Apple M2 SoC Analysis - Worse CPU efficiency compared to the M1

[Retskrad]

Apple M2 SoC Analysis - Worse CPU efficiency compared to the M1

The Apple M2 is here and after our reviews of the MacBook Pro 13 and the MacBook Air M2, we want to have a closer look at the new M2 chip. What changed, which improvements can you expect, and how does the increased power consumption affect the efficiency, especially compared to the previous M1 chip?

www.notebookcheck.net

Review of the M2 and how it stacks up against the M1, M1 Pro, M1 Max and latest offerings by AMD and Intel. Very interesting read.

 

[PauloSera]

It's a larger chip with higher clocks. They require more power.

 

[dugbug]

Seriously, the title is missing the point entirely. Do you want a 25% faster GPU? there you go, bigger chip.

-d

 

[Boil]

Seriously, the title is missing the point entirely. Do you want a 25% faster GPU? there you go, bigger chip.

Not to mention the addition of the Media Engine...

 

[PauloSera]

Seriously where the hell does this article get less efficiency from?

They list all of the following changes/improvements over M1:

• improved 5 nm process (probably N5P at TSMC).
• increased number of transistors (now 20 vs. 16 billion)
• a new Media Engine with support for 8K & ProRes files
• faster Neural Engine
• more powerful GPUs with 8 or 10 cores
• LPDDR5x RAM
• improve the maximum memory bandwidth from 68.25 GB/s to 100 GB/s and you can now get up to 24 GB unified memory
• the efficiency cores now reach up to 2.4 GHz (previously up to 2.1 GHz)
• The performance cores also reach a higher maximum clock of up to 3.48 GHz (previously 3.2 GHz)
• increased their L2 cache from 12 to 16 MB
• higher clock for the GPU cores (both 8 and 10-core versions) of 1398 MHz compared to 1278 MHz for the old M1 GPU

And then go on to say:

"The higher clock requires more power, because the old M1 consumed ~3.7 Watts and the new M2 requires ~5.3 Watts....This means that the single-core efficiency of the new M2 is about 26 % worse compared to the old M1 processor."

Lol an increase of 1.6 watts is deemed "26% worse". Based on what? The chips are not equal. You cannot claim the M2 is less efficient simply because it consumes more power. It is also doing significantly more than the M1. The argument could be made that it is even more efficient.

 

[leman]

Lol an increase of 1.6 watts is deemed "26% worse". Based on what? The chips are not equal. You cannot claim the M2 is less efficient simply because it consumes more power. It is also doing significantly more than the M1. The argument could be made that it is even more efficient.

Whats wrong with that claim? The performance cores are practically identical between M1 and M2 (not counting few incremental new features), M2 runs at a higher clock, so obviously it will use more power and since power increase is quadratic, will be less efficient. The new RAM. Ought be more power-hungry as well. So it’s a perfectly valid conclusion to make. After all, efficiency is performance per watt, and while M2 is faster it pays for it with a proportionately higher increase in power consumption.

But the folks who see all doom and gloom here are being ridiculous. It’s just Apple working in their vertical scaling. The power consumption is still ridiculously low compared to any other high-performance chip on the market, so there is nothing wrong with tweaking the power targets a bit. The hardware can take it and the users get faster software. What’s not to like?

 

[Darkseth]

Lol an increase of 1.6 watts is deemed "26% worse". Based on what?

It seems, you didn't read and/or understand the Article at all.

26% worse is NOT the difference in raw power consumption, but the graphic below, "Points per Watt".

The chips are not equal. You cannot claim the M2 is less efficient simply because it consumes more power.

When doing the exact same Task, you 100% can. That's even the definition of efficiency. Performance per Power consumed.

It is also doing significantly more than the M1.

I wouldn't say "up to 18%" at CPU Performance is significant.
it's a "bigger" Chip. More Clock speed, more GPU Cores, more Neural Cores, Media Engine. That requires power for just existing.

The argument could be made that it is even more efficient.

No it couldn't.
Just as you couldn't make an Argument that $50 is more than $100.

You can measure power efficiency pretty easily, and Notebookcheck did it. Comparing 2 numbers really isn't difficult and doesn't leave any room for argumentation.

297 points per Watt is less than 407 points per Watt.


Even aside Benchmarks, just read the full Review. Batterylife with Wifi webbrowsing or Videoplayback went down. Same Task, higher Power consumption = less efficiency.

