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.
M3 Chip Generation - Discussion Megathread
- Thread starter scottrichardson
- Start date
- Sort by reaction score
Just making a note for prosperity - so far we're seeing a range of:
SC min to max: 2543 - 2999
MC min to max: 6626 - 7779
on https://browser.geekbench.com/search?page=1&q=iPhone16,1
Quite a range of GB scores.
SC min to max: 2543 - 2999
MC min to max: 6626 - 7779
on https://browser.geekbench.com/search?page=1&q=iPhone16,1
Quite a range of GB scores.
A common theory is that benchmarking a new phone soon after acquiring it often has many background installation and indexing tasks running that can lower the scores. There isn't much a user can do to boost their iOS Gb score, but quite a few things can interfere and lower a score.
The 2543 is curious because it scores extremely low in text processing.
| Single-Core Score | 2960 | 2543 | 116.4% |
|---|---|---|---|
| File Compression | 3004 | 2810 | 106.9% |
| Navigation | 2777 | 2786 | 99.7% |
| HTML5 Browser | 2906 | 2952 | 98.4% |
| PDF Renderer | 3122 | 3069 | 101.7% |
| Photo Library | 2698 | 2644 | 102.0% |
| Clang | 3601 | 3596 | 100.1% |
| Text Processing | 2727 | 463 | 589.0% |
| Asset Compression | 2761 | 2688 | 102.7% |
iPhone16,1 vs iPhone16,1 - Geekbench
browser.geekbench.com
Minimum scores have no meaning. If you run enough other stuff, the minimum score of every chip is zero.
So we can round the results to:
SC = 3000
MC = 7777
Benchmarks are a sore point with me and I strongly distrust them. I think most processors (and many other things too) are designed and optimized to make benchmark results look better and are not reflective of actual performance gains user experience. As a result benchmarks are typically biased to a particular architecture or manufacturer.
I am interested in basic chip characteristics (clock speed, cache size, codec encoders/decoders) but forget benchmark results. And I’m interested in other features to: battery size, interfaces, size , weight, cost.
They also didn't with the A16 and the A15...
They've clearly concluded there is no upside in that sort of talk. The tech-savvy people will immediately track GB6 numbers, while too many non-tech-savvy people will treat any numerical claims as eventual grounds for a lawsuit.
In vague, non-actionable terms.
They throw out a number every year, but the exact number changes every year (against a 3yr old Apple chip, against "competitors", about the GPU or about the ANE, etc) but they're clearly avoiding anything especially specific.
3,000 score is wild. Absolutely wild. It'd be the second fastest CPU in ST in the world after the 13900KS.
On top of that, it is scored from a mobile phone.
If the 3,000 score holds, I will have to apologize to Apple. Or they will have to apologize for sandbagging with the 10% claim. It's more like 15 - 19%.
3,000 score would translate to about 3,400 for M3.
3,000 score would translate to about 3,400 for M3.
Of course it holds. You can’t really overclock an iPhone. At the same time, less than 5% of scores are that high. Most appear to be around 2900. The 3000 seems to be a peak score only attainable in ideal conditions.
Edit: this is what I mean. These are all currently published GB6 scores, with lower peaks removed. Anything above 2950 looks like an outlier. We will probably see median score improving a bit as time goes by, but I doubt it will surpass 2950.
Last edited:
Correction, it could be around 3,688 for M3.
A15 --> M2 Max is 23.5% increase in ST.
If the 23.5% holds true for A17 Pro to M3 Max, then M3 Max would score 3,688.
That their nearest competitor can't beat a three year old iPhone processor is pretty specific information and way more telling than some random number like 3000 g-b-six benches. Nobody knows what a benchmark score is supposed to mean, unless you compare it to the result of the same benchmark running on another device. And then you still need to translate the numbers into percentages and visually represent them in a bar graph.
Granted, they don't show how unimpressive a 10% year-on-year increase looks like. But marketing is the communication of reasons to buy. And it's not necessarily always directed at people, who already own last year's iPhone Pro. Maybe they wanted to convey this information to Android users? That's perfectly legitimate and in no way misleading or vague.
We will probably see median score improving a bit as time goes by, but I doubt it will surpass 2950.
Yay, I am medianicity!
I'm not the only one with a similar thought. Jon is former Nuvia, and now ARM @ Google.
I no longer think A17 Pro is a die-shrink. But it does seem like A17's speed improvements aren't as good as expected. It "feels" like the major upgrade will come with the N3E SoC and the A17 was slightly rushed due to N3B being an emergency node for Apple.
If M3 comes out using N3E, then it could actually have a different architecture like what Jon is suggesting. If it's using tech meant for the A18, then it could be drastically faster than expected. On the other hand, if Apple plays it safe and ports A16 to N3E, then it'll be a disappointment. Last option is that M3 will simply use A17 Pro and N3B.
Note that N3E ramps in second half of 2023, which is basically now and M3 is rumored to come out in Spring 2024, which lines up.
To summarize:
Best case: M3 is N3E with A18 features
Middle case: M3 is N3B with A17 Pro features
Worst case: M3 is N3E with A16 features
Last edited:
M3 will be N3B, because N3E is 100% a no no
So the safest bet will remain N3B but...there are chances for the A16 based...hopefully not
So the safest bet will remain N3B but...there are chances for the A16 based...hopefully not
I hope so. Threads like this one would be more interesting and educational if Apple had different cores for iPhones and computers.
Apple plans to release iPhones for another 100 years. They are fully content with 10% y/y single core improvement. The Intel-M1 performance jump was an anomaly and we shouldn’t see it as a normal occurrence. All these people on Twitter who expect the next years iPhone to be the “real” upgrade is setting themselves up for disappoint.
And on those rare occasions when such huge jumps occur it is typically associated with some fundamental architectural change and then those same people complain about software compatibility. People just like to complain .
I think the improvements are reasonable given the already high IPC of Apple Silicon. We get around 5% boost in IPC, not excellent, but certainly not nothing. In certain workloads the IPC improvement approaches 10%.
I also doubt that there is anything rushed here, they had three years to iterate on a new u-arch. So I am quite certain that we are seeing the new u-arch. Next iteration will probably be in two years or more.
But it is a die shrink to 3 nanometers. If reality doesn't match with your expectations, check your assumptions! N3E promised an 18% speed improvement at the same power draw OR a 32% power reduction at the same speed over N5. Not both and not on a different chip with a lot more features. Naturally Apple can't achieve the full 18% when they want to save energy for other new tasks.
Let's say TSMC told Apple in 2017 that the real N3 node will be ready by 2023 and Apple is given rough design guidelines so they can start A17 design. Suppose that in 2019, TSMC says the real N3 node won't be ready, but instead, they will release a stop-gap node called N3B in 2023. Apple is given brand-new specs.
Apple just lost 2 years of design and planning.
Just all speculation of course. But this could have happened and we have an A17 that was designed with less time than a typical generation.