Apple M2 Studio stupidly powerful, M3 will be ridiculous

[Theo.M]

Bit of a story here, but I'll put a TLDR; in the final paragraph for those who don't want to read it.

Hi all, my wonderful and kind sister bought this for me on her interest free a month back (with her daughter's 10% educational discount too, yipee, and yes, Apple said it was OK for her to buy it and instantly transfer it to me, it was all above board), so I have 2 years to pay it off at a bit under 200 AUD a month, and I never, ever thought I'd be back on Mac when I started using Windows 5 years back to game. Long story short, I fell ill, sold my iMac pro and some other hardware, was bed bound, bought a windows gaming laptop and a Microsoft Xbox controller to game as it really helped pass the time when I was stuck in bed, then I upgraded it to a new one 3 and a half years later to keep up with the latest games. I am not greedy, I just want 1440P/high/60 FPS. When I started feeling a little better about 2 months back, well enough to at least start music again, I was very unhappy with Windows 11 and the gaming laptop performance.

It was a weird experience, as initially when I had the iMac Pro, Windows 10 via bootcamp was crushing the performance of OSX (then Mojave) on the very same iMac Pro! Pro Tools for example would get double the plug in instances on Windows on the iMac Pro vs OSX.

The reason was that Windows would constantly keep all iMac Pro cores at their maximum turbo speed, whereas OSX just wouldn't, and there was a ton of discussion about it on various audio boards at the time. Pair that with Core Audio's low buffer size issues, and Windows just performed way better.

I was convinced that Apple Silicon would be the same old story, and I was also going via cinebench scores, as PC Audio builders say that CB is about the closest we can get to measure daw performance/plugin scaling.
My laptop is around the same cinbench as the M2, pretty much dang on actually, but the performance differences are just ridiculous (favouring the Mac this time round).

My laptop does not have DPC latency issues - I have tuned it via bios to disable speedstep and C states, and it's an unlocked processor and I have all cores manually set to 5ghz, no thermal throttle, and no DPC spikes. The average is between 50 and 100 microseconds which is actually really good for a laptop.

I am able to run it at 8 buffer size on my Arturia Minifuse 2 no issue at all, the very lowest buffer, without pops and clicks. This is the sort of scenario where DPC issues would cause audible problems in the audio stream, and I am happy to report mine is fine.

But when I test, I do it all fairly, same buffer size and plugins, etc, and same DAW if possible. For example, PT is available both on PC and Mac.

Anyway, I really was wrong, so wrong. I was absolutely sure in my head that Core Audio would still have low buffer issues and performance would be compromised vs PC.

All I can tell you is this. IF you are a Mac user or interested in them and want to know whether to go Apple Silicon, it's a resounding yes.
It could not be more different to previous Intel Macs. People were trying to drum it into my head how good these were for audio and to be honest I truly thought it was just Apple fanboys spouting a bunch of BS. I didn't believe any Apple Silicon lover and what they were claiming.
I firmly believe that if Apple were still using Intel, the roles would be reversed and Windows would be performing better.

I am kind of stunned and had to re test and re test as I could not believe my results.
I mean just think this - my laptop is running 13th Gen intel cores at 5ghz and the Mac is 3.5ghz Peak P cores and 2.4ghz E cores and is crushing it. Even the PC E cores are at 4.2ghz (slightly overclocked), a higher clock speed than the Mac P cores!

So mhz for mhz, X86 just is not even close to Apple's take on ARM and I did not believe it till I used it myself. Seriously.

I want to give you an example to show you how huge the difference is.
On the PC, in Pro Tools or Reason 12 latest, the MOST IK Tape Model 80 I can get is 20 instances, and 18 in Reason at 256 buffer (remember, unlike Pro Tools, Cubase etc, Reason runs at the live audio buffer at all times, whereas *almost* all other DAWs have a hybrid buffer with something like 1024 samples on non record armed tracks, or a playback buffer).
Now you will see many videos on YouTube talking about the IK tapes and how great they sound but that they are so ridiculously intensive that even though they are emulating multitrack tape, one can't use them as multitrack in any realistic project scenario.

