H/W Buffer setting not really affecting CPU Load?

[LDS]

Originally Posted by tmuzik

Dual buffer? I did not know - I will need to investigate that. I have only ever had one buffer setting. Makes total sense, to separate playback/record.

What is odd is I can now record tons of inputs at 64 buffer @88.1 - I made a video showing how changing my buffer from 2048 to 64 does not change the cpu load what-so-ever, by the end I was recording 32 mono + 32 stereo w/ 64 sample buffer - it is crazy.

https://youtu.be/rUMzFY0AHz4

I also opened an old 44.1 session and I now have the option to go as low as 32 samples, I did not have that option in the past. I had always assumed it was because of my old audio interface and it's drivers.

I don't think my aging system is capable of that low of a buffer with that many tracks and plugins.

You will only ever have one buffer setting that you can change. That is the recording buffer. Pro Tools handles the playback buffer automatically, under the hood. Playback tracks are just audio files sitting on a hard drive, so Pro Tools will automatically start them playing early, applying a longer sample buffer to them, giving the CPU more time to process them.

Obviously the same doesn't apply to record channels where the audio is a live, realtime input into the DAW so you need to be able to adjust how big or little that sample buffer is. As long as both the record and playback channels hit the DAW output at the right time, you and I and anyone else listening won't notice the magic that Pro Tools is weaving under the hood.

I do think there has been substantial improvements with the way the dual sample buffer works in Pro Tools since they introduced the hybrid engine. The hybrid engine needs to divide the record channels from the playback channels, then put the former on hardware DSP and keep the latter on CPU. Seems like an ideal opportunity to also improve the way native systems handle recording and playback channels too.

[audiolex1]

Originally Posted by tmuzik

Dual buffer? I did not know - I will need to investigate that. I have only ever had one buffer setting. Makes total sense, to separate playback/record.

What is odd is I can now record tons of inputs at 64 buffer @88.1 - I made a video showing how changing my buffer from 2048 to 64 does not change the cpu load what-so-ever, by the end I was recording 32 mono + 32 stereo w/ 64 sample buffer - it is crazy.

https://youtu.be/rUMzFY0AHz4

I also opened an old 44.1 session and I now have the option to go as low as 32 samples, I did not have that option in the past. I had always assumed it was because of my old audio interface and it's drivers.

I don't think my aging system is capable of that low of a buffer with that many tracks and plugins.

Well, that is the thing. The higher the sample rate the lower the buffer goes.

If you have something that shows the In and out latency, not pro tools, you will see that you can get better buffers at higher sample rates.

Back in 2014 I was recording with 88.2 and the buffer size was better and I can't remember why exactly.
That is native not using HDX.

I literally had to DuckDuckgo it (no Google here)

https://sonalsart.com/does-higher-sa...rease-latency/

There is a cost to CPU usage as it will increase as well as hard drive space, both which I don't think matter as much anymore.

Whether you think high sample rates sound better or not, mathematically they should. If you record at 88.2, 44.1 is exactly half.
I've noticed differences in sounds working in 88.2 with plugins and synths for what I am hearing.

For me, the latency benefit and I'm hearing more detail with the plugins then at 44.1.
There harmonics and aliasing that behave differently, the nyquist limit. I haven't thought about all of that in a long while.

I'm sure I'll be trolled for just posting that there are benefits above 44.1. I know a well known engineer/mixer who recorded a d Punk album at 88.2 and mixed it at that.

Play around with it. It might surprise you more than confuse you.

[LDS]

Originally Posted by audiolex1

Well, that is the thing. The higher the sample rate the lower the buffer goes.

If you have something that shows the In and out latency, not pro tools, you will see that you can get better buffers at higher sample rates.

Back in 2014 I was recording with 88.2 and the buffer size was better and I can't remember why exactly.
That is native not using HDX.

The improvements in latency at higher sample rates are associated with hardware more than sample buffers. Things like internal processing in converters, etc where a task might take 7 samples to complete, so running higher sample rates push those 7 samples through in much shorter times.

When it comes to sample buffers, things are incredibly uniform within DAWs and audio interfaces. The actual sample buffer 'value' that you can choose in a DAW is a variable based on the chosen sample rate. What matters is the actual buffer duration/latency that the sample buffer represents.

[tmuzik]

Steering back on point - This thread has drifted into a buffer conversation. ¯\_(ツ)_/¯

The original issue still remains - AND it is bad for mixing, excellent for cutting multiple tracks.

H/W Buffer DOES NOT increase or decrease CPU load which it should do.

Real easy to test, verify or prove it is a "me" only problem, and I am just a crazy old coot.
To Recreate:

• Create a New session
• Create 128 Stereo Audio Tracks
• Set the H/W Buffer to something you can record with 32/64/128/256
• Go to Window > System Usage - to bring up CPU meter
• Add enough plugins to create a decent CPU load 40-70%
• Click Play - Notice the cpu load
• Now set the buffer - as high as possible - emulate mix - this should bring the CPU load down - it does not
• Drop the H/W buffer to 128 or 64
• Record arm 32 tracks - then 64 then 96 - then 128

I can tax my CPU to 45% then record arm 100+ stereo tracks and record with 20% CPU still remaining.
Which is fantastic if I had 100 preamps, mics and a reason to do that.
I would rather have the buffer go towards my processing power.

Can anybody at AVID explain what is going on?

[tmuzik]

Update from AVID support ticket.

Thank you for reporting this issue. The issue has now been reproduced with Engineering and logged as PT-279941, for your reference.

I hope they fix it soon.

Cheers

[Darryl Ramm]

Oh thanks for reporting this and the heads up here.

[StudioTalk]

Originally Posted by Ben Jenssen

Tracks record enabled or not, I've never seen, in over twenty years w/PT, that the playback buffer size has any fundamental influence on CPU usage. The only relationship I see between the two is that as you reduce the buffer size toward the limit of what the computer can handle, the cpu usage indicator bars will become more and more unstable, more spikes, and eventually produce errors. You don't get a more capable cpu if you increase the buffer size. That's not how it works.

Agreed, that said, I noticed this too when doing test for my channel. It's really weird, the meters don't seem to give correct info until you hit record. Weird.

[thin ice]

Originally Posted by StudioTalk

Agreed, that said, I noticed this too when doing test for my channel. It's really weird, the meters don't seem to give correct info until you hit record. Weird.

Surely this is because of dynamic plugin processing. Would it give the correct info if you press play instead of record?

[Darryl Ramm]

Playback in Pro Tools uses the large fixed size disk IO buffers, active inputs either through input monitoring or being in record use signal chains going through the user adjustable IO buffer. It seems counterintuitive why record would start showing CPU processing but input monitoring would not, or why dynamic plugin processing could have an effect here. Now if input monitoring was not enabled on all the same tracks then sure it’s simply dynamic plugin processing doing its thing. But sure try disabling and see the difference. And a reminder that Pro Tools CPU meters are already largely useless for predicting actual CPU errors.

[tmuzik]

Originally Posted by thin ice

Surely this is because of dynamic plugin processing. Would it give the correct info if you press play instead of record?

No, I tried dynamic plugin off, in fact I tried every relevant tick box, in all combinations. I also moved all plugins and let PT rebuild with stock.
PRESS Play question: The first half of video I posted is "play" oriented.
No cpu relief going from 64 to 2048 - I actually did the video backwards -
I loaded the system with 2048 buffer (0:30 mark) - switched to 64 buffer (1:40 mark) and there was no noticeable change on system load. A 2048 buffer with a 75% load should crash at 64k.

Buffer size does have a direct correlation to CPU load. The less time you give the processors to "process" the harder they have to work.