Survey - Buffer Size

[Timeline]

if the 2048 and + would be available , that would be it...... Always 1024 , thanx to Cue mix.....

FYI:

IO buffer limits are set with the particular IO you are using. I'm told RME IOs allow buffers over 5000.

I'm more partial to Thai buffers although more expensive.

[FM]

here in NYC we have the best chinese buffers in the world.
down in chinatown you can get the tastiest Dim Sum-ing buffers this side of the Yangtze.

FM

FM will buy some furniture and give the cat a name.

[RhythmRmixd]

Recording 256-512
Once everything is on audio tracks, always bump up to 1024.

[qo]

Yeah, I agree with Timeline that we should at least be including sample rate. I'm starting to move to 96/24, but current projects are in 44.1/24. I don't have enough experience at 96/24 at this point to be able to say what buffers I'm using.

One thing that I did notice though. Larger buffers at higher sample rates have less latency than the same sized buffer at a lower sample rate. Which makes sense since the buffer is simply storing samples temporarily.

So, for example a 256 sample buffer at 44.1kHz, should contribute, ideally, 256/44100 = .005 seconds of latency.

The same buffer at 96kHz is contributing 256/96000 = .002 seconds of latency.

So, roughly, the buffer size could be a little over twice as large at 96 and still result in the same perceived latency (all other things being equal).

That's only math though. There are other contributing factors with DP's implementation. For example, I'm pretty sure the Host Buffer Multiplier creates multiple parallel buffers, one for each "host" or interface. But, I'm not at all clear on whether HBM increases latency in some way (where the buffers are parallel, but DP's buffer service routines introduce additional delay). Perhaps magicd could chime in?

<small>[ August 04, 2005, 02:06 PM: Message edited by: qo ]</small>

[Shooshie]

128 buffer when doing MIDI
256 when recording multiple tracks
256-512 when mixing a few tracks
1024 when mixing lots of tracks

I usually don't switch buffers until either latency drives me to it or CPU meter threatens to shut me down. The above is just sort of the average of how it works out.

I work mostly in 48k/24bit, but occasionally at 88.2k. Then the buffers are different--basically x2.

Dual 1GHz G4.
1.5GB RAM

Shooshie

[Timeline]

Hi Shooshie.

My plan is to also cut any new sessions 88.2 instead of 96.

I'm pretty convinced that 96 is a waste but 88.2 is better than 48 sonically per my ears & Nyquest theorem plus, I might be able to eek out a few more goodies and keep thing crankin.

<small>[ August 04, 2005, 05:25 PM: Message edited by: Timeline ]</small>

[qo]

EDIT: this hung in mid transfer and ended up a partial post. Fixed...

Yeah, I agree Timeline. Really, I wonder why they didn't come out with something like 64kHz as a standard. Some say that multiples are easier to downconvert, but other say that any but the simplest conversion techniques don't work that way anyway. Rather they use a common multiple between the two sample rates, upconvert to the common multiple and then downconvert from that.

If you HAVE to work with both 44.1 and 48, then 64kHz has a lower common multiple against these than either 88.2 or 96 do. i.e.:

(44.1 * 48 * 64) &lt (44.1 * 48 * 88.2) &lt (44.1 * 48 * 96)

And, 64kHz is well above Niquist for our range of hearing, and would tax our DAWs less, etc.

Now, granted, if you're ONLY dealing with 44.1, (and you subscribe to the notion that downconverting is simply division) then 88.2 is your number.

<small>[ August 04, 2005, 06:50 PM: Message edited by: qo ]</small>

[joconnel]

1024 unless i am doing VI's...

[Timeline]

64kHz is well above Niquist for our range of hearing, and would tax our DAWs less, etc.

Maybe if IO manufacturers up the bits someday they will think about that too.

Reminds me when we used to think about changing the speeds of our 24 tracks from 30ips down to 24ips to match film standards and increase LF response for sound. Top end saturation would be perfect too.

Never did it though.

[bigdaddy]

64 for everything...Okay, so I'm kidding - So sue me.

Dual G4 450 w/1.25 GB ram. M-Audio AP 2496 and Delta 66 w/Omni I/O.

With the new combo of the AP and Delta, I've finally been recording more than 4 tracks at once. All at 44.1 16 bit.

If I'm tracking w/softsynths, I start at 128 and can sometimes work up to 512 (I save pads and slow attack sounds for last so latency doesn't matter).

Mixing from 20 - 30 tracks I can stay at 512 w/minimal FX plugs. More than 24 tracks w/lots of plugs requires 1024.

I've only recently started tracking 5+ tracks at once (drums), thanks to my adding the Delta 66. DP can easily track 5 - 6 tracks on my machine at 128 (with a few audio tracks playing back as well) with no probs. I plan on trying a full 10 tracks at some point, but haven't needed it yet.

I imagine that I won't have to go above 256 for that.

[qo]

Originally posted by Timeline:

64kHz is well above Niquist for our range of hearing, and would tax our DAWs less, etc.

Maybe if IO manufacturers up the bits someday they will think about that too.

Yeah! 64kHz/32bits standard. I love it.

[pcm]

I am on 1024 full time, for many years now. I monitor out to a console, through Cuemix, so there is no latency in my world. Plus, I don't use virtual instruments.

[grimepoch]

Question, concerning nyquist limit, as I did study this in school. If we record at 44.1kHz, nyquist says that the highest thing you can record would be 22.05kHz. This means if you actually had a signal at 22.05kHz, you'd get two samples for it.

Obviously, both things run asynchronously to each other, so depending on when the samples happened, you could be sampling at the zero point, or any other points which mean your 22.05kHz sound could be lost.

Has anyone tried recording a 20kHz signal multiple times to see what the amplitude does?

Even worse, with the jitter, I could see getting a waveform that doesn't stay consistently at an amplitude. I'd think any really high frequency components would suffer from this, for lack of a better term, amplitude modulation due to sample points.

[MT]

512 for both. Just seems to work the best for my setup.

I use the Lynx mixer for zero latency monitoring during tracking.

MT