 

[Xiao_Xi]

Is this the first time Apple has replaced an SoC with a less efficient one?

 

[Pressure]

Geekerwan also compared the M1 and M2 to other Windows based systems and shows that the M2 MacBook Pro actually achieves higher battery life than the M1 MacBook Pro on their script based battery life test. Their battery test is based on 1 hour cycle that streams music constantly, 10 minutes each to Microsoft Office (Excel, PowerPoint and Word), 10 minutes web browsning, 10 minutes online video streaming in 4K and 10 minutes of instant messaging.

 

[evertjr]

No it couldn't.
Just as you couldn't make an Argument that $50 is more than $100.

You can measure power efficiency pretty easily, and Notebookcheck did it. Comparing 2 numbers really isn't difficult and doesn't leave any room for argumentation.

297 points per Watt is less than 407 points per Watt.


Even aside Benchmarks, just read the full Review. Batterylife with Wifi webbrowsing or Videoplayback went down. Same Task, higher Power consumption = less efficiency.

This is a pretty ridiculous assumption, what the heck I'm supposed to do with "watt points"? Apple doesn't sell processors so these numbers doesn't really matter, what matters is the products being sold with the M2 chip currently are faster and have the same battery life (or better).

It uses a little more power but can accomplish more in important areas like graphics and videos encoding.

So yes, in this case $50 worth more than $100.

 

[bcortens]

This is a pretty ridiculous assumption, what the heck I'm supposed to do with "watt points"? Apple doesn't sell processors so these numbers doesn't really matter, what matters is the products being sold with the M2 chip currently are faster and have the same battery life (or better).

It uses a little more power but can accomplish more in important areas like graphics and videos encoding.

So yes, in this case $50 worth more than $100.

From the perspective of someone just interested in the technology the reduction in efficiency is something to be a little concerned about. As leman said, they are still ridiculously low power compared to the competition but it is the wrong direction. Hopefully this was mostly due to the delays on the 4N (refined 5N) and 3N process and that when we get the chips on 3N we see efficiency climb back upwards.

It is impressive to see that the extra GPU cores actually lead to an increase in points per watt on the GPU front even with the increase in power available to the GPU.

 

[Xiao_Xi]

Geekerwan also compared the M1 and M2 to other Windows based systems and shows that the M2 MacBook Pro actually achieves higher battery life than the M1 MacBook Pro on their script based battery life test.

Do the M1 13" MacBook Pro and M2 13" MacBook Pro use the same battery? Are they interchangeable?

 

[Pressure]

Do the M1 13" MacBook Pro and M2 13" MacBook Pro use the same battery? Are they interchangeable?

They pretty much use all the same parts, which is why comparing the M1 and M2 in MacBook Pro 13" should give us the best picture on whether it really is a regression with regards to power efficiency.

 

[PauloSera]

It seems, you didn't read and/or understand the Article at all.

26% worse is NOT the difference in raw power consumption, but the graphic below, "Points per Watt".
View attachment 2117440

When doing the exact same Task, you 100% can. That's even the definition of efficiency. Performance per Power consumed.

I wouldn't say "up to 18%" at CPU Performance is significant.
it's a "bigger" Chip. More Clock speed, more GPU Cores, more Neural Cores, Media Engine. That requires power for just existing.

No it couldn't.
Just as you couldn't make an Argument that $50 is more than $100.

You can measure power efficiency pretty easily, and Notebookcheck did it. Comparing 2 numbers really isn't difficult and doesn't leave any room for argumentation.

297 points per Watt is less than 407 points per Watt.


Even aside Benchmarks, just read the full Review. Batterylife with Wifi webbrowsing or Videoplayback went down. Same Task, higher Power consumption = less efficiency.

It seems you didn't read the article at all. The article makes its actual claim based solely on the power used and pretends like the M1 and M2 are identical. You might as well say the M1 Pro is "worse" than the the M1 too, since it uses more power. The article is idiotic.

 

[PauloSera]

Is this the first time Apple has replaced an SoC with a less efficient one?

It's not less efficient. It does more. If the M1 and M2 had the exact same performance characteristics, then you could make this claim.

 

[PauloSera]

Whats wrong with that claim?

They're not identical chips. That's whats wrong with the claim.

 

[leman]

It seems you didn't read the article at all. The article makes its actual claim based solely on the power used and pretends like the M1 and M2 are identical.

