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.
Mark Gurman says the M2 chip is supposed to be a stopgap chip
- Thread starter Retskrad
- Start date
- Sort by reaction score
M2 Pro/Max rumours seem hot so if that happens I can't see M3 Pro/Max in Spring. Maybe M3 devices in Spring and M3 Pro/Max (maybe Ultra?) at WWDC. Then M4 in fall? Would put things back on track compared to the M1 launch at least.
I read somewhere (maybe here on Macrumors?) when the M1 chip was first released that they were aiming for an 18 month cycle on their chips. Since M1 was released in Nov-20 and M2 in June-22 that seems to make sense.
However this is nothing I can back up more than “I read somewhere” so I might be mistaken and it just happened to be delayed with 6 months from a 12 to 18 months cycle.
However this is nothing I can back up more than “I read somewhere” so I might be mistaken and it just happened to be delayed with 6 months from a 12 to 18 months cycle.
....as is my lot in life
.
They need to have someone who knows how to type the tickets into Jira.No, as long as they have this guy it's all good.
Apple Leadership - Johny Srouji
Johny Srouji is Apple’s senior vice president of Hardware Technologies and a member of Apple’s executive leadership team. Learn more about Johny.www.apple.com
Gurman was a treasure trove of information in this video
I agree M2 does sound like a stop gap, an M1.5 or M1S, based on fact it will have a shorter shelf life than M3 possibly will, and M1 had / and is having.
M1->M2 is extremely incremental, and who knows for Pro And Max M2 could be a big boost but maybe not too?
He doubled down on 14 and 14 Max with A15 and A16 only going to Pro's. that kinda sucks but whatever I wonder what iPhone 14 will have to offer over iPhone 13 - only colors? the 14 max is a new style of phone we haven't seen a light more budget friendly Max size yet.
--
I wish he spoke more about iPads. info seemed anemic there.
M2 iPad? OLED mini? upgraded 11" Pro screen? a huge 15" iPad? nada.
--
the 15" MBAir should be cool! looking forward to that. and the m3 12" ooo la la - but in 2024 -- ouchie!
I agree M2 does sound like a stop gap, an M1.5 or M1S, based on fact it will have a shorter shelf life than M3 possibly will, and M1 had / and is having.
M1->M2 is extremely incremental, and who knows for Pro And Max M2 could be a big boost but maybe not too?
He doubled down on 14 and 14 Max with A15 and A16 only going to Pro's. that kinda sucks but whatever I wonder what iPhone 14 will have to offer over iPhone 13 - only colors? the 14 max is a new style of phone we haven't seen a light more budget friendly Max size yet.
--
I wish he spoke more about iPads. info seemed anemic there.
M2 iPad? OLED mini? upgraded 11" Pro screen? a huge 15" iPad? nada.
--
the 15" MBAir should be cool! looking forward to that. and the m3 12" ooo la la - but in 2024 -- ouchie!
The rumor is that the M2 Pro/Max/Ultra will be built on TSMC's 3nm process. If that's true, then the M2 on the 5nm process would be a "stopgap."
TSMC had some difficulty with 4nm and 3nm. Not to the same scale as Intel's issue with 10nm, but enough to delay shipping chips in volume. Assuming Intel stays on course, it should get back to parity with TSMC by 2025 (and could be ahead if TSMC has any delays). I wonder if Apple would consider using Intel Foundries in the future (to produce Apple Silicon - Apple isn't going back to x86).
I highly highly doubt this rumor.
N3 requires new design rules. This means Apple would have to completely redesign the A15 for N3 just for Pro, Max, Ultra. What's the goal of completely redesigning an 18-month old SoC for a new node? It takes a ton of resources to design a chip on a cutting edge. Why do it with your lowest volume chips? Why not just wait a few more months and do it with the M3?
In my opinion, it's far more likely that M2 Pro/Max/Ultra will be on N5P and M3 will be on N3.
Even so, it's interesting that the A16 will most certainly be on a variation of N5 but its corresponding M series will be on 3nm. This means Apple had to design the A16 for both the N5 and N3 processes simultaneously. There's a non-zero chance that M3 will still be on a variation of N5.