So what do you think I got on the M2 in PT?

80 stereo instances, one per stereo audio track, and there was still room to spare but I actually said "well OK, this is stable as anything and not one dropout, but I can "feel" the GUI starting to bog a little now, so no need to push it more". I am sure I could have squeezed 5 to 10 more instances IMO.
The thing is, Pro tools and Logic do NOT process plugins on E cores, and Cubase and Reaper do, so I am predicting 100+ instances in those DAWs. ALL cores are being used on the Windows laptop at all times by all DAWs tested.

I mean this is not just some small difference, this is monumental.

I also found out, even though 99% of the plugins I use *are* Apple silicon native in 2024, Logic in Native mode can actually still load the intel plugin holdouts via a bridge in the same session. I don't know of any other DAW that can do this, you need to run the actual DAW in Rosetta to use Intel plugins on Apple Silicon. So if you are a Logic user and worried about upgrading, don't be, the performance is mammoth, just run Logic in native mode.

The M3 chips coming to the Mac studio are going to have another 20% higher clock speed and 30 percent more Performance cores, so I can only imagine what insane beasts they will be.

So If you want to use IK Tape in realtime, for say an average size 32 track project and with one on every track, and still be able to have CPU headroom for VIs and other FX, you CAN on Mac, and you can NOT on Windows, even on the most powerful 14th gen desktops!

I know this is only one example but this is the most power hungry single effect plugin on the market, and to experience this performance was crazy.

I very much look forward to installing Reason on the Mac and doing all sorts of DSP tests soon.

BTW, for this testing both PC and Mac main buffer were at 128, but the playback buffer in PT is 1024, same as Logic's default setting and Cubase Asio Guard.

TLDR;
I am experiencing at least 4x the CPU plugin performance with what I have tested thus far on an M2 Studio VS a 13th gen PC laptop. Both computers have near identical Cinebench CPU scores and core counts, and the cross platform DAW I have tested in thus far is Pro Tools. Reason tests are coming soon. The M2 runs at a lower clock speed, is completely silent, yet is crushing the PC in audio plugin performance. One prominent example is IK tape, 80 vs 20 stereo instances!

[EdGrip]

I keep a little list on my notes app of which VSTs I have and like and use, so that when I replace my computer - or it suddenly dies - I don’t have to try and remember them all (good practice in general - M4L devices too).

Given the like likelihood my next computer will be an Apple silicon Mac, I put a little asterisk * next to every one of those VSTs that are Apple silicon native. At this point it’s either the whole list, or nearly the whole list.

[Chi-Individual]

I came across this vid a while back cause I was interested in how well M3 handled audio since I have an M1 Pro. His results were interesting. Someone smarter than me would need to tell me if his testing methods are valid.

[PhillipOrdonez]

He is not wrong, chi individual, because the issue is the number and type of cores available. Whereas before the pro chips were add good as the max sans the GPU, they used to be the winning chip for audio and a Mac mini would be sufficient for most audio tasks. But apple changed strategy and slowly reduced the number of performance cores on the later generations, meaning you either get the max or you miss out, and would have been better off with a previous chip than the current pro chip. Max and ultra have always been better but they are making the difference even larger with every generation of chips. You know what I mean?


I've been following the m saga with desires to buy into apple for the first time, and have been pretty set on the studio as the best option for future proofing. However I'm not sure how going to wait for the prices on the m3 studio to think about it.

[EdGrip]

It’s mainly about the extra ports, for me. I could outsource them to a hub, but by the time I’ve spec’d a Mac Mini up, I might as well have bought the base level Studio.

[Theo.M]

My worry was single core performance. As we all know, your DAW can only be as fast as your fastest core, cause once one core is overloaded, you can have 20 cores free and it means nothing.

It could be a multithreading issue in Windows actually. What I need to do is see how many I can get on a SINGLE track on Mac., On Reason with the windows PC it was 5 instances stable on a single track i.e. a single core, and I presumed therefore with all those other cores free that one on each track would spread through the cores one by one till it got to core 1 again and then the same process till full.
But plugins just don't seem to be multithreading their instances all that well on Windows for me. That said, after a nap, I'll check the Mac on a single core and compare it to Windows.

