Apple introduces M2 Ultra

[Jim Roseberry]

M1 Pro
Cinebench R23

• multi-core = 12,359
  single-core = 1,532

M2 Pro
Cinebench R23

• multi-core = 11,791
  single-core = 1,654

13900k
Cinebench R23

• multi-core = 41,039
  single-core = 2,274

Loaded Kontakt (96k using a 64-sample buffer size) and opened The Grandeur (piano library).
M.2 SSD (3500MB/Sec) was used for streaming samples (Mac and PC).
Played heavily sustained fast runs up/down the keyboard (voice count ~100).

M2 Pro
CPU use ~25% and always glitches (audible pop) when releasing the sustain pedal.

13900k
CPU use ~8% with zero glitches. You can get your Rachmaninoff fix... with zero worries about glitches.

This is one of the first audio specific tests that I do... as it's quick/easy and gives an idea of how much disk-streaming polyphony is possible (under demanding circumstances).

I had planned to do significantly more testing... but there's no point.

On the Windows side, performance gain from adding cores doesn't scale 1:1.
IOW, Doubling the number of CPU cores doesn't double the processing power.
Extrapolate and say that Apple has done an amazing job and that doubling the number of cores with M2 Ultra doubles the M2 Pro's performance. No doubt the Studio model has more robust cooling and that would also help.
Double the M2 Pro's performance would put the M2 Ultra in the realm of the 12700k.
Considering the tight enclosures, that's really not bad.
If Apple wants the M2 to compete head-to-head with the 13900k and 7950x, they're going to have to crank the clock-speed and TDP (just like Intel and AMD).
My guess is Apple won't do this... because that's not their target market (which is a lot larger than those chasing the performance dragon).

[sqigls]

which audio interface?

desktop cpu vs laptop... what kind of cooling and power consumption do you need to maintain that CPU performance?

[Jim Roseberry]

which audio interface?

desktop cpu vs laptop... what kind of cooling and power consumption do you need to maintain that CPU performance?

Audio interface was a Presonus Quantum (Thunderbolt).
At 96k, Round-Trip Latency with 64-sample buffer size is 2ms.

M1 Pro was MacBook Pro
M2 Pro was Mini
13900k was full tower (no thermal-throttle using top notch large AIO)

Obviously TDP on the 13900k is **significantly** higher.
It's not Threadripper high (where you've also got active-cooler motherboard chipsets), but it's far higher than M1 or M2 (laptop or "slab").
Mac Studio is ~double the volume of the Mini... so I'd expect more robust cooling... and resultant performance increase.

[Jim Roseberry]

Mac or PC; I do not expect Laptops or Mini form-factor machines to compete head-to-head with performance of a full-tower.
It's unrealistic given the laws of Thermodynamics.

[sqigls]

exciting times though
for computer audio

[revvy]

Mac threads on KVR now read like PC threads.

Which motherboard should I use in my Mac? Any suggestions for a good thermal paste? Why is the case of my Mac so f**king ugly? Can someone suggest a good Mac anti-virus?

[PAK]

say that Apple has done an amazing job and that doubling the number of cores with M2 Ultra doubles the M2 Pro's performance.

The base M2 Pro is 6 performance cores. Ultra has 16. So Ultra's more than double the cores. Memory bandwidth is also quadruple the Pro (800GB/S Vs 200GB/s). For apps, like Kontakt, that further impacts performance beyond core count.

Double the M2 Pro's performance would put the M2 Ultra in the realm of the 12700k. Considering the tight enclosures, that's really not bad

This isn't what DAWBench's tests showed. Their testing showed the M1 Ultra beat the 12900k even with their more general "DSP INT" test. However it loses to both 13900k and 7950X, otherwise, except in Kontakt.

If Apple wants the M2 to compete head-to-head with the 13900k and 7950x, they're going to have to crank the clock-speed and TDP (just like Intel and AMD).

Outside of the GPU, almost all of the M2's performance gains (Vs M1) are already through a clock speed boost. It's a fairly unimpressive upgrade, and any Apple user (who is not hurting to upgrade) should wait for M3.

M3 will be the first "true" update to the M1. It will also be a 3nm part. An increase to 24 (or more) performance cores is not unlikely with the size reduction. Provided your chosen host can handle all those cores well (not an insignificant factor), it should thus see a significant boost even without clock increases.

To some degree I don't get a sense Apple are too bothered by what Intel or AMD does. They appear more concerned with keeping "creative types" on board. I'm not sure they'd even bother with a desktop product otherwise

[2DaT]

Loaded Kontakt (96k using a 64-sample buffer size) and opened The Grandeur (piano library).
M.2 SSD (3500MB/Sec) was used for streaming samples (Mac and PC).
Played heavily sustained fast runs up/down the keyboard (voice count ~100).

M2 Pro
CPU use ~25% and always glitches (audible pop) when releasing the sustain pedal.

13900k
CPU use ~8% with zero glitches. You can get your Rachmaninoff fix... with zero worries about glitches.

Is this emulated (Rosetta) or native? For a fair comparison it should be native, because emulation has a heavy overhead.

[machinesworking]

Audio interface was a Presonus Quantum (Thunderbolt).
At 96k, Round-Trip Latency with 64-sample buffer size is 2ms.

