I think M1 & M3 series chips are more "mobile" in nature than M2 series chips

[sam_dean]

I say this as when pushed 5nm M1 & 3nm M3 series chips uses less power & generates less waste heat because of the their input power relative to their respetice die shrink as compared to 5nm M2 series chips.

So if I were to buy a Mac with Apple Silicon I'd opt for a M1 series Mac for MBP & MBA while M2 series Mac for Mac mini, iMac & Mac Studio.

When the 2021 Mac Studio came out I spoke up that I wish Apple overlocked the M1 Max & M1 Ultra to take advantage of the desktop form factor. It appears that wish will be fulfilled with the 2023 Mac Studio as early as June 2023.

 

[Boil]

I say this as when pushed 5nm M1 & 3nm M3 series chips uses less power & generates less waste heat because of the their input power relative to their respetice die shrink as compared to 5nm M2 series chips.

Pretty cool you have access to the as-yet-ureleased M3-series of ASi SoCs...!

So if I were to buy a Mac with Apple Silicon I'd opt for a M1 series Mac for MBP & MBA while M2 series Mac for Mac mini, iMac & Mac Studio.

If one were buying their first ASi Mac it would only make sense to go for a M2-series SoC, unless one found an especially good deal on a M1-series SoC...

When the 2021 Mac Studio came out I spoke up that I wish Apple overlocked the M1 Max & M1 Ultra to take advantage of the desktop form factor. It appears that wish will be fulfilled with the 2023 Mac Studio as early as June 2023.

ASi is not designed for overclocking, any increase in clock speeds needs to be designed for, not just an arbitrary "ramp the voltage for more speed"; this ain't Intel...

 

[leman]

I thought reviewers agreed that M2 Pro/Max models had better battery life?

 

[Xiao_Xi]

ASi is not designed for overclocking, any increase in clock speeds needs to be designed for, not just an arbitrary "ramp the voltage for more speed"; this ain't Intel...

What prevents Apple Silicon from under/overclocking?

 

[Boil]

What prevents Apple Silicon from under/overclocking?

Hardware side, by design; software side, no BIOS/UEFI to access...?

 

[Xiao_Xi]

Hardware side, by design;

What do you mean by design?

I thought reviewers agreed that M2 Pro/Max models had better battery life?

From Dave2D's Apple M2 MacBook Pro -14" vs 16" video

 

[leman]

What do you mean by design?

The circuitry doesn't tolerate higher clocks. To put differently, x86 CPUs are designed to operate under a vast range of clock frequencies (but trade some performance for it), while Apple Silicon's frequency ceiling is much lower (but it can do more per clock).

 

[Scarrus]

I think you worry way too much about these things. Honestly if a M1 MacBook hasn't got enough battery life for you then I doubt the M3 is going to solve that issue.

That said I think we will see a nice bump in performance and probably battery life too with the M3. Still, M1 and M2 Macs will still be fast and power efficient computers, even by then.

 

[Xiao_Xi]

The circuitry doesn't tolerate higher clocks. To put differently, x86 CPUs are designed to operate under a vast range of clock frequencies (but trade some performance for it), while Apple Silicon's frequency ceiling is much lower (but it can do more per clock).

Is there any proof that it is a hardware limitation? What makes you think it is a hardware limitation if you cannot test it without the necessary software?

 

[leman]

Is there any proof that it is a hardware limitation? What makes you think it is a hardware limitation if you cannot test it without the necessary software?

There is indirect evidence. If these chips could easily go higher, don't you think Apple will allow higher clocks on the Pro/Max? Outperforming x86 desktop in single-core benchmarks would make for some great publicity.

Other bits of evidence is that Apple has been continuously tweaking sizes of the internal structures in the CPU as well as the circuitry layout. So this is very different than "take M1 and crank up the frequency".

 

[Xiao_Xi]

If these chips could easily go higher, don't you think Apple will allow higher clocks on the Pro/Max? Outperforming x86 desktop in single-core benchmarks would make for some great publicity.

I don't think so, because it would reduce performance/watt and Apple cares more about efficiency than performance.

If I'm not mistaken, x86 CPUs can achieve higher frequency by increasing the voltage. So, even if Apple's SoCs ran at the highest possible frequency, wouldn't they achieve higher frequency by increasing the voltage?

 

[sam_dean]

Hardware side, by design; software side, no BIOS/UEFI to access...?

Overclocked by the manufacturer.

 

[sam_dean]

There is indirect evidence. If these chips could easily go higher, don't you think Apple will allow higher clocks on the Pro/Max? Outperforming x86 desktop in single-core benchmarks would make for some great publicity.

