At what buffer setting does it negatively impact recording using Maschine's pads?

At what buffer setting does it negatively impact YOUR recording using Maschine's pads? I have been having a robust debate with hardware users regarding whether Maschine can match the response and timing of dedicated hardware.

Using Maschine at 512 Samples...Input 19.2 ms, Processing 11.6 ms, Output 1.9 ms for a total of 32.7 ms, I can't hear or feel any major differences to using dedicated hardware. Likewise, taking my audio interface down to 32 Samples...Input 8.3 ms, Processing .7 ms, Output 1.9 ms for a total of 10.9 ms, I don't hear or feel any noticable improvement.

NI has these David Haynes videos. I am curious to know his buffer settings/report latency of his audio interface.

 

for the most part, i use 512 on my Macs and it feels pretty instantaneous. i have tested at much lower sizes in terms of audio quality, but have not needed to reduce the buffer sizes for musical reasons, but then i am not a finger drummer.

i hope flux posts on this thread.

sowari

 

This thread is not about that. I even modified my post to ensure it is not about that. If you modify yours, there will be no mention of said product.

 

My views on this neverending software latency debate:

Staying under 8-10 ms total latency is the ultimate goal, i heard an human can perceive latency if greater than 15Ms. You need latency as low as possible for "recording", but high enough in order not to have cracks when "playback/processing vst".

512 sample seems a lot to me (from a pc perspective) i would recommand to go down until you start hearing cracking sounds within a big project, then go up one step.

I think on finger drumming demos latency must be as low as possible (32 sample) as it is using sampler features only (no vst, not much "processing time" added => go for low latency).

I'm on pc with an RME babyface and i run maschine with 128sample@48Khz for a total latency of about 7,8Ms; i can't feel any difference with hardware.

Within a full DAW environment t i can go from 96 to 256 samples as the project goes "bigger"

 

So when have RME set to 128 samples @48Khz, what does Maschine report specifically:
1. Input
2. Processing
3. Output
4. Overall

I keep my Mac at 512 because setting it to 128 or lower and I hear/feel no differences.

Same question....what does Maschine report when you have your interface set to 512 ms?

 

Here's what i got running @ 48/44.1

input:3.8/4.1
processing:2.7/2.9
output:1.3/1.4
Overall:7.8/8.4

From what i know rme card are running internally @ 48Hz and so it need extra dsp processing to "downgrade" to 44.1, that's why latency is a little bigger @ 44.1Hz.

For some card it can be the exact contrary; some high-end pro cards will work natively @96hz and will have a drop of performance running@ 44.1/48hz.

If you hear no difference running 128 samples then go for it then, the day you hear clicks/pops it means your cpu gets maxed out then go for 256.

 

This is a personal thing on what point you will hear the difference. Everyone has different levels of perception or what they consider acceptable.

A glass of water always seems clean until you put it next to one that really is. Also note that some sound cards add double the buffers either on input or output or even both. These extra buffers are not always reported to the host. You may not feel much difference between settings because some cards simply can't get rid of the latency no matter the setting. Another thing is if you monitor your hardware through your sound card then it is also subject to the latency of your soundcard. You have to plug your phones directly in to the hardware to compare the difference.

On my sound card I use 64 buffers @48k and feel no latency at all. This gives 4.8ms total (2.7 input +1.3 precessing + 0.8 output.) and this card does not add extra buffers. If I use 32 samples for 3.5 ms I can't hear any difference. 128 buffers (7.5ms total) I can feel the delay but it's workable. Anything over that is unusable for me because it throws off my timing between what is heard and what I play.

 

As far as hardware goes, there's latency there as well- after all, the internal processor still has to deal with note ons and offs, etc.