Why do you assume that M2 Pro etc. will be based on A15? If the 3nm rumor is true then it’s almost guaranteed that they are based on A16.
If they're 3nm and A16, then they will call it M3. The performance will be significantly different than M2. It would make no sense to call a 3nm and A16 chip as M2.
Also, if the M2 Pro/Max comes out later this year as expected, then there's zero chance it will be 3nm.
There is absolutely nothing preventing them from naming A16-derived prosumer chips “M2 Pro/Max” and then the scaled down consumer chip “M3”. These are just marketing names after all. Even has the advantage that the prosumer chips are faster and more capable and helps them get the roadmap back on track.
Either way we will see soon enough. The only thing is clear - if the upcoming prosumer chips are still A15-based, Apple will get behind the competition in terms of performance. With Zen4 and Raptor Lake on the horizon, not to mention the new GPUs, A15 cores are not cutting it anymore in the prosumer segment.
There's nothing preventing Apple from doing anything. That doesn't mean it makes any sense.
Zen4 isn't expected to ship to laptops until Spring 2023 at the earliest. This is because mobile Zen is always 6 months behind desktop Zen launch.
Same for Raptor Lake. Desktop Raptor Lake will launch first.
By Spring 2023, M3 (based on A16) should be out and ready to compete.
Lastly, I highly doubt that Apple could change their schedules so fast based Zen4 and Raptor Lake release info. SoCs are planned years in advance.
Which is exactly my point. If this years prosumer chips are 3nm then there is no change of schedule. Maybe that was the plan all along. I think it would make sense to base the prosumer chips on a more advanced platform to increase their value proposition.
Of course, this is all conjecture and wishful thinking. I was too optimistic before hoping that M1 Pro would be more advanced than M1 and it wasn’t. I do hope for more substantial improvements in M2 prosumer chips but who knows.
I personally think that if Apple can get to a place where they are releasing a new M series once every year, there's nothing AMD and Intel can do that can catch Apple. That's it. Just update once a year. Piggy back on top of the A chip.
I mean, even Zen4 and Raptor Lake wouldn't touch the M1 in terms of efficiency. They might beat it in raw performance but it doesn't really matter that much in a laptop, which has far higher volume than desktops.
But Zen4 and Raptor Lake will likely go up against M3, not M2.
In multicore efficiency Zen3+ already gets fairly close, simply because AMD can throw more cores at the same problem and undervolt them in the process. Benchmarks show that M2 at 10W and 6800U at 12-13W get comparable scores in Cinebench R23. And sure, that's a test that maximally favours x86 and maximally disadvantages Apple, but still this suggests that the real-word efficiency difference at lower wattages is probably under 30% now. I am fairly confident that Zen4 with architectural improvements and 5nm node can reduce this to 10% (again, to be clear — I am talking about performance at low wattage). So yeah, Apple absolutely needs to continue improving their architecture to stay on top. I am curious to see how the future products will shape up. Definitely exiting things happening!
Last edited:
Is it still true? The changelog of the latest version of Embree says "Using 8-wide BVH and double pumped NEON instructions on Apple M1 gives 8% performance boost"
Release Embree v3.13.4 Release · RenderKit/embree
Using 8-wide BVH and double pumped NEON instructions on Apple M1 gives 8% performance boost. Fixed binning related crash in SAH BVH builder. Added EMBREE_TBB_COMPONENT cmake option to define the co...
github.com
cmaier claimed that all else equal, x86 had a 20% efficiency deficit to ARM because of all the legacy CISC instructions to support. That's consistent with what you observe with the M2 vs. 6800U. I think both are on TSMC's 5NP node.
Yes it's still true. That version of embrue isn't in Cinebench yet. It will probably have to wait until a new version of Cinebench comes out. Apparently they can't alter it too much mid-version otherwise it invalidates the other scores. I'm afraid we'll have to put up with the wholly unsuitable "benchmark" being used by more clueless testers.Is it still true? The changelog of the latest version of Embree says "Using 8-wide BVH and double pumped NEON instructions on Apple M1 gives 8% performance boost"
Release Embree v3.13.4 Release · RenderKit/embree
Using 8-wide BVH and double pumped NEON instructions on Apple M1 gives 8% performance boost. Fixed binning related crash in SAH BVH builder. Added EMBREE_TBB_COMPONENT cmake option to define the co...