See I thought the low clock speed would really hamper the Mac native silicon performance, but it just doesn't. It's so powerful for the speed and wattage and heat it puts out.

I have a feeling the coming M3 may throttle or be louder to compensate, cause the Mac Studio M2 can run silently with all cores 100% loaded (Cinebench 10 minute run), @ about 90c degrees. Higher M3 clock speeds and more P cores at those clock speeds might overpower the cooling system as Apple won't change it at this stage (and it's great for what it is, and IS dead silent with the M2), so my suggestion is, when the M3 Studio gets announced, see if you can get an M2 at a blowout price as they just work as they are and match perfectly with the silent cooling. It's so quiet it's almost silly, I don't even know it's on, honestly.

The Mini gets noticeably hotter, has 2 less USB C/TBolt ports and just isn't worth the small saving once you equal the Ram and CPU core specs (and still you get less GPU cores). That's the other thing. 6 USB C and 2 USB A on such a tiny thing is pretty crazy in itself, and the thunderbolt buses are one bus per port, so the bandwidth is huge. In Intel models, the iMac Pro over 4 Tbolt ports for example was 2 busses, where as the Studio is 4 or 6 individual busses. Pretty crazy indeed!

Also my old UAD works FINE over thunderbolt 2 with an apple adaptor, the same adapter I had for the iMac pro all those years ago. I haven't even turned the thing on in 5 years so was glad it still worked (phew). I could not use my UAD on my Windows, as thunderbolt 2 devices do not work with thunderbolt 4 ports on windows even with the adaptor, as they do on Mac. I actually thought my device had died till I found the info out that it can't be done. Intel forced this thunderbolt 2 compatibility out of TBolt 4 ports purposely on the Windows side. You need a thunderbolt 3 native port or an interface with thunderbolt 3 itself or higher for a modern Windows PC.
Apple have done some sorcery with thunderbolt that even thunderbolt 1 devices will work on the latest Apple Silicon Macs (or any Intel thunderbolt Mac).

[Theo.M]

I came across this vid a while back cause I was interested in how well M3 handled audio since I have an M1 Pro. His results were interesting. Someone smarter than me would need to tell me if his testing methods are valid.

I'll watch it now and share my thoughts. Cheers.
I do know that currently the higher core M3 chips are for laptop only, and they get loud and throttle as I have seen on Max Tech and other sites, so, the M3 won't perform to its full capacity in a Macbook Air or 14". The best bet if you want an M3 right now is the 16" Macbook, as it has dual fans.

Edit, Oh Ok I have seen that video. It's not a big deal because I use app tamer to relegate ALL tasks besides Pro Tools to E cores, so maybe that's why it's so stable, the P cores are literally completely dedicated to the DAW.

But yes, you need Reaper or Cubendo at this stage if you want to use all the E cores in the DAW also, but remember, the M3 will have more E cores. For example, the M2 Ultra has 16 P and 8 E, and M3 Ultra will have 24 P @ 4ghz. As long as it's able to be cooled sufficiently, it's going to be another massive performance leap for EVERY daw due to so many P cores.

[Theo.M]

He is not wrong, chi individual, because the issue is the number and type of cores available. Whereas before the pro chips were add good as the max sans the GPU, they used to be the winning chip for audio and a Mac mini would be sufficient for most audio tasks. But apple changed strategy and slowly reduced the number of performance cores on the later generations, meaning you either get the max or you miss out, and would have been better off with a previous chip than the current pro chip. Max and ultra have always been better but they are making the difference even larger with every generation of chips. You know what I mean?


I've been following the m saga with desires to buy into apple for the first time, and have been pretty set on the studio as the best option for future proofing. However I'm not sure how going to wait for the prices on the m3 studio to think about it.

The studio is the best Mac on the market, period. The Mac Pro is a complete waste of money unless you have a real need for lots of internal pcie slots, and there aren't any extra native drive slots which is crazy. It's not worth it. Mini gets hot.
The best Macs right now are the 16" M3 Max or Studio M2 Max or Ultra. That's a fact (imo).

