Apple M2, M2 Pro, and M2 Max SoCs

A rumor (Google Translate) from the Commercial Times provides information on the next two generations of Apple SoCs.

Code:

SoC            M2         
	
M2 Pro        
	
M2 Max 
Codename       Staten     Rhodes        Rhodes 
Process        4 nm       4 nm          4 nm 
Release date   2022 H2    2023 H1       2023 H1 
Products: 
Notebook       
	
MacBook    
	
MacBook Pro   MacBook Pro 
All-in-One     
	
iMac       iMac Pro      iMac Pro 
Desktop        
	
Mac mini   
	
Mac Pro       
	
Mac Pro 

According to this rumor, the M3 generation will be released "after an 18-month cycle," so presumably sometime in 2024 depending on the chip. The rumored Jade 2C-Die and Jade 4C-Die for the ARM Mac Pro, which are widely speculated to consist of multiple M1 Max dies, are not mentioned here.

 

Going to be interesting if the M2 debuts at the March 8th event called "Peek Performance".

 

I am expecting hardware accelerated Raytracing.

 

Why do you expect hardware accelerated ray tracing?

 

Nobody expects the hardware accelerated ray tracing!
Sorry, i had to.

 

I am expecting hardware accelerated Raytracing.

 

Double confirmed!

 

Eventually, they'll have to follow the market stadards. But they may not be in a rush. But even if they ignore gaming altogether, creative and media production software will eventually adopt HW RT acceleration, and it just won't make sense for apple not to offer it.

But apple being apple, they might wait untill that is an indispensable necessity to adopt it, and will turn that into something to brag about. The older models becoming obsolete is, for them, a bonus. Buy the new one.

 

It's "Peak". Peak is in 'summit' or 'top', rather than peek as in 'glimpse'. Apple's event titles are already tricksy enough!

 

https://wccftech.com/apple-unveils-...k-gpu-on-par-with-nvidia-rtx-3090-less-power/

 

Those performance statements tell me next to nothing at all...

• Prerelease Compressor 4.6.1 tested using a three-minute clip with 5K Apple ProRes RAW media, at 5760x3240 resolution and 24 frames per second, transcoded to Apple ProRes 422.
  
• Performance measured using select industry‑standard benchmarks.
  
• Performance was measured using select industry‑standard benchmarks.
  
• Prerelease Final Cut Pro 10.6.2 was tested using a one-minute picture-in-picture project with 18 streams of Apple ProRes 422 video at 8192x4320 resolution and 30 frames per second, as well as a one-minute picture-in-picture project with nine streams of Apple ProRes 422 video at 8192x4320 resolution and 30 frames per second.
  
• Power was measured using a representative workload in a commercial application.

 

I would guess that the two "industry-standard benchmarks" are Cinebench and Blender.

 

Here is some Geekbench scores, if you care about shitty benchmarks. 1793 Single-Core Score / 24055 Multi-Core Score for the M1 Ultra. That's 16 high-performance cores and 4 high-efficiency cores. 800GB/s memory bandwidth.

 

The big issue going forward for M1 Max/Ultra and the M2 gen of chips is that the software support still isn't there for high performance 3D graphics in macOS.

Apple needs to pull its finger out and just adopt Vulkan. They're never going to have the influence to get devs to port their games to Metal or Metal2.

With the release of the Steam Deck, there is no better time to get serious about supporting Vulkan in macOS.

 

Someone help me out if I'm understanding this wrong... Your description of the benchmarks seem to be FMV transcoding performance? How is this a reasonable way to to interpret "as fast as a 3090" when a 3090 really isn't the spec for something like this?

I mean, props to them for getting all the FMV raster work to a much lower power envelope. Still, IIRC the NV transcode hardware is supposed to be identical horizontally across same-gen GPUs right? So a 3050 has the same transcoder as a 3090?

 

Me either, but I guess these metrics are aimed at the people who use the applications for which the performance was cited.