Has Cinebench been updated to use the latest version of Embree? I doubt it will happen any time soon. But even if it gets updated, it’s still going to be a worst case benchmark for Apple. It still uses an AVX-to-NEON layer rather than hand optimized ARM SIMD, and it can maximally utilize the wide SIMD units of x86 CPUs as well as benefit from their faster clocks and SMT. This is the kind of workload where x86 can play all to of its advantages.
My intuition (I know, it’s not much) says that current-gen Apple Silicon would probably be 15-25% faster than now if embree were hand-optimized for ARM. But I wouldn’t be surprised if x86 CPUs (especially the high-end mobile and desktop) were still faster. As I said, massively parallel SIMD processing is not the primary strength of Apple Silicon. Nor is it the primary area of interest for most users. It’s quite funny that a benchmark that solely focuses on this one domain became the standard measure of CPU performance.
Zen3+ is N6, which is an improved N7 if I understand it correctly. I think a lot depends on which point on the efficiency curve we are comparing. The primary reason why 6800U can get closer in efficiency to M1/M2 at 10W is because it has 8 performance cores. So AMD can afford to aggressively downclock these cores more, losing out some of the per-core performance in return to a significant boost in efficiency. It we had a 8-core Zen3+ running against an 8 P-core M2 at 10 or 15 or 20W AMD would have no chance whatsoever.
But this again illustrates that performance does not exist in isolation. It all boils down to design tradeoffs. Apple has subjectively superior tech but it’s also more expensive. AMDs tech is not as good but it’s cheaper so they can afford to put more cores in an affordable package which allows them more flexibility in multicore efficiency.
The reality is that the M2 was always going to be an "incremental" update, simply because of the need to mass produce at scale for the MacBook Air and 5nm being incredibly stable. Apple will start moving to 3nm this year, but the first year of a new node, especially in this Chip-apacololypse means Apple needs to hedge, and use 3nm in products that aren't going to be as high in demand. iPhone Pro's will probably take up most of that capacity for the year. It will then probably move over to M2 Pro and M2 Max.
No inside information.
I'd have to bet that lots of money is being spent building out 3nm so they can continue to beat their competitors.
Their competitors will get access to 5nm relatively soonish.
No inside information.
I'd have to bet that lots of money is being spent building out 3nm so they can continue to beat their competitors.
Their competitors will get access to 5nm relatively soonish.
Ah friend, we meet again. I don't know if you remember us debating on here, but we typically were arguing the same side of the same argument.
I have heard it from many people that Microsofts insistency on keeping up support for legacy code basis is holding back CISCo/x86. In order for AMD to be competitive it would have to I'm speculating that AMD has to support those legacy items as well. But I could be proven wrong. I'm highly analytics, and read lots of chip white papers, but a chip designer I am not. I would almost think it would be better for Intel, AMD, and Microsoft would be better served running such legacy software with a software stacks that emulate legacy behavior there by freeing up the hardware stack, and still giving legacy app users still a better experience compared to running it on hardware natively. The thing that brings me to this conclusion is when the first M1 MacBook Pro came out, it was able to run just about everything X64 in Parallels in Windows better than many (if not most) laptops within a $500 price swing of the CISC/x86 market, in emulating, inside of virtualization. I've heard people say the M1 was built with Rosetta in mind, but I don't know if anyone knows of the sacrifice to the devil, or voo doo magic they did to do it.
The greatest benefit Apple has, is they are shameless to tell developers "read the tea leaves, if you don't modernize, you can't run on our platform".
The opposite can be said about Microsoft/Intel/ARM.
I believe there is a way for the big 3 to make it work, Apple did. But Apple was working on it in secret behind close doors, and it's been 18 months since people saw what Apple can do.
If the logic for legacy code was taken out of the hardware (largely) and used for something else that enhances efficiency, I think the big 3 would be better served as long as they worked with Microsoft on a software solution.
Just my 2 cents.
It's more than that. They are using the new generation of cores, and theres more of them. I don't expect there to be enough of anything in the M2 Max to make me want to buy it immediately though.