[selig]

My 2022 Studio M1 Max is already more power than I've ever needed (so far)!

[Chi-Individual]

Since I’m basically an ITB Reason user my 14” MPB M1 Pro is still super overkill. But I like to keep my eye on these things just in case RS shocks us on the next update and does something truly unexpected DAW-wise. Though I’m not holding my breath on that

[Theo.M]

Please if this topic interests you at all, do read this post as it's all about single core performance discovery and thread scaling.

So, it gets more interesting. Of course I am streamlining to one plug in at present, but those who know me already know I am a dsp test addict so I will be broadly expanding all of this as time goes on

I just tested Pro Tools on OSX vs Reason on Windows, however - I set Reason to 1024 buffer to be fair as Pro Tools internal hybrid playback buffer is 1024 samples.

A single track (i.e. core) of the M2 Mac can get 6 IK Tape @ 96K, and 9 @44.1K in PT 2024.3 or Logic 10.8.1

Pro Tools and Logic (unlike I think all other DAWs), will not play compromised audio at all, they will trigger an overload message and will not allow a single pop or click. S1, Reaper, Reason, Live, Bitwig, anything else I have tested, will still play even if the audio sounds like it's going through a Cheese greater which itself is going through a 4 bit reduction effect. They just don't care.
Therefore I had to be really careful in listening to Reason, but I think the results will interest you.

Reason @ 96K, 6 instances (yep, identical), 7th adds very noticeable pops and click
Reason @ 44.1K, 9 instances. Now, 8 was stable as, and when initially adding the 9th, it popped and clicked till the device was in place i.e. just for a couple seconds, then pops and clicks disappeared and was stable.
10th was pop and click haven.

So this is REALLY interesting to me.

Indeed, the M2 is still impressive, as its single core DAW performance @3.49 ghz is matching an X86 13th gen @5.21ghz (since only a single core was being used, it went into the higher turbo mode).

And at about a quarter power usage. So of course, this is VERY impressive to say the least.

BUT, as I suspected, it's not some miracle computer and so immensely more uber powerful than a Windows PC.

What is happening here is better multithreading on the Mac DAWs, and I am sure that has something to do with the OS's own multithreaded scheduling, not just the DAW code.

Core for Core they are neck and neck, and now the near identical Cinebench single core results make sense. An app like Cinebench which properly uses all cores does match the Mac. And more up the PC scale, a14900K will obliterate an M2 Ultra in Cinebench (~40K vs ~28K, CB R23) yet it's not translating to DAW performance whatsoever right now.

Remember, I have tried this with almost every well known DAW on Windows. I downloaded the Ableton 12 30 day demo, DP11 30 day demo, Bitwig 30 day demo, Reaper demo, Nuendo 13 60 day demo, S1 6.5 30 day demo and more! Bar Nuendo and Reaper, these have all expired now unfortunately, so I can't test further in them. I have access to unrestricted R12 and PT2024 right now and of course the very generous Reaper demo, so that's what I'll be testing on.
On Mac it's Logic and PT, I have NOT installed Reason yet. I'll get to it, but each day is a very slow burn for me and I get just a little by little done as my health slowly improves (or tanks for a day, c'est la vie).

Now a very, very respected DAW machine builder was telling me that the latest version of Cubase had major multithreading issues on Windows with high core counts, but it seems ALL Daws currently do as of writing this post and on Win 11 build 23H2. But I do know this person has been with Cubendo from the early days and has always done performance testing with it. I believe he uses Reaper now also.

Look at the math - 9 instances on a single core @44K on Mac is translating to 80 instances over 16 P cores (E cores not being used at all).
On Windows, 9 instances on a single core @44K is translating to 20 total instances across 24 cores/32 threads in Nuendo 13. See how it's not properly scaling even though Windows Resource monitor is showing all cores used?
So there is an issue with Windows multicore scheduling concerning real time audio apps specifically, or it's a problem the DAW developers need to improve on themselves, I don't know.

