Ableton Forum

djsynchro wrote:

epjl2000 wrote:Is 2.36ms decent?

2.36 is shite my latency is 0.000000000000096 us (MU=micro)
I am on the latest Jappletosh coca-cola cooled octocore though.

Dude, you should use Jolt Cola instead, it'll give you negative latency. Live will actually play your chops before you do. Unfortunately, Jolt has gone bust. But you could always try eBay.

When I play my guitar/bass/keyboard etc.. It sounds like I'm playing it right now as opposed to say 0.3 seconds ago.

That’s my overall latency.

I've never calculated this in numbers, always 'set my latencey' doobry at the point just before it starts to crackle to keep it low.

(Acer Dual Core Travelmate (2. 2 GB ram - Fast Track Pro)

Man, when I first plugged my guitar into my old PC many moons ago I never thought i could actually use the technology as the latency was shocking!

Now I can play me bat and it comes out of the speakers when I play just like it did in the 80's. Sorted.

To the OP:

ASIO4ALL is a wrapper and still uses your driver program. The wrapper adds latency to the value that you see reported in Live. So..... your reported value of 2.36 (output?) latency is actually higher.

BTW what's your total value reported and how many VSTs and audio channels can you support running at that level?

AD/DA is never mentioned in latency figures (usually around 1ms each way, so 2ms), any hidden buffers (naughty, naughty) are not always declared, and don't forget, the VST buffer for softsynths which is seperate from audio, but the default is 'set to same as audio' which effectively doubles the latency when using softsynths - you can access this manually and change it but often taking it to 64 samples or less leads to choking with more demanding softsynths such as Omnisphere or Ace.

Typically anything under 256 samples is fine for most intents and purposes, a reported round trip latency of 11-13ms (but probably closer to 20ms with all the above added, not including VST buffer).

Do your tracking first and save piling on plugins for your final mixdown helps a lot, when mixing you can step up the latency so as not to strain the CPU.

How does Live calculate the displayed latency anyway? With my Echo Audiofire 2, 64 samples gives me 2 x 4.35ms. Shouldn't it be 2 x 1.45?

Not that Live is usable with this latency anyway! The 80% CPU tone test suggests I need at least 192 samples (2 x 7.26ms).

But it seems that I can get half the latency with ASIO4All (128 samples and 2 x 3.63) than with my native drivers. That's can't be right, can it?

Either way, I've never really felt it's been low enough for playing virtual instruments live. There's always been an annoying lag.

What people also fail to consider when publishing 'I can get X' latency is using multiple tracks, especially multiple vsts and also that non Live devices, i.e. vst synths etc require an additional buffer which by default is the same as the audio latency (effectively doubling latency) but if you manually adjust it 32 samples is the lowest, but many synths in Live need a bigger buffer than this or crap out. Too many people give their low latency figures based on a very light Live set usually based on audio tracks, which is less demanding than midi tracks with vsts.

If you can use sub 128 samples (with a 64 samples vst buffer) on a 20 track set with 5 or more vst instruments, then you have pretty much unbelievable performance and shouldn't worry. Even 256 samples with the above would be manageable.

As for drivers, asio4all can sometimes significantly improve on native drivers, but isn't always the best option for mixing down - I find on my Alesis IO14 that the asio4all driver drops out less, but does things like narrows the stereo image when it is getting strained and somehow seems less clear - this might not apply on every interface but is definitely an issue for me, so I don't mind tracking with asio4all but when it comes to mixdown, I step up latency for smoother mixing with less CPU load and use the native drivers.

I wrote something else in this thred; there was several days latency.
Not that anybody cares.

I misspelt "thread"