will take CPU cycles (and hence time, therefore latency) - I think there is a way to reduce latency below what we consider ultra-low latency under the current paradigm.

Process audio in smaller chunks. I believe this can result in lower latency for processing within plugins / DAWs etc. Data I/O such as PCIe, USB3 and TB3 should certainly support very small buffer sizes.

Of course, your CPU will be working *very* hard to achieve this!

If any one entity were to accomplish a move in this direction and have a chance of success, it would almost certainly be Apple. Not because they have any magic technology but simply because they control the whole stack from hardware to the OS right down to having an in-house DAW.

I welcome any criticism or further info that might support or invalidate my hypothesis.

Most latency will have nothing that Apple can solve because as you acknowledge, physics plays a part and without the ability to pause time while numbers are crunched, there is always time needed to calculate sounds on the fly. Apple may have been the wonderbread 15 years ago when every CPU cycle mattered and many standard for computers were still forming, but now users have just as many issues and hangups on Mac as any OS, especially given their update cycles tend to break everything and make things a mess for months. It is now just what one prefers to use.

There is also the case where people need to recognise there is the production stage of a track and the mixing stage and these are separate things. Many processing plugins add latency which cannot be avoided regardless of the OS. Also, focus on efficiency in production is just as important using only what is needed (does every track need a compressor and a FabEQ?).

[subparuser]

Most latency will have nothing that Apple can solve because as you acknowledge, physics plays a part and without the ability to pause time while numbers are crunched, there is always time needed to calculate sounds on the fly. Apple may have been the wonderbread 15 years ago when every CPU cycle mattered and many standard for computers were still forming, but now users have just as many issues and hangups on Mac as any OS, especially given their update cycles tend to break everything and make things a mess for months. It is now just what one prefers to use.

There is also the case where people need to recognise there is the production stage of a track and the mixing stage and these are separate things. Many processing plugins add latency which cannot be avoided regardless of the OS. Also, focus on efficiency in production is just as important using only what is needed (does every track need a compressor and a FabEQ?).

Yes, ecuk provided the information regarding circumstances where a certain fixed latency will always apply - we're in agreement here.

I would say the only wonderbread Apple provided 15yrs ago for audio users was a hefty vector processing unit on the CPU *that actually got used* by audio devs. I may be misremembering, but SSE/2 code on the Windows side was something of a rarity (and the Intel SSE hardware was a bit anaemic by comparison to VMX (AltiVec)).

However . . . and it's important to note - I am making no claim that Apple have any secret sauce or wonderbread solution to the problem beyond their control of the entire stack, soup to nuts. I guess other DAW / plugin vendors could decide to process audio in smaller chunks (where the nature of the operations do not require discrete time processes as per ecuk's excellent summation) but Apple not only control the direction of a major DAW, they also control a plugin standard.

Your point about separating production-phase from mix-phase is well taken - but some people work differently, and there is absolutely nothing wrong with that. I believe I've already outlined a use-case where lower latency would be a benefit - multiple roundtrips out of the audio interface and back in real time.

Slapping a compressor and a FabEQ on every track doesn't seem the best example for the purposes of this discussion* as I'd assume the FabEQ falls under ecuk's example of FFT-based processing and per-track wouldn't add latency cumulatively?

Anyway, it's been an interesting topic despite its . . . ummm . . . inauspicious beginnings!

* Narrowly, smaller buffers / processing chunks.

[Ast3rix]

The basic way computers operate hasn’t changed that much in decades - the hardware has got faster and more efficient and the software often better optimised, but until and unless quantum computing becomes a usable thing we’re stuck with the technological approach we have.

I think there are ways around many of these issues. I have seen very stable performance as well as lower latency in studio one 5 and it uses steinberg's vst model. As an extreme model look at Bitwig they are thinking more inline with how you should handle third party apps.

I look forward to seeing the details of your time machine patent - or your patent for a completely new universe with different physical properties to the one we inhabit, whichever is easiest to implement. A guitar plugged into an analogue effect and then an analogue amplifier effectively has zero latency - though the speed of light and how fast an electric field propagates through the circuitry imposes some latency it’s so tiny as to be ignorable even if your guitar leads are miles long.

I am working in this realm of reality. I have multiple DAW's to prove my point. I'm just tired of the one that I use the most being the most unstable as well as being the worst with latency. I can't blame it on my audio interface because it performs way better in studio one 5.

I’ve not had Live 10 crash on me in ages. And I can get the same latency times out of Live as I can Logic Pro. Often a bit better.

Now this is a funny statement!!! I wish I didn't experience crashes... I use a lot of third party plugins and they are all 100% licensed and updated. My problem is real.... it's painful and it sucks...

