Idiotsville United

[elvisfridge]

I wanted to up the anti re sample rates and projects from 24bit / 44.1 khz to 24bit / 96 khz.
Keen on the latency benefits also and processor hit (around 40 tracks audio, 10 tracks virtual instruments (not pussys) enough decent plug ins for Bob Clearmountain)

Tired of some of the bad info online. This is for people who know about this stuff (it goes in and out and takes a certain amount of time to get round) with a few observations thrown in for good measure, and a few questions.

Ended up doing a bit of testing re buffer and processor capacity size and what my MBP could take (not interested in monster track counts) and using a ball park mix of audio, non Logic Plugs and Virtual Instruments)

Went to 96khz, buffer 256 (8.8ms delay) buffer 128 (6.1ms delay) buffer 64 (4.8ms delay) buffer 32 (4.1ms delay)

At no point did the system (M3Max) get to 50%. That was on the main CPU gauge in the Control Bar and subsequently the Performance Meter when I figured out how to open it.

I reset Logic before the test so no clever moves re processor were going on (I did enable Complete Features)

To the untrained Coal Mine Eye the latency difference is nothing (roughly 4ms - 9ms) it's a no brainer tbf.

After all this I realized that the MBP was recording all the audio and running on its internal drive (two of the VI,s were running off an external fast drive)

Have I totally underestimated the power of my MBP or got this all wrong.

 

[Markdvc]

FWIW I've been recording projects at double sample rates for over 20 years. Back then using Logic 8 on a G5 with rotating hard drives.

regards

Mark

 

[elvisfridge]

I get that but I could never get near those lower buffer sizes and latency figures years ago.

I underestimated the Silicon processor for sure but couldn't find much on buffer / latency regarding model / processor / setting / delay in real world scenario (if there are reams of stuff on this please point me at it, I would like to see where others are at)

I thought it might be of interest to someone, if its common knowledge to everyone and / or in the wrong place please feel free to remove it.

Thanks for chipping in and please direct me to anything along the lines I originally posted.

 

[elvisfridge]

How is Tahoe going? Currently on Sonoma 14.8.1 seems reasonable. Upgrade option on my MBP is straight to Tahoe. Not sure if all my plug ins and software are there yet, interested to know how it's going for music people who have taken the leap. I am sure Logic will be fine but not sure that 3rd party developers will be. Any constructive info would be fab.

 

[paulnajar]

All things being equal doubling your sample rate or even 2.1x in your case should deliver about half the round trip latency at the same buffer setting. This is obvious because a single sample at 2x rate represents half the amount of milliseconds of time.

But the other important factor is the quality of your interface’s driver. How low a buffer setting you can go and still get work done depends on the cpu overhead the interface driver imposes on your system. Ultimately the point at where you start to get clicks and pops at low buffer settings is due mostly to this driver quality. In this respect there are worlds of difference between most interface driver brands (poor to average performance) and a really good interface driver brand.

At the top of the good end are brands like RME, Lynx, Audient (sometimes) maybe others as well but this list is short. I’ve been an RME boi for more than 20 years. Most of my projects start and finish at 96Khz. On my M1 ultra I can record and mix all day with hefty track counts, virtual instruments and audio processing plugins in use at a 64 sample buffer giving me a RTL of about 2.5ms with my UFX 3.

I do use amongst others a lot of UAD-2 plugs and they are worth a mention not just because they offload their processing to other processors but also internally even when working at 1x sample rates they upsample for better processing quality. When I first went from 1x to 2x rates I expected their processor usage to double but in reality the usage only increased by around 10 - 15% because of this internal upsampling. Other native plugins my use a similar scheme so you may do better going from 1x to 2x than basic maths would have you believe.

If you want to crawl down the wormhole of interface driver performance the most exhaustive database of work on this I’m aware of is the Low latency performance database thread over at gearspace. While the testing is windows centric there is a LOT of Mac based data intermingled in the many pages of the thread. Be prepared to spend days trolling through to find the stuff you’re looking for.

Kind regards