Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
2021 has not been a grea5 year for apple’s marketing team. Someone needs to be replace.

this is a huge performance increase year on year with zero mention in the video. Fail.
 
Why would that be?
I just said why in the same post. The way defects work is that they are randomly distributed across area, with clustering of certain types. The bigger the area of a structure the higher the likelihood of a defect in that structure.

If you believe apple is selling 4-core chips where the fifth core is broken, you have to believe that there is a big supply of chips where there happens to be a broken gpu core but nothing else on the chip is broken. That is highly unlikely. It will happen once in awhile, but a very small percentage of the time. In most cases, if there is a defect in a gpu core, there will be other defects in the chip - defects cluster. The chance that the only other defects also happen to be in that same gpu core, given the small size of the gpu core, is small.

Where this sort of binning happens is with much bigger structures, or with structures that have much more repetition - in other words, the chip has 256 MB of ram, but some blocks don’t work, so you sell it as 128 MB. Or in situations where you have 2 cores, each of which is roughly half the die area, so that clustered defects are likely to break one but not the other.

Here, each GPU is small, they are close together (so that a cluster is likely to break more than one), they share logic and metal (so certain types of defects in one will break more than one)… The supply of “golden 4-core” chips is just going to be way too small.
 
How does having 1 additional GPU core (+25%) result in +34% higher Metal score?
The math doesn’t make sense to me either. Perhaps there really are two variations of the processor, one having a different gpu design? Or those benchmarks are simply wrong/fake? Or the gpus are clocked faster on the pro? Or that one is running an unreleased iOS update that that has optimized metal for the new processor?

So many possibilities. 🤔😁

I would bet the pro has the gpus clocked faster along with the extra core.
 
How does having 1 additional GPU core (+25%) result in +34% higher Metal score?
Yeah this seems real odd to me, I don't think we saw a similar jump with the 7 core M1 -> 8 core M1 GPU. My guess is there is the core difference along with a clock speed increase...
 
Battery life gain is from bigger battery more so than the same 5nm node. Even TSMC 4nm in 2022 won't see much of a battery difference compared to bigger battery.
While you are right about the battery size being the biggest part of the explanation, video streaming increased 35% while the battery capacity increased 15% between iPhone 12 and 13 models.

Which leads me to think the processor also plays a key role.

The Pro model also has this variable refresh rate ProMotion display, which leads to even bigger gains. If they ever put this in Macs, I wouldn't be surprised if we got over 25hrs with some MBP models.
 
Last edited:
There is no way this is the next gen Macbook Pro CPU. It says the internal GPU is 5.2 tflops. The current top end GPU for 16" MBP is 5.3 tflops. Apple will never ever release something that's slower than the previous gen, especially after waiting 2 years.

I expect something close to 7-8 Gflops in the next 16" MBP.

Or this might be the base config which can be extended further.
I don't know (because no one outside Apple knows) but those benchmarks make sense to me. Tflops don't tell the whole story, as we've seen with the M1. In some tasks it exceeds discrete GPUs and in others it lags behind. If the next chip is an 'eXtension' of the M1 (AKA the A14) and not based on the M2 (AKA the A15), I think this is likely where it will scale. 5+ Tflops with tile-based rendering and unified memory is going to scream for a lot of workflows. In 2022 we're likely to see some class beating GPU performance as the A15 hints at a bigger leap in not just GPU cores but GPU interconnects.
 
  • Like
Reactions: anticipate
Not sure what you mean. Both iPhone 13 and 13 Pro have the same A15 Bionic chip. They are not different.

The Pro models have a 5-core GPU compared to 4-core in the non Pro's, which benchmarking reveals about a 25% increase in performance with the Pro
 
I don't know (because no one outside Apple knows) but those benchmarks make sense to me. Tflops don't tell the whole story, as we've seen with the M1. In some tasks it exceeds discrete GPUs and in others it lags behind. If the next chip is an 'eXtension' of the M1 (AKA the A14) and not based on the M2 (AKA the A15), I think this is likely where it will scale. 5+ Tflops with tile-based rendering and unified memory is going to scream for a lot of workflows. In 2022 we're likely to see some class beating GPU performance as the A15 hints at a bigger leap in not just GPU cores but GPU interconnects.
I do realize Tflops don't tell the whole story, but still Apple never released anything that's slower on paper than the previous generation. The Geekbench, or Metal Compute scores have always gone up.
 
