In the first article in this series, I explained the major differences between the chips and CPUs in Apple silicon Macs and their Intel predecessors, focussing particularly on performance and their two core types. The underlying purpose in much of their design is to deliver that performance with high power efficiency, the theme of this article.

Total power

Apple provides detailed power figures for one series of models, Mac mini, over the period from the original PowerPC G4 model in 2005 to the current M2 Pro model from 2023. As these were measured at a steady state, heat output from the minis tested equalled power used, readily confirmed if you convert the figures given in BTU/h (a British unit abandoned by Britain over half a century ago) into Watts. I show Apple’s reported power data in the chart below.

The upper sets of lines and points are those for what Apple terms CPU maximum, when running a maximal CPU-intensive task, and the lower lines and points are when the Finder is idle with default power settings. Those shown in black are for single and 2-core processors, and those in red are for 4-core and greater.

Historically, Mac mini idle power consumption has fallen steadily, until it rose with the Intel 6-core model in 2018. Following that, all three Apple silicon models have had extremely low idle power use at around 7 W. Power used at CPU maximum remained between 85-110 W for much of the past, until the 6-core model in 2018, when it rose to 122 W. Apple silicon models with the base M1 or M2 chip have had maximum power figures of about half those of Intel models, and even the 12-core (8P+4E) M2 Pro model uses only 100 W at CPU maximum.

Compare those with the remarkable rise in typical Geekbench scores over the last few years. Single-core scores have risen from 1,200 for the Intel 6-core to 1,900 for M2 models, and multi-core from 6,000 for the Intel to 9,000 and 15,000 for the M2 and M2 Pro respectively:

• Intel 6-core 1200, 6000
• M1 4P+4E 1750, 7700
• M2 4P+4E 1900, 9000
• M2 Pro 8P+4E 1900, 15000

where the first score is single-core, and the second multi-core.

As complete systems, Apple silicon Macs have better performance (even given my reservations about multi-core benchmarks) for lower power consumption, hence their heat output has fallen significantly.

How is low power achieved?

CPUs in M-series chips achieve efficiency in power use in several ways:

• by running low-QoS threads on cores designed to use less power,
• by varying the frequency (and possibly voltage