So there's the issue. They'd basically be neck & neck if PC side sorted out its multithreading.

Also, I have to test the Mac at the lowest possible buffer, as the PC is stable @ a buffer of 8 and a total roundtrip of 3ms from the in put of the interface (USB) through the DAW and to the output (all USB interfaces have safety buffers, the 3ms is real world figure, well, 3.11 to be exact) and can be loaded to about 50% or so DSP usage on the DAW meter before any cracks and pops occur. Remember, a DAW usage meter is the Asio time on the PC side and Core Audio on the Mac side and not real CPU usage, but nevertheless DAW usage meters are great as they are relative and accurate for that DAW.

So tomorrow when the others that live here are awake, I will test @ whatever the lowest buffer is on Mac, then match that same buffer to Windows, as traditionally, Mac was always *awful* at low buffers in OSX. The entire reason Avid, Steinberg etc created their hybrid buffers was because the performance was so poor otherwise on Mac, seriously. This is true.

Has Apple's native silicon, with its new memory architecture and powerful watt for watt processing mitigated the issue? We will see!

I am absolutely stunned just how poorly the PC is scaling, so I am also going to do tests with 8 E Cores disabled and hyperthreading on the P cores disabled, i.e. just 16 "real" cores instead and see if that's actually BETTER! We shall see.

PS in 5 days I won't be able to test IK tape anymore as the demo runs out. They will still work but with hideous hissing, amplified across instances, so if there's something in particular you want me to test with them before the 5 days is up please let me know.

[Theo.M]

It never ceases to amaze me how stuff gets unravelled by just discussing about it. I would have never thought to try disabling cores till I wrote this topic.

I just disabled hyperthreading and 8 E cores in bios, so now the OS sees 16 cores rather than 32, and I am getting identical instances of the plugin.
Identical.
Maxing out at the same amount of 20.

Work that out!

So this is entirely a high core count scaling issues with Windows and Daws. Whatever the reason, Mac is not suffering from that issue. All that remains to be tested is Mac native Apple silicon at very low buffer sizes with tracks record armed. We shall see!

[PhillipOrdonez]

Reason ONLY utilizes performance cores on Mac. Maybe it use the same story on windows?
Reaper does utilize all cores.

This in case you didn't know. Keep on testing.

[Theo.M]

Reason ONLY utilizes performance cores on Mac. Maybe it use the same story on windows?
Reaper does utilize all cores.

This in case you didn't know. Keep on testing.

Nope, only 16 P cores on Mac are being used vs 32 on Windows. Mac E cores aren't being used at all, so 8 cores free, yet all PC cores being used, which makes it even worse. Reaper maxes out at 22 instances on Windows but the interface is very sluggish. 20 instances like the other DAWs to be fully responsive.

I have tried about 14 Daws for 30 days before the demos expired, as I explained, but reason, PT, Nuendo and reaper are still installed and all get 1/4 of the instances of the Mac. The issue is windows and DAWs and high core counts. Mac is just crushing it in this regard. The issue is scaling.

Windows resource monitor shows all 32 threads being hammered across 20 instances, yet I get 20 instances also across 16 threads as mentioned and it shows the same hammering.

So there's something going on with real time audio and scaling across cores in windows. Cinebench probably scales so well as it's not a real time audio process. I am convinced it has something to do with asio or Windows.

Actually I am going to try Windows WASAPI and see if it's any different. That would be interesting.

BTW this isn't just me with these plugins, there are SO many vids out there with people saying to only use on busses as it's impossible to multi track them. Not on Apple silicon Mac it's not!

[Theo.M]

MME made no difference. I used the onboard realtek audio via MME in case the arturia minifuse asio was the issue, I got one less instance with MME.

[PhillipOrdonez]

Interesting.

[Przemyslaw]

It's interesting what you write about scaling.
In the context of what you wrote, I will write about my recent experiences. Until recently, I was working on Windows 10 Intel 7940x (14 + hyperthreading), system optimized 2 years ago, maximum settings, Reason using 14 cores or more. I work on large projects, so I started thinking about changing the processor to a newer one. However, a month ago my system disk died, I installed everything from scratch, also upgraded to Win11, and while trying to optimize the system, it turned out that a new guy appeared on YouTube and talked about optimizing Windows for DAW. Following his advice, I'm currently working in Reason with 6 cores, 128 samples in full projects with mastering done, and I'm not thinking about changing the processor for now.