That has nothing to do with profits. Pro model has 120Hz Promotion display and non-pro model does not in addition to pro model having extra camera features. That’s the reason for the pro model having one additional GPU core. There is no crippling or disabling.
I suspect Apple made a mass order of the exact same chip (5 core GPU). To maximize yields, any with defects due to 5nm process would then have a core disabled and put in the non-pro chip bin.
This is called chip binning. Allows Apple to mass order a single chip to minimize costs, and any chips with defects can have the GPU core disabled and go into the non-pro iPhones.
I highly doubt a 120hz display requires all that GPU power. However, the ProRes video might.
 
I suspect Apple made a mass order of the exact same chip (5 core GPU). To maximize yields, any with defects due to 5nm process would then have a core disabled and put in the non-pro chip bin.
This is called chip binning. Allows Apple to mass order a single chip to minimize costs, and any chips with defects can have the GPU core disabled and go into the non-pro iPhones.
I highly doubt a 120hz display requires all that GPU power. However, the ProRes video might.

As i‘ve explained elsewhere, highly unlikely that the difference between 5-core and 4-core chips is yield fallout. The GPUs are small, and the chance that a die happens to have one and only faulty GPU and that nothing else is broken due to defect clustering is vanishingly small. Almost certainly, the vast majority of the 4-core chips are simply chips with one core disabled intentionally.
 
Graphics: 4-core A15 15% faster than A14, 5-core A15 55% faster than A14

Single-core CPU: 10% faster than A14

Multi-core CPU: 18% faster than A14

While 5-core A15 is indeed impressive, I am not sure rest of the A15 numbers can be called "pretty dang impressive" over A14.
10% year on year is great. Some years have been Moore but that is still great.
 
  • Like
Reactions: SantaFeNM
Just imagine if they put M1 on this phone. Definitely will buy but I will pass the A15 bionic.
It's not hard to imagine what would happen if they put the M1 in the iPhone 13. It's essentially an A14 with more cores. Completely unnecessary to have 8 cores for the iPhone (not to mention the M1 operating in the iPhone's thermal constraints would throttle hard), so you'd have literally no performance benefits, terrible battery life, and a chip from last generation.

The M1 would be worse than the A15 for the iPhone in every single way.
 
  • Like
Reactions: kirk.vino
Not sure if anyone beat me to it but if you run the numbers the A15's single core performance improvement is scaling almost linearly with clock speed vs the A14 (2998Mhz A14 vs 3233Mhz on the A15), lending credence to the idea that the A15 is probably more like an A14+ (which is totally fine.) I'd imagine the increase in the multicore score is down to higher all core frequencies. (Source: https://www.anandtech.com/show/16226/apple-silicon-m1-a14-deep-dive/2)
Speaking of Anandtech can't wait for their annual deep dive!

GPU performance on the 5 core A15 variant is impressive, I'm surprised they didn't compare it to A14 directly... (although I guess that wouldn't meant admitting the 4 core A15 isn't all THAT much of an improvement)
It seems like most people here are missing the whole point. First of all yes the CPU cores themselves seem to be close to identical to the A14, but they are getting better throughput just because of the clock increase. Secondly, the multi core performance is significantly higher. This is probably a result of the doubling of the system cache. The GPU speed is a lot higher in the pro models especially to accommodate for Promotion but that raw power is also there. The Neural Engine is 40% faster which itself last year was 200% faster so in two years the Neural Engine has sped up over 280%. And that’s used for all kinds of things on the phone. And the ISP and encode an encoder engines are all new as well.

Theres a LOT more to A15 than just the CPU cores. There’s whole SIP design, RAM, cache, coprocessing, node improvements… people saying the two chips are identical or close to it is like saying Google doesn’t watch what you search. It’s asinine and contrary to the facts.
 
Last edited:
  • Like
Reactions: Cosmosent
As i‘ve explained elsewhere, highly unlikely that the difference between 5-core and 4-core chips is yield fallout. The GPUs are small, and the chance that a die happens to have one and only faulty GPU and that nothing else is broken due to defect clustering is vanishingly small. Almost certainly, the vast majority of the 4-core chips are simply chips with one core disabled intentionally.
You could be right. But I doubt Apple would pay the extra cost to accommodate an entirely different production line for the exact same chip. Binning is an industry practice…can’t see how this isn’t being used on the A15 as well. Remember, not all wafers will have any defects. But some may have 1 or 2 defects. If disabling a GPU core solves the defect, then yields go up.

