Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
3nm?
- Thread starter gojilla
- Start date
- Sort by reaction score
No, it's more logical to use A17. The primary reason is that A16 was designed for 5nm from the start and A17 should be designed for 3nm from the start. It takes years to design and validate a chip. M3 will use 3nm - not 5nm. So it makes more sense that A17 will be used for M3.
My opinion is that it was a failure because of how late it was. It was ~8 months late. If it launched on time, it would have been fine but by the time it launched, its single-thread performance was lept by Zen4 and Alder Lake.
So pretty good then? Since the M2 isn’t much of a failure. My M2 is 21% faster in multi core and 13% faster in single than my M1 in Geekbench 6. Compute is even better at a 43% improvement in the GB6 Metal test.
Didn’t the PugetSound Adobe benchmark show a 40+% improvement of the M2 over the M1?
Don't use benchmark use real world test. For export videos barely any difference
I keep looking for that "Real World Benchmark" but can't find it anywhere. 
In my corner of the real world, I like to see what Lloyd Chambers (Mac Performance Guide) and Art Suwansang (Art is Right on YouTube) have to say. They are both tech savvy photographers who do a lot of testing on Adobe software, and related programs like Zerene Stacker, that I use.
Lloyd performed a number of tests on the M1 Max in early 2022 and recently posted results on the 16" MacBook Pro M2 Max. The overall results for the M2 were impressive, but with the tests separated by about a year, it was hard to say how much of any improvement over the M1 was related to gains in Apple Silicon as opposed to software improvements over the past year.
Lloyd's conclusion (based on a raw SHA-512 hashing speed test) was that the M2 Max has a 7% performance improvement through 8 CPU cores that extends to 23% in the "rare" case where a job can use all cores.
"Clearly the 2022 MacBook Pro M2 Max is a better performer than the 2021 MacBook Pro M1 Max, but only marginally so. Other new features aside, a 2021 MacBook Pro M1 Max discounted by 15% offers better price-performance value."
Art's recent video compares the M1 Max and M2 Max and his results were a real mixed-bag. Some tests showed significant improvement for the M2 (Lightroom Classic 1:1 previews, 38% better and Lightroom CC Export, 15%). Some improvements were modest (Lightroom Classic Export, 3.5% better). Some tests were a push (Final Cut Pro and Lightroom Classic HDR Merge). And some (Lightroom Classic Panorama Merge and the Photoshop "huge" benchmark) showed a significant advantage for the M1 Max. Why? Art's M2 had 32GB of RAM, while his M1 had 64GB and the M2 bogged down on these memory intensive tests.
From my vantage point, the M2 looks like a nice, but modest upgrade. That said, what I really like about it is the ability to get it with 96GB of RAM. The Adobe products I use can be real memory pigs, so having the extra headroom is attractive.
In my corner of the real world, I like to see what Lloyd Chambers (Mac Performance Guide) and Art Suwansang (Art is Right on YouTube) have to say. They are both tech savvy photographers who do a lot of testing on Adobe software, and related programs like Zerene Stacker, that I use.
Lloyd performed a number of tests on the M1 Max in early 2022 and recently posted results on the 16" MacBook Pro M2 Max. The overall results for the M2 were impressive, but with the tests separated by about a year, it was hard to say how much of any improvement over the M1 was related to gains in Apple Silicon as opposed to software improvements over the past year.
Lloyd's conclusion (based on a raw SHA-512 hashing speed test) was that the M2 Max has a 7% performance improvement through 8 CPU cores that extends to 23% in the "rare" case where a job can use all cores.
"Clearly the 2022 MacBook Pro M2 Max is a better performer than the 2021 MacBook Pro M1 Max, but only marginally so. Other new features aside, a 2021 MacBook Pro M1 Max discounted by 15% offers better price-performance value."
Macintosh Performance Guide: Articles & Reviews
Mac Performance Guide: choosing, configuring, how-to, upgrading, performance of memory, hard drives, RAID, solid state drives, especially for photographers
macperformanceguide.com
Art's recent video compares the M1 Max and M2 Max and his results were a real mixed-bag. Some tests showed significant improvement for the M2 (Lightroom Classic 1:1 previews, 38% better and Lightroom CC Export, 15%). Some improvements were modest (Lightroom Classic Export, 3.5% better). Some tests were a push (Final Cut Pro and Lightroom Classic HDR Merge). And some (Lightroom Classic Panorama Merge and the Photoshop "huge" benchmark) showed a significant advantage for the M1 Max. Why? Art's M2 had 32GB of RAM, while his M1 had 64GB and the M2 bogged down on these memory intensive tests.