In my country, the calculation regarding Mac vs. PC is very simple, I build the PC myself and expand it myself as I want, when I want and with what I want, according to current prices, Mac is about 10-20x more expensive than PC, is it 10-20x better? ...

[Theo.M]

It's interesting what you write about scaling.
In the context of what you wrote, I will write about my recent experiences. Until recently, I was working on Windows 10 Intel 7940x (14 + hyperthreading), system optimized 2 years ago, maximum settings, Reason using 14 cores or more. I work on large projects, so I started thinking about changing the processor to a newer one. However, a month ago my system disk died, I installed everything from scratch, also upgraded to Win11, and while trying to optimize the system, it turned out that a new guy appeared on YouTube and talked about optimizing Windows for DAW. Following his advice, I'm currently working in Reason with 6 cores, 128 samples in full projects with mastering done, and I'm not thinking about changing the processor for now.

In my country, the calculation regarding Mac vs. PC is very simple, I build the PC myself and expand it myself as I want, when I want and with what I want, according to current prices, Mac is about 10-20x more expensive than PC, is it 10-20x better? ...

Of course it all depends on usage, needs, even preferences.

I am talking purely about plug in power and so far testing a very small subset of plugins,. I didn't mean to infer that Mac was better or anything like that.

Of course even a 4 core 5 year old PC can make great music. We have bounce in place, and the music is up to the skills of the composer, not the PC.

Cheers!

[Przemyslaw]

What I meant was that the key is to properly optimize Windows for DAW, thanks to which the system properly addresses power and speed to drivers and music programs. Maybe in Mac it is the default, but in Windows you need to discover its hidden potential.

[dvdrtldg]

I haven't done any scientific testing, all I know is that pre-Apple silicon I would regularly hit the CPU ceiling, and post-silicon (currently M2) it's never happened. Interesting you should mention IK Multimedia's Tape Machine 80, because it's one that I use a lot. As I say, the CPU overload is a problem that's simply vanished

[tiker01]

One issue is that M3 has less performance cores than M2. This is a bit problematic for audio processing. I am waiting with my M1 Pro probably till M4.

[Theo.M]

One issue is that M3 has less performance cores than M2. This is a bit problematic for audio processing. I am waiting with my M1 Pro probably till M4.

The M3 Ultra will be all P cores though and no interconnect die, the rumours are 24 P cores @4ghz and MAYBE a "super" version with 24+24 for 48 cores.

Anyway, what is happening here is Windows DAWs are not scaling across cores. Raw power core for core, the M2 @3.5ghz seems to be on par with a 13980HZ @5ghz, so it's more efficient, but the PC still plenty powerful.

I am trying to explain the math to people that 9 on ONE track which means on one core, then 22 on 24 cores doesn't make sense. No we never get linear core scaling but it shouldn't be anywhere near that bad either.
7950X desktop user at GS is getting 26 instances. A 7950X. Now it's not just the tape 80, this user is able to play projects that are hammering his 7950X on an M1 8 core. No jokes. Yes it's completely maxed out as well but do you see the difference? In real world he's not getting better plugin performance on a desktop 7950x than he is on a small macbook M1. It doesn't make sense and it is involving core scaling, and Cubase is not the only culprit.

[Heater]

I have an MacBook Air M1 with only 8Gb of memory and have never ran out of performance. It’s a beast.

[PhillipOrdonez]

And that's the binned pro chip

[moalla]

Still ordered a Ryzen 7840hs machine with 32gb ddr6400 and 2tb ssd, for the price of slower
8gb 512gb MacBook Air 1350€
My new notebook is on par with the m2. Pro chip what I’ve read. I was hard thinking to buy a Mac, but the delicatessen prices for upgrades are cheekiness….2tb ssd upgrade 920€