M1 Pro was MacBook Pro
M2 Pro was Mini
13900k was full tower (no thermal-throttle using top notch large AIO)

Obviously TDP on the 13900k is **significantly** higher.
It's not Threadripper high (where you've also got active-cooler motherboard chipsets), but it's far higher than M1 or M2 (laptop or "slab").
Mac Studio is ~double the volume of the Mini... so I'd expect more robust cooling... and resultant performance increase.

A quick google of current online results of the M2 Ultra Studio Cinebench is posted at CPU Monkey at 28967.
This is more on line with what I would think in terms of Apples thermal achievements VS Intels top dog. The other elephant, both machines you posted are half the memory bandwidth of the Studio Max and 1/4 the M2 Ultra. 200GB/s VS 400 and 800GB/s.

[Jim Roseberry]

A quick google of current online results of the M2 Ultra Studio Cinebench is posted at CPU Monkey at 28967.
This is more on line with what I would think in terms of Apples thermal achievements VS Intels top dog. The other elephant, both machines you posted are half the memory bandwidth of the Studio Max and 1/4 the M2 Ultra. 200GB/s VS 400 and 800GB/s.

Yep, Apple's soon to be released top CPU (June) will be in-line with Intel's top 12th Gen CPUs.
I built a mini-ITX 12900ks machine (Meshlicious case) that scores 29k on Cinebench multi-core. It's larger than a Mac Studio... but can house 3 internal NVMe drives (7000+MB/Sec) and a RTX-4000 video card.

Video rendering makes very efficient use of multiple cores and RAM.
ie: Running a DAW stress-test, you'll never achieve 100% load across all cores.
You'll hear audio glitches well before that (Mac or PC).

I will agree that those scores are impressive... for the given TDP.
I maintain that to compete on highest level, Apple will have to move away from small form-factor. Laws of physics

It would certainly be nice if AMD and Intel would get TDP lower.
That would allow more flexibility in cooling and case choices.
Both the 7950x and 13900k are a "go big or go home" scenario.

[pierb]

I maintain that to compete on highest level, Apple will have to move away from small form-factor. Laws of physics

Apple still hasn't released their top desktop chip. My guess, it's going to be released with the M3 in a proper Mac Pro.

I think the latest Mac Pro with the M2 Ultra is more of a devkit so that manufacturers of PCIe devices can start testing their products and drivers. In terms of performance, there's no difference from the Mac Studio.

[machinesworking]

I maintain that to compete on highest level, Apple will have to move away from small form-factor. Laws of physics

Apple still hasn't released their top desktop chip. My guess, it's going to be released with the M3 in a proper Mac Pro.

I think the latest Mac Pro with the M2 Ultra is more of a devkit so that manufacturers of PCIe devices can start testing their products and drivers. In terms of performance, there's no difference from the Mac Studio.

IMO who cares? chasing the fastest computer out there is always a zero sum game, less than a year later it's no longer the fastest. I doubt Apple ever will compete directly for this crown, as they never really have. When they used Intel chips they weren't aggressively releasing the fastest chips every three months and people were complaining. You would get Windows users chiming in on threads like this about how the latest x machine is faster than the Mac Pro, that Windows on Mac had 2-5% increase in performance on cross platform applications. etc. etc.

Apple has always cared more about stability in their hardware, despite being able to find youtube clickbait to the contrary, they spend a lot of money on R&D specifically to eliminate as much as possible hardware malfunctions, to stop returns etc. So as shown, a Mac Studio is competing decently against the best that Intel has to offer, at a much lower thermal output.

Again, the claims made are not there, no one said "Apple released the fastest chip ever made", yet here we are, with the typical KVR straw man shit show every time Apple announces anything really.

[Jac459]

[quote="Jim Roseberry" post_id=8692803 time=1686833315
It would certainly be nice if AMD and Intel would get TDP lower.
That would allow more flexibility in cooling and case choices.
Both the 7950x and 13900k are a "go big or go home" scenario.
[/quote]

Today in my day job, I was attending a laidback workshop, my high-end i7 asus on my lap when I felt a burn on my legs. I was using onenote... wtf....
TDP matters....

[jamcat]

It would certainly be nice if AMD and Intel would get TDP lower.

But then they wouldn’t be able to compete against Apple Silicon for speed.

[pierb]

IMO who cares? chasing the fastest computer out there is always a zero sum game, less than a year later it's no longer the fastest. I doubt Apple ever will compete directly for this crown, as they never really have. When they used Intel chips they weren't aggressively releasing the fastest chips every three months and people were complaining. You would get Windows users chiming in on threads like this about how the latest x machine is faster than the Mac Pro, that Windows on Mac had 2-5% increase in performance on cross platform applications. etc. etc.

Apple has always cared more about stability in their hardware, despite being able to find youtube clickbait to the contrary, they spend a lot of money on R&D specifically to eliminate as much as possible hardware malfunctions, to stop returns etc. So as shown, a Mac Studio is competing decently against the best that Intel has to offer, at a much lower thermal output.

Again, the claims made are not there, no one said "Apple released the fastest chip ever made", yet here we are, with the typical KVR straw man shit show every time Apple announces anything really.

I kinda agree but my point is a) if you're going to compare a high end desktop CPU it shouldn't be vs a laptop CPU and b) Apple still hasn't released a high end desktop CPU to be able to do an even remotely fair comparison.