Hi,
I just got an M2 Studio Ultra last week, 64GB RAM, 1TB SSD. For many mostly happy, stable years I have been running a 2009 8-core cheesegrater with an HDX card and a UAD Octo card, still running PT 2021.6. Despite the HDX card, the vast majority of my processing is from the CPU, or the Octo card. But I have been kind of amazed by how many Native AAX plugins I can run before I get errors. But I wanted to upgrade my plugins and especially Kontakt/Komplete for some upcoming projects - VI's definitely slow down my Intel 5,1 system - so thought it was time to get on the Silicon ARM chip bandwagon.
I have just started setting up the M2 Studio, installing PT 2023.12.1, installing plugins. The HDX and Octo cards are still in the old computer. Based on everything I'd been reading over the last few years I expected the M2 to BLAZE! It doesn't seem exponentially faster to me. Admittedly, I am so far only running a tiny, un-scientific test Pro Tools session. And I have to run in Rosetta Mode. But I'm starting to get a little nervous, wondering what I am getting for my MANY thousands of dollars in upgrade costs. (Haven't bought the Sonnett expansion chassis yet, thankfully) Just now I took a look at System Usage on the M2 for the first time and I was disappointed to see I'm hovering around 17% total usage on a VERY small session - couple Boom plugs, couple Xpand2, BlueCat Patchwork, VPROM LM-1 emulation, one Altiverb and one Lexicon Native Reverb. One audio track. A few other plugins, nothing intensive. 17% seems high for that. So I exported a copy of this PT Session and loaded it up on my old 5,1 rig. The System Usage looked the way I expected it to look on the M2, barely tickling any of the CPU cores, and 2% total usage. Does this make sense?? Am I missing something? I've attached screenshots of the system usage windows. Thanks for any thoughts!!

I'm likely to transition from Windows to OSX either in 2024 or 2025, and I've been slowly preparing for that transition by making sure that before doing so everything I use has an ARM-native version. I suspect the performance issues are largely attributable to running PT in Rosetta mode.

Actually, these are better examples. A moderator on the DUC suggested I run the Avid Pro Tools Demo Song session as a proper test, so I downloaded it. I ran it on the Cheesegrater and the M2, in ARM and Rosetta mode. There was no appreciable difference. CPU usage was higher in Rosetta, but both of them, ARM and Rosetta, were WAY less efficient than the old Cheesegrater. The issue is not simply a question of Rosetta vs. ARM. It's something going on with the M2 itself. Here are the screenshots from running the demo song:

Are you running Sonoma? Pro tools isn't supported on Sonoma yet. Could that be part of the reason? What happens if you open the logic demo project on each?


My M2 Pro runs circles around my 2016 MBP. Different comparison though obviously.

Edit: Pro tools known issues with Sonoma: https://avidtech.my.salesforce-sites...a-Known-Issues

looking at system usage is the wrong way to test your system...

I'm running Ventura. I'm open to the idea that System Usage isn't the best way to gauge performance. So then - what is...??
As a side note, the very first time I pressed record on the new M2 system was on that tiny session - the screenshots from my first post - on a MIDI track - 64 Buffer @ 96K. I instantly got a CPU error. Probably due to the one Lexicon Native Reverb in there. But still, disappointing.

Over 2000 tracks with Kontakt on an Ultra with Logic.

https://m.youtube.com/watch?v=3CBIdw0L7mM&t=2728s

Same process with other DAWs resulted in half the count it seems.

That box is FAST.

So I suspect there is some fundamental issue with something in the software you use.

I ran some M2 tests yesterday (thoroughly unscientific), and only with Pro Tools, but thought I'd share some results.
I did two main tests yesterday.

TEST #1 :
New 44.1K / 24-Bit PT Session:
- Loaded a mono audio track w/ FF Pro-Q EQ, FF Pro-G Gate, FF Pro-L2 (**16X Oversample**), Eleven MkII, MDWEQ6
- Master Fader = EQ3-7, ChannelStrip, ProComp, ProLimiter, MDWEQ6, F670, EQP-1A, MEQ-5, Maxim, POWr Dither

Then I tried to see how many times I could duplicate the audio track before the session crapped out with CPU overload:
1. 2009 CHEESEGRATER 8-Core = 22 Tracks [Buffer = 512]
2. M2 STUDIO - ROSETTA MODE = 64 Tracks [Buffer = 64]
3. M2 STUDIO - ARM MODE = 110 Tracks [Buffer = 64]
4. M2 STUDIO - ARM MODE = 122 Tracks [Buffer = 512]
5. [Buffer setting seemed to make no difference in M2 ROSETTA Max Track Count?]

TEST #2 :
Tried to see how many times I could duplicate ALL the tracks and folders in the Avid "Low Road" Demo Song (except the final Mix Aux and Master fader)
1. 2009 CHEESEGRATER 8-Core = 7X @ 512 Buffer (266 Voices)
2. M2 STUDIO - ROSETTA MODE = 8X @ 64 Buffer (304 Voices)
3. M2 STUDIO - ARM MODE = 23X @ 64 Buffer (874 Voices)

No idea if these results seem as they should be? Good? Bad? Does it seem my M2 / Pro Tools setup is operating as it should? I really wish more plugins were ARM compatible because Rosetta is clearly a poor second. Curious that my 15-year old Cheesegrater could run 7X instances of the full demo song, and M2 Rosetta could only run one more at 8X! I have System Usage screenshots if anyone is interested.

Don't see what's curious about that - it's a miracle it's faster. If you think about what went on behind the scenes to achieve that backwards compatibility, it's amazing.

You need to ensure everything is running in ARM. Rosetta is incredibly fast for emulation BTW - it’s an amazing feat of engineering that it’s as fast as it is.

But it’s not meant to remove the need for proper ARM binaries.

Now set the cheesegrater to 64 buffer and see how fast the ultra really is in comparison