No, they make the claim based on power used and performance delivered using that power. What's the confusion?

It's not less efficient. It does more. If the M1 and M2 had the exact same performance characteristics, then you could make this claim.

I think your understanding of what constitutes efficiency might be quite different from its common definition. Efficiency = work done per unit of energy, so claims of M2 being less CPU-efficient than M1 on CB23 are absolutely correct. It's entirely different issue that these comparisons bear very little useful information.

They're not identical chips. That's whats wrong with the claim.

From the perspective of the P-cores they might as well be. The IPC is identical. M2 supports some additional instructions and the caches are slightly different, but that's about it. If you run it at the same frequency as M1 you'll get the same exact performance.

As leman said, they are still ridiculously low power compared to the competition but it is the wrong direction.

I disagree that this is the wrong direction. In fact, I welcome this direction as it gives the user more performance without any meaningful drawbacks. I mean, it would be definitely worrying if this trend continues but Apple can easily go up to 8-10W on the P-core and still maintain 2-3x efficiency headroom over the competitors. The actually worrying thing about M1 for me is the lack of any vertical performance capability on the P-cores and M2 is a proof that Apple is not committed to run their laptops at the same power level as a smartphone.

 

[PauloSera]

It's entirely different issue that these comparisons bear very little useful information.

I disagree.

 

[Xiao_Xi]

It's not less efficient. It does more.

The key question is whether the increase in M2 performance is greater than the increase in consumption. If so, M2 is more efficient. If not, it is less efficient.

According to Notebookcheck, the increase in M2 performance is less than the increase in consumption, so it is less efficient.

 

[leman]

I disagree.

You do? I must say, your posts give a very different impression. Aren't you claiming that these efficiency claims are meaningless? If so, how do you disagree when I say that these efficiency claims are meaningless?

 

[PauloSera]

You do? I must say, your posts give a very different impression. Aren't you claiming that these efficiency claims are meaningless? If so, how do you disagree when I say that these efficiency claims are meaningless?

The overall claims are broader, and the point they try to imply is false. The actual results are so insignificant to almost make the analysis itself a waste of time. But instead it is used to imply that the M2 is somehow worse than the M1.

 

[leman]

The overall claims are broader, and the point they try to imply is false. The actual results are so insignificant to almost make the analysis itself a waste of time. But instead it is used to imply that the M2 is somehow worse than the M1.

So you don't disagree after all? That's exactly what I have been saying.

 

[PauloSera]

So you don't disagree after all? That's exactly what I have been saying.

You said:

"It's entirely different issue that these comparisons bear very little useful information."

And I said, I disagree. It is not an entirely different issue. It is the entire issue. The article and its subsequent posting here with its misleading title intact, have an entirely purpose that does not align with reality.

 

[TPadden]

You said:

Tomato, potato…

 

[Love-hate 🍏 relationship]

Seriously where the hell does this article get less efficiency from?

They list all of the following changes/improvements over M1:

• improved 5 nm process (probably N5P at TSMC).
• increased number of transistors (now 20 vs. 16 billion)
• a new Media Engine with support for 8K & ProRes files
• faster Neural Engine
• more powerful GPUs with 8 or 10 cores
• LPDDR5x RAM
• improve the maximum memory bandwidth from 68.25 GB/s to 100 GB/s and you can now get up to 24 GB unified memory
• the efficiency cores now reach up to 2.4 GHz (previously up to 2.1 GHz)
• The performance cores also reach a higher maximum clock of up to 3.48 GHz (previously 3.2 GHz)
• increased their L2 cache from 12 to 16 MB
• higher clock for the GPU cores (both 8 and 10-core versions) of 1398 MHz compared to 1278 MHz for the old M1 GPU

And then go on to say:

"The higher clock requires more power, because the old M1 consumed ~3.7 Watts and the new M2 requires ~5.3 Watts....This means that the single-core efficiency of the new M2 is about 26 % worse compared to the old M1 processor."

Lol an increase of 1.6 watts is deemed "26% worse". Based on what? The chips are not equal. You cannot claim the M2 is less efficient simply because it consumes more power. It is also doing significantly more than the M1. The argument could be made that it is even more efficient.

what is it that you consider worse in efficiency then ? the power consumption in relation to the performance is the most absolute and definitive way to calculate a chip's efficiency afaik