Cross-platform (PC/Mac, Windows/macOS, x86/ARM) benchmarks are pretty useless anyway and when all three are used, i.e. trying to compare an AMD/Nvidia GPU running on Windows under x64 against an Apple GPU running under macOS on ARM, how would these ever be remotely equivalent?

 

Some things to consider. Was reading tomshardware forums and their quite much poking holes everywhere regarding these benchmark claims by Apple.
First of all, the Zen2 3990x scores north of 35.000 in Apple's own geekbench 5 app, the M1 Ultra aint even nowhere close. As many on the toms forums wonder is where Apple has these numbers from.
https://browser.geekbench.com/v5/cpu/search?utf8=✓&q=AMD+Ryzen+Threadripper+3990X+

Another is, Zen3 TR was officially announced yesterday/today depending in which timezone you live. The comparison to Alder Lake is kinda strange aswell, not just price but also in core-count. The M1 Ultra is in the server/redering station class, not consumer gaming/creator cpu class. Heck, the M1 Ultra doesnt even outperform the 3970X. Apple is comparing to old Zen2 architectures, in tests and benchmarks that favor Apple sillicon, like the media accelerators, more on that later.

And as many have pointed out there (and partially here) Geekbench 5 is kinda useless to compare different platforms using different architectures. I cant myself thinkoff any pc hardware review site using geekbench to gauge pc performance. To note aswell, geekbench was created by the owner of a Mac review site.

What another poster shared:

''I just checked out Geekbench, and a lot of the benchmarks actually do things that Apple probably does as part of the SoC instead of the main CPU, or has done main CPU optimization for iOS. Image compression, ML, PDF rendering, Camera processing, etc. It was hard to tell if this was using the main CPU, or other parts of the M1 SoC. Don't get me wrong: it's VERY clever of Apple of leveraging the SoC in a way that others can't.''

Forum topic

https://forums.tomshardware.com/thr...iX2XscUAvMUKjINmQ-AzFGjKAUsfj-3gcU3pKm_I5XzdY

Indeed, many others have said the same thing (forum link above). Apple could just have said its as fast as a 3050.... The 3080/3090 or equal AMD gpu will be far, far ahead in just about everything, aside from some specific Apple media acceleration stuff, which these other GPUs also have aswell in their own forms.
Oh and again price ofcourse. youre looking at around 5k if you skimp abit on memory. For the 128gb version with 1tb (lol) youre going to be looking at very high prices. Here in Europe were looking at almost 6k for the entry Ultra studio mac. Almost 10k for the complete experience (128gb/800gbs/64core soc). For the half the price a X86 system will decimate that in just about everything, oh and you can game on it aswell and you have ray tracing and other such important features that these days not only matter in gaming. In actual gaming and other non Apple tasks, i'd be impressed if the maxed Ultra would be close to a RTX2060 (dgpu). 3090 has close to 1TB/s for its memory alone.

Some speculate that the Mac Pro would double the performance of the Studio, and that would compete with the Threadrippers most likely, like the 3995wx and Zen3 variant and whatever Intel has to offer in that space. Imagine the price of the upcoming Mac Pro.

 

LTT video on the matter

One of the posts on how Apple compares in their ads to intel HW.

''jadoei13
3:18 So what they seem to be saying is that if you set both the M1 Ultra and a 12900K to 60w, it is 90% faster. That's not very difficult given how far the 12900k is away from it's maximum power consumption and thus clockspeed. It would be interesting to see a comparison of the 12900K and 5950X in the usual benchmark suite though.''
210

 

First this is not an Apple app.

Second those 35k scores obviously are overcloked results; a result closer to reality is ~25k as shown on Geekbench processor summary page.

So what?

That guy is full of shit. And it's not that hard to prove it: compare a random Arm result against an Apple one, for instance this. Do you see any subtest going 5 to 10 times faster (whcih what you'd see if there was HW acceleration)? The highest speedup is on HTML5 (about 2x); do you think they have an HTML accelerator?

Frankly get over it, M1 is a great CPU, no need to look for cheating. I guess my rant is useless, you'll keep on posting such stuff, but I'll keep on answering