If, as you say, they have a separate 4 core GPU production line, any defects on those chips would go into the trash bin…increasing costs and lowering yields. Something I highly doubt Apple is doing here.
 
You could be right. But I doubt Apple would pay the extra cost to accommodate an entirely different production line for the exact same chip. Binning is an industry practice…can’t see how this isn’t being used on the A15 as well. Remember, not all wafers will have any defects. But some may have 1 or 2 defects. If disabling a GPU core solves the defect, then yields go up.

If, as you say, they have a separate 4 core GPU production line, any defects on those chips would go into the trash bin…increasing costs and lowering yields. Something I highly doubt Apple is doing here.

I didn’t say anything about an additional production line.

I said the 4-core chips are 5-core chips with 1 core intentionally disabled, and that the disabling has nothing to do with defects.

Disabling a core does not require another production line. It’s done using fuses, at the end of the production line.

As for your other point, the chance that a die has 1 or 2 defects, and that defect is in one GPU, and happens to hit something that does not affect anything other than that one GPU is essentially 0. Defects cluster, on a wafer scale. If there is a defect, there are other defects “near by” - and “near by” is determined at the wafer scale, not the die scale.
 
It seems like most people here are missing the whole point. First of all yes the CPU cores themselves seem to be close to identical to the A14, but they are getting better throughput just because of the clock increase. Secondly, the multi core performance is significantly higher. This is probably a result of the doubling of the system cache. The GPU speed is a lot higher in the pro models especially to accommodate for Promotion but that raw power is also there. The Neural Engine is 40% faster which itself last year was 200% faster so in two years the Neural Engine has sped up over 280%. And that’s used for all kinds of things on the phone. And the ISP and encode an encoder engines are all new as well.

Theres a LOT more to A15 than just the CPU cores. There’s whole SIP design, RAM, cache, coprocessing, node improvements… people saying the two chips are identical or close to it is like saying Google doesn’t watch what you searich. It’s asinine and contrary to the facts.

There are whispers that the MMU was redesigned and there is LPDDR5. If so, in combination with the SLC changes, then the bandwidth has increased tremendously in all likelihood, meaning the CPUs do not get starved. This allows more clock speed headroom (e.g. M2 may run at a very high clock) but also improves multi-core (and graphics) performance a lot.
 
I didn’t say anything about an additional production line.

I said the 4-core chips are 5-core chips with 1 core intentionally disabled, and that the disabling has nothing to do with defects.

Disabling a core does not require another production line. It’s done using fuses, at the end of the production line.

As for your other point, the chance that a die has 1 or 2 defects, and that defect is in one GPU, and happens to hit something that does not affect anything other than that one GPU is essentially 0. Defects cluster, on a wafer scale. If there is a defect, there are other defects “near by” - and “near by” is determined at the wafer scale, not the die scale.
My apologies. I mis-read.
 
  • Love
Reactions: cmaier
Seems like the era of diminishing returns in terms of performance continues, and Apple has now broken their cadence with S releases - probably to over-hype minimal changes. It used to be that major non-S releases were significant changes, this is less of a change than say the 6-6S, or X - XS.

There are three things I see as primary reasons to go for a 13 and it has nothing to do with the SoC. That's the improved 5G modem (Qualcomm X60), larger base storage, and assuming Apple's assertions are correct the longer battery life. Of those 3 that modem is probably the biggest reason by a wide margin, yet it gets little press attention. These phones should have noticeably better reception and 5G performance than the 12.
 
  • Like
Reactions: SantaFeNM
10% year on year is great. Some years have been Moore but that is still great.
Yes and this means for me the M2 will be 10 % faster than the M1 in CPU performance.

So probably no reason to upgrade for all the M1 air owners for example.


edit: ok 10% in Cinebench or similar where only raw CPU performance counts.
 
Last edited:
Pretty dang impressive, the A14 has been killer in performance and battery life in the 12 Pro Max (leaps ahead of the 11 Pro Max) and looks like they outdid themselves yet again.
Except not. In real world use there was no difference between the 11 Pro and the 12 Pro. If anything, the smaller battery in the 12 meant less battery life compared to the 11, and that's something that can be verified with published fact and not just anecdotal hyperbole from an Apple faithful.
To quote my 12 year old, "I just can't even with this mentality."
 
  • Like
Reactions: gadsfan37
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.