Live is good at telling you how much “built in” latency a plugin has, which is something many plugin manufacturers are a bit shy about revealing (Waves being a noticeable exception). Personally I find a round trip under 12 milliseconds with nearfields about 4ms away from me to work OK.

I on average without plugins have 12ms of latency. It is noticeable on near fields and then it progressively gets worst when I add plugins. My only work around is to use UAD's realtime features. It adds an extra step to my workflow that I hate. It's the only way I can play live with Ableton live 10. When in studio one 5 its reduced down to 9ms. It does not get worse as I add plugins. So yeah studio ones audio sub system is more modern.

[subparuser]

I think there are ways around many of these issues.

Okay . . . name one then.

[Ast3rix]

Re-code the audio sub section to bring it up to today's standards.

[jestermgee]

Re-code the audio sub section to bring it up to today's standards.

Recode what exactly and how to match which standards that you refer to specifically? Seems like a pretty generic/vanilla suggestion

It's one thing to throw out a suggestion, but another to have some useful substance to it. Why not just say "recode ableton from the ground up to work on every system perfectly in every way... simple!".

[Ast3rix]

[/quote]

Recode what exactly and how to match which standards that you refer to specifically? Seems like a pretty generic/vanilla suggestion

It's one thing to throw out a suggestion, but another to have some useful substance to it. Why not just say "recode ableton from the ground up to work on every system perfectly in every way... simple!".
[/quote]

I don't mean to make this some generic unrealistic topic, but the standards I refer to are those that both Presonus and Apple have created. If you have spent any time in either DAW you will find that they have spent time optimizing the audio sub system as the long standing issues of latency have been managed in a very logical way. I know that every musician that uses computer based recording would appreciate if all of the DAW makers would join forces to tackle this 30 year old problem so that we all can continue forward. The buffer workarounds don't work for many of us and those that have minimal resources struggle even more.

[jlgrimes]

Is there any chance we will see a zero latency Ableton in our life times? I was hoping that Ableton would be the first to claim zero latency. No adjusting buffer sizes and turning off monitoring or spending excessive money on analog summing gear.

Question should probably be will we ever reach a chance where latency effects are miniscule on all modern systems. Id think that's a more realistic goal.

My guess is probably once USB4.0 gets adopted, latency will get dramatically decreased.

Also maybe at some point better driver technologies, faster CPUs will help as well.


That said latency is a very normal issue that people had to deal with since large orchestras and a computer will always take a finite amount of time to do a calculation, read data, perform actions.

But at some point it probably will get miniscule. But this also goes against the grain of demands of more higher quality plugins, effects, programs that are more complex.

Practical Low latency is already achievable but does require knowledge of the end user to manage resources.

Someone performing live will have different demands than a mixing engineer who probably enjoys a higher CPU overhead at the cost of higher latencies.

At some point though maybe it will get to a point where raising buffer sizes won't have much effect on CPU overhead, then it will be simpler to make an "easier" driver but right now, there is still a good performance boost when going from 64 samples to 512, but like 20 years ago, it was difficult to have a system that was capable of running at 64 samples reliably, where now it is pretty normal on many systems even using a low end audio interface.

I'd guess maybe 10 years out this will be more normal once USB specs are improved at a low cost to end user and interface manufacturers start adopting these.

Ableton (or any DAW alone) can't cure the latency issue.

[properLofi]

I don’t really see that a new USB standard is going to dramatically decrease latency. The throughput of USB is more than adequate to shift the data rates required for high sample rate, high bit depth audio to a pair of monitors. That’s not really the bottle neck. It’s more to do with the computations required to create the audio buffer and the speed with which that buffer can be converted into an electrical current for your monitors. Good interfaces are extremely efficient in converting the digital buffer to an analogue signal and provide robust and efficient drivers to support the work of the CPU. The motherboard buses (PCI, USB) have been able to convey multi-channel audio for a long time but the calculation and conversion takes time. The time taken is restricted by the clock rate of USB rather than the transfer speed. Thunderbolt alleviates that to some extent by increasing the bus clock speed from USB.

I think we are realistically reaching the latency limits of software. I’ve worked extensively in Logic and Ableton and see very little real world difference in terms of latency other than when using non-ASIO devices on Windows.

However, there is a solution and that appears to be interfaces that alleviate CPU reliance by building devices that can utilise DMA (direct memory access). This allows the interface to pull samples directly from the memory of the system bypassing CPU and alleviating buffer latency. PreSonus claim to have done this with the Quantum which on paper (never tried one) allows sub 1ms round trip with processing in a DAW. At $600 for 8 in / 8 out it also seems relatively affordable.