From my vantage point, the M2 looks like a nice, but modest upgrade. That said, what I really like about it is the ability to get it with 96GB of RAM. The Adobe products I use can be real memory pigs, so having the extra headroom is attractive.
See, for real meaning improvements from M1 Max wait for 3nm tech and a brand new CPU and GPU design.
Given that he was comparing two dissimilar machines (32GB RAM vs 64GB RAM), you can not discount RAM as a factor in some of the performance discrepancies. That's admitted to with respect to the latter two tests, yet the overall comparison between the two seems to overlook that discrepancy in favor of a value judgment (which was probably predetermined prior to the testing). To be a fair and equitable comparison, both machines would need to have the same amount of RAM and likely the same amount of storage to be as comparable as possible.
Yeah, I wish Art had compared two 64GB RAM machines, but his comparisons are still helpful.
Interestingly, Lloyd's comparison wasn't apples-to-apples either. In his, the M2 Max has the advantage of 96GB Ram compared to 64GB on the M1.
Lloyd is a big proponent of maxing out RAM on his Macs. After using some of the same software he uses on some really large stacks and merges, I can see why. That's why, if I was in the market for a Max computer, I'd be inclined to get the 96GB M2 even though the overall upgrade from the M1 seems modest.
Of course, YMMV.
China Times reports that Apple will use TSMC N3E, the second generation of N3, for A17 and M3.
www.chinatimes.com
領先群雄 蘋果跨入3奈米應用 - 財經要聞
手機市場今年市況最快要到下半年才有望開始復甦，使手機晶片大廠普遍將重兵押注在2024年，其中聯發科、高通（Qualcomm）皆在今年下半年推出的新品採用台積電4奈米製程，唯獨蘋果（Apple）單獨跨入3奈米世代，並在今
China Times reports that Apple will use TSMC N3E, the second generation of N3, for A17 and M3.
領先群雄 蘋果跨入3奈米應用 - 財經要聞
手機市場今年市況最快要到下半年才有望開始復甦，使手機晶片大廠普遍將重兵押注在2024年，其中聯發科、高通（Qualcomm）皆在今年下半年推出的新品採用台積電4奈米製程，唯獨蘋果（Apple）單獨跨入3奈米世代，並在今
So what about the N3B wafers that have been in production for the last three months...?
Still hoping for M3 Ultra / Extreme / Whatever based ASi Mac Pros...
Yes, that's what I've been wondering. Where did all those wafers go? It's actually 4 months of wafers: https://www.macrumors.com/2022/12/29/tsmc-3nm-mass-production-begins/
China Times reports that Apple will use TSMC N3E, the second generation of N3, for A17 and M3.
領先群雄 蘋果跨入3奈米應用 - 財經要聞
手機市場今年市況最快要到下半年才有望開始復甦，使手機晶片大廠普遍將重兵押注在2024年，其中聯發科、高通（Qualcomm）皆在今年下半年推出的新品採用台積電4奈米製程，唯獨蘋果（Apple）單獨跨入3奈米世代，並在今
Same China Times that said that a iMac with a dGPU was coming real soon now over two years ago in the M1 era?
Report: Apple Silicon iMac With Custom GPU to Launch in Second Half of 2021
Apple will debut a powerful custom-designed graphics procesing unit in its first Arm-based iMac which is set to launch in the second half of next...
www.macrumors.com
I wouldn't 'bet the farm' on them being highly accurate on what is going on inside of TSMC on Apple projects.
Not that China Times has a good accuracy track record , but the reports of N3B in very high volume production are a bit thinly sourced too. The Digitimes leak a while back said that TSMC was only going to get to 50% utilization by end of March. ( Going to get to just half of the production capacity sitting idle as opposed to more previously. )
So making something that is closer to 0.5M (or lower) than 10M around M1 sized chips per month.
It would be a better match to the lower production capacity being used early on in the first 1-3 months.
Also something for the AR/VR headset. (again relatively low run rates versus plain Mn sized die ). [ reportedly there are two SoC chips in the headset. One dedicated to sensor, screens 'control', and associated inferencing, while the other to run apps and networking. No real good reason to 'go cheap' on the custom sensor fusion chip as it likely will/could be a huge product differentiator (if done right). ]