And, if you're talking hardware synths, the difference between first and last note ons can easily exceed 15 ms, which you won't see in software; everything will come out at the same time.

ew

 

True, but it depends on the hardware. Typical workstations like Fantom, Motif or MPC have such low latency that it is unnoticeable. Usually under 2 ms after DA. A product like Maschine that is used to sequence primarily percussion in real time makes latency an important issue. Of course it all depends on your sound card and computer but you can achieve what even a picky type would perceive as "zero latency" on Maschine.

 

Found this on the subject:

http://www.presonus.com/community/Learn/The-Truth-About-Digital-Audio-Latency

and this:

http://www.focusrite.com/answerbase/en/article.php?id=161

Wondering how one's can measure the latency of the human ear with precision ^^

 

Interesting. But a false statement here.

Did the blind test and I can hear the difference between 5 ms and 11 ms 10 times out of 10 and I am not the only one. Depends also on the material. Percussion playing is much more sensitive vs playing pads etc.

The Presonus article is very informative and kind of goes into how a sound cards latency setting can't be taken at face value due to safety buffering. Not all sound cards do that though.

 

i find it tough to get good grooves when fingerdrumming if the latency is over 15ms.

for me, this is because the audio i hear and the movement of my fingers have enough of a delay that it causes my brain to go 'huh?' and i get out of the groove - you know the feeling

a good compromise for me has been
sample rate: 44.1k
latency: 160 samples
input > 6.7ms | processing > 3.6ms | output > 3.0ms | overall > 13.3 ms

it really depends on your audio device to find the compromise between latency and no dropouts.

 

Hello,

don't know if i am on the right side of thinking, but i think if you Fingerdrum with maschine, only the half processing and the output latency is the delay between touch the pad and the signal out of audio interface.

Ive forget the delay of signal from hit pad to signal over USB to PC, but this has nothing to do with the audio interface delay i think !

 

What sound card/audio interfaces are you guys using with Maschine to get your ultra low latency?

 

Don't forget, there's also unreported latency (ADC/DAC/Safety buffers/Safety Offset) which can be quite big with some audio interface manufactures, especially working with USB or Firewire devices.
The Machine software doesn't know about this unreported latency so it won't report it.
You can measure the total unreported latency quite easy though, with Centrance's LTU or by a roundtrip recording and measure the difference in samples yourself.

Make the following sum to determine the unreported latency.

Outcome roundtrip in samples - 2x ASIO buffer size in samples (In & Out) = unreported output latency in samples / through sample rate = unreported output latency in milliseconds.

 

here im using edirol fa-66 at 48K Hz sample rate

the best i can reach is a 256 samples buffer. but if i run jackrouter in exclusive mode (hog mode) i can reach down to 128 samples because audio device is not shared (there are no device interrupts)

i dont know if coreaudio can give a program exclusive access to the audio device. i only saw this kind option in very few programs such as Jack, Audirvana, ASIO4ALL (win), ...


_________________

not sure but i think you are wrong (please someone confirm this). also, there are a lot of different algorithms to resample a waveform.

the more the rate of the sample, the bigger the buffer size is, the more the time to process a buffer.

 

The higher the samplerate, the faster samples wil make their way through the buffer(s).
So in theory
256 / 44100 = 0,0058 (5,8ms)
256 / 88200 = 0,0029 (2.9ms)

Of course the higher the samplerate the more calculation cycles are required from your CPU.

 

couldn't you just go down as low as you can, record your drum pattern, and then bump it back up to cpu efficient buffer rate for playback?

 

yeah i was menanin that. thanks nilesp for clarification.

 

I'm using a Duet at the moment and I usually have my buffer setting at 128 to record and bump up to 512 when editing. Before ML I usually always had my settings at 256 which was about 10. something in latency which isn't bad imo. However anything higher than that is way too noticeable and for some reason sometimes the latency when recording gets higher or lower depending on what I have going with the project (which tend to get very heavy for me. I like a lot of instruments). When I'm playing a keyboard controller the higher latency doesn't bother me all that much so I tend to run Maschine at 512 when doing instrumentation and 128 when doing drums.