Other bits of evidence is that Apple has been continuously tweaking sizes of the internal structures in the CPU as well as the circuitry layout. So this is very different than "take M1 and crank up the frequency".

Or they do not want to use any resources for more variables in their supply chain.

Binned M1 Max in laptop or desktop are identical... which I find jarring because Intel/AMD did not do that.

 

[leman]

I don't think so, because it would reduce performance/watt and Apple cares more about efficiency than performance.

Just think about it and you’ll see that it’s a weak argument. Other manufacturers let their CPU cores run at 20 watts or more, Apple Silicon only does 5-6 watts. Why not double it and claim not only unmatched efficiency but also the lead in absolute performance? It won’t make the computer run any hotter and won’t sacrifice battery life.

Basically your argument is that Apple holds their hardware back. There is no good business sense in doing that. It’s not like there will be a M2 plus with slightly higher clocks (a strategy often used by PC vendor to sell old chips as new revision). Even if you’d argue that Apple wants to keep conservative clock for mobile, why not give them more juice for the desktop? Just 20% more performance and M2 Pro would make a much better figure agains with top desktop CPUs. That would be impressive marketing!

No, since we don’t see Apple do any of this, to me it’s very clearly an inherent limitation of their silicon. Thats the price you pay for supreme energy efficiency. Can’t have it all.

If I'm not mistaken, x86 CPUs can achieve higher frequency by increasing the voltage. So, even if Apple's SoCs ran at the highest possible frequency, wouldn't they achieve higher frequency by increasing the voltage?

You can’t increase voltage arbitrarily, there are limits. And I’m not talking about heat but about stability. Different parts of the hardware won’t be able to synchronize the execution past certain speeds, not to mention that at higher voltages you might get various electrical malfunctions. These limits are different for different designs.

To use an ever popular car analogy. For a sports motorbike running an engine at 7000 RPM is the norm, they won’t reach max speed until 9000rpm or higher. If you tried to run a diesel car engine at that speed it would probably fall apart.

 

[leman]

Or they do not want to use any resources for more variables in their supply chain.

Binned M1 Max in laptop or desktop are identical... which I find jarring because Intel/AMD did not do that.

It’s possible that at small proportion of these chips can run at slightly higher frequencies. But since Apple is not binning by speed it probably means that it’s not worth it. So they decided to go for consistency across the models and bin for amount of cores instead.

Interestingly, M2 Max does appear to run slightly higher clock than M2 Pro.

 

[Xiao_Xi]

It’s possible that at small proportion of these chips can run at slightly higher frequencies. But since Apple is not binning by speed it probably means that it’s not worth it. So they decided to go for consistency across the models and bin for amount of cores instead.

It is statistically impossible for all SoCs to run at the same frequency, so Apple has to choose a lower common frequency so that all SoCs have the same frequency. That would at least mean that some SoCs could be overclocked.

Basically your argument is that Apple holds their hardware back. There is no good business sense in doing that. It’s not like there will be a M2 plus with slightly higher clocks (a strategy often used by PC vendor to sell old chips as new revision). Even if you’d argue that Apple wants to keep conservative clock for mobile, why not give them more juice for the desktop? Just 20% more performance and M2 Pro would make a much better figure agains with top desktop CPUs. That would be impressive marketing!

No, since we don’t see Apple do any of this, to me it’s very clearly an inherent limitation of their silicon. Thats the price you pay for supreme energy efficiency. Can’t have it all.

Manufacturing differences between SoCs aside, I find it highly unlikely that Apple could design an SoC at its maximum voltage and frequency and achieve the best efficiency.

Interestingly, M2 Max does appear to run slightly higher clock than M2 Pro.

The difference is not between M2 Max and M2 Pro, it is between M2 Max on 16" MacBook Pro and 14" MacBook Pro. Geekbench detects higher frequencies in M2 Max on Mac 14.6 (~3.68 GHz) than on Mac 14.5 (~3.48 GHz).

 

[kiranmk2]

As Apple doesn't sell chips, it makes sense that they clock the chips as fast as they can to hit the power/therm specs of the machine. There is evidence of them down clocking their chips from as far back as the A5 (down-clocked for example in the AppleTV).

For the binned chips, I'm not aware if they are binned because some cores don't work, or because the power draw for all the cores is out-of-spec.

 

[maflynn]

There is evidence of them down clocking their chips from as far back as the A5

It goes back further as I remember having discussions eons ago regarding Apple's powerbooks and Apple down clocking the AMD GPU

 

[Scarrus]

Don't M1 & M2 Pros and especially Max run at around 20W?

Anyway, it's also possible that the chips are so over engineered that you could add another 20W on top for 40W as an example and you would only get like 10% better performance.