M1 MacBook Air Gets No Price Drop, 8-Core GPU No Longer an Option

[Fomalhaut]

Everyone was expecting an Apple Silicon Mac Pro at this WWDC.
So ok, the M3 is delayed along with the Apple Silicon Mac Pro. My $$$$ stays in the bank.
Apple missed their 2 year transition goal. Period.

IIRC, Tim Cook said "about 2 years", when M1 was announce in June(?) 2020. I thought they might announce the Mac Pro at WWDC, but it's not entirely surprising that they didn't. They still have until the end of the year to stay within the "about 2 years" approximation, and I don't think they are likely to be punished by delaying a few months given the global circumstances outside Apple's control. Stock prices have hardly fallen off a cliff, ergo, not many people are bothered by this....except you apparently.

 

[Fomalhaut]

Completely wrong. Multi-core performance is NOT simply the sum of single core performance. Multi-core measures how efficient each core can work without impacting (blocking) other cores. For example, It makes no sense if all other cores must wait for a single core to access memory. Multicore metrics (tests) strive to be independent of Single core because multicore measures the efficiency of the memory bus architecture.

I did not say that multi-core is simply the sum of single core. Please re-read my comment. Multicore scores, as you rightly point out, are dependent on the overall system bandwidth and how this is affected by running all the cores simultaneously. But each individual core contributes to that score - not linearly - but the throughput of each core will affect the multi-core score.

You haven't yet demonstrated mathematically how you can have a higher multi-core throughput than the M1 without increasing the throughput of individual cores. Where do you think the multi-core improvement comes from? Please show us your calculations.

 

[Fomalhaut]

Check it out: The new so called "M2" MacBook Air base configuration only has EIGHT GPU cores, just like the original M1 MacBook Air with 8 cores (no longer available).
It will be interesting to compare the new "M2" 8 core GPU against the M1 8 Core GPU system. That's when this so called "M2" chip will earn the name M1 Plus, which will be a more accurate name.
All the performance improvement metrics presented in the keynote compared the 10 core GPU against the 8 core GPU M1. Apple's sleight of hand at work.

Why does the M2 have to align with your arbitrary expectation of what it should be? I'm baffled....

It was pretty much accepted that the next base-level Apple Silicon was likely to have a similar improvement to the A14->A15 (or maybe A15->A16?) and have two more GPU cores. This is exactly what it has.

What were you honestly expecting? Something that would be almost as good as the M1 Pro?

Intel and AMD have no problem in increasing their CPU generation names/numbers for modest (5-10%) improvements, so what stops Apple doing this?

What exactly are you after? If you think you can do better, kindly submit your resume to Apple. You clearly have an advanced degree in microprocessor design and will no doubt be an invaluable asset to Apple's engineering department....

 

[Fomalhaut]

Might also compare the "M2" 8 GPU core vs the "M2" 10 GPU core, which is a $100 USD difference. There is an 8 core GPU option for this so called "M2" chip although most peeps seem to be missing this.
The 7 GPU core M1 option is $200 or $300 (vs 8 or 10 GPU) USD less without the options you listed.
You can order the 10 core GPU with 256GB SSD if you select the 8 core unit then bump the GPU option on the BTO page.

Summary: I think the overall MacBook Air update is good. But calling this new new chip an "M2" is BS. It should be called the M1 Plus. Apple did good work on the Pro, Max, and Ultra, so "refreshing" the original M1 would have been completely fine. This "M2" chip really doesn't contain any "internal" tech that significantly outperforms the Pro, Max, or Ultra chips. This will really be evident when this "M2" MacBook Air w/8 GPU cores is compared against an M1 MacBook Air with 8 GPU cores. Sure it will be faster, but only slightly - and thus only a minor performance bump.

Single core performance (currently unknown) will become an issue going forward because Intel is not sitting still. More importantly, all these performance metrics (single core multi core, etc... ) are really just "building blocks" used to configure a system appropriately for a user's needs. I'm a software dev so I need more CPU cores to build large open source libraries with thousands of source files. Intel hyper threading is great for this because one core could compile 2 files concurrently. Apple Silicon doesn't have hyper threading so I need more real cores to do the same job.

I agree with the bold text....but it's exactly what most of us expected. The recently-released M2 is the next generation of the base M1. It will be slightly faster than the M1, but I don't think many people were expecting any major architectural improvements, and certainly nothing that would allow it to compete against the M1 Pro, Max or Ultra.

What was your expectation, and why? I'm sorry you're disappointed....I think you must have be misled about what M2 was going to be, hence your determination to declare that it's not the "real M2". It's just the next one after "1", i.e. "2". the one after will be....<drum roll>... "3".

Considering Intel post-2015 was offering 5-10% year-on-year improvements, if Apple gets over 15% they are doing quite well in my opinion.

 

[Abazigal]

That's the thing to keep in mind with Apple keeping an almost two year old system in the lineup: these things have gotten such a good performance that the average consumer will struggle to reach a performance bottleneck.

It doesn’t hurt that almost two years later, the competition still doesn’t have anything remotely close to the M1 chip in terms of performance and battery life.

 

[d686546s]

It doesn’t hurt that almost two years later, the competition still doesn’t have anything remotely close to the M1 chip in terms of performance and battery life.

Absolutely. Though I have a Windows machine with a CPU released in 2017 for work -- I unfortunately have no say in the matter -- and tbh for office tasks (video calls, spreadsheets, presentations and documents) even that is more than enough. Battery life is horrible and I prefer the Mac, but there certainly isn't the pressure on performance anymore as there used to be.

Your mileage may obviously vary and I'm quite aware that my private and professional needs are far from demanding but I doubt that I'm exceptional in that regard.

 

[developer13245]

I did not say that multi-core is simply the sum of single core. Please re-read my comment. Multicore scores, as you rightly point out, are dependent on the overall system bandwidth and how this is affected by running all the cores simultaneously. But each individual core contributes to that score - not linearly - but the throughput of each core will affect the multi-core score.

You haven't yet demonstrated mathematically how you can have a higher multi-core throughput than the M1 without increasing the throughput of individual cores. Where do you think the multi-core improvement comes from? Please show us your calculations.

This isn't a mathematical issue, its plain Computer Science.
Effective Multicore performance measurement should strive to be agnostic of single core performance. Yes, that's a fact because multicore performance is a measurement of memory bus architecture/implementation performance, NOT CPU performance.
Proof (you can look this up): The 2009 Mac Pac Pro upgraded to a new Intel CPU that allowed each CPU to access memory without blocking other CPUs from doing so. This transition was the first introduction of tech that departed from old style "Front Side Bus" FSB memory controller architecture. FSB required each CPU to wait while another CPU had access to memory. I (sorta) remember Intel's architecture was named something like "Quick Processor Interconnect" (QPI) or something... Multicore also measures CPU memory cache performance, which is part of and helps the overall memory controller architecture.
I've been writing commercial software for 35 years (began on Motorola 68000 series architecture - assembly!), using concurrent programming techniques.

 

[developer13245]

Why does the M2 have to align with your arbitrary expectation of what it should be? I'm baffled....

It was pretty much accepted that the next base-level Apple Silicon was likely to have a similar improvement to the A14->A15 (or maybe A15->A16?) and have two more GPU cores. This is exactly what it has.

What were you honestly expecting? Something that would be almost as good as the M1 Pro?

Intel and AMD have no problem in increasing their CPU generation names/numbers for modest (5-10%) improvements, so what stops Apple doing this?

What exactly are you after? If you think you can do better, kindly submit your resume to Apple. You clearly have an advanced degree in microprocessor design and will no doubt be an invaluable asset to Apple's engineering department....

My exact point: Comparing the MB Air M1 8/8 CPU/GPU vs "M2" 8/8 CPU/GPU performance will show the M2 is not that much better.
And: Apple's comparison of M1 8/8 vs "M2" 8/10 was disingenuous. Especially because they have the "M2" 8/8, and could have presented that comparison, which would have been fair. They were going for higher numbers to look good.

 

[Fomalhaut]

This isn't a mathematical issue, its plain Computer Science.
Effective Multicore performance measurement should strive to be agnostic of single core performance. Yes, that's a fact because multicore performance is a measurement of memory bus architecture/implementation performance, NOT CPU performance.
Proof (you can look this up): The 2009 Mac Pac Pro upgraded to a new Intel CPU that allowed each CPU to access memory without blocking other CPUs from doing so. This transition was the first introduction of tech that departed from old style "Front Side Bus" FSB memory controller architecture. FSB required each CPU to wait while another CPU had access to memory. I (sorta) remember Intel's architecture was named something like "Quick Processor Interconnect" (QPI) or something... Multicore also measures CPU memory cache performance, which is part of and helps the overall memory controller architecture.
I've been writing commercial software for 35 years (began on Motorola 68000 series architecture - assembly!), using concurrent programming techniques.

I'm probably much closer to you than you think and also remember programming the 68000, and even the Z80, 6502 and 6809 in assembly language before that.

You make some good points on the effect of bus architectures on multi-core benchmark scores, but unless I am completely mistaken about how tools like GB5 are implemented, they run the same code against either a single thread, or multiple threads. I suppose it *is* conceivable that an improved multi-core score could be solely down to architectural improvements without any increase in actual single core speed. However, there is no indication that this is the case (or you haven't provided one) and pretty much every single GB5 benchmark that I've seen shows a relationship between improved multi-score and improved single-core. I'm happy to receive a counter example if you can find one.

 

[Fomalhaut]

My exact point: Comparing the MB Air M1 8/8 CPU/GPU vs "M2" 8/8 CPU/GPU performance will show the M2 is not that much better.
And: Apple's comparison of M1 8/8 vs "M2" 8/10 was disingenuous. Especially because they have the "M2" 8/8, and could have presented that comparison, which would have been fair. They were going for higher numbers to look good.

I can't disagree with that....but that is what Apple does. Their performance comparisons are hardly rigorously scientific and clearly driven by marketing, although they do generally include the test details in the (very) small print at the bottom of the web page.

My point is that the M2 is an update to the M1, albeit a small one, so Apple is within their rights to call it the M2, especially since it is 2 years after the M1. Calling it an M1+ or similar would send the wrong message to buyers. It's just a name, like Intel's core-i generations. It doesn't mean anything in absolute terms.

 

[Eidorian]

Absolutely. Though I have a Windows machine with a CPU released in 2017 for work -- I unfortunately have no say in the matter -- and tbh for office tasks (video calls, spreadsheets, presentations and documents) even that is more than enough. Battery life is horrible and I prefer the Mac, but there certainly isn't the pressure on performance anymore as there used to be.

Your mileage may obviously vary and I'm quite aware that my private and professional needs are far from demanding but I doubt that I'm exceptional in that regard.

My work laptop is from 2017 and my desktop from 2018. I admit to having a 32" + 24" display for my desktop. That's really all I need at my desk.

The laptop is just used for connecting to our inventory and video conferencing when I'm on the road. 2 cores, 4 threads with a SSD is more than enough for that. I have a brand new battery though and yes it lasts all day on the road.

 

[Rychiar]

I can’t even imagine working on a 256gb machine in 2022. My iMac at work is 512 and we use network storage for all of our files yet I still find myself running out of drive space when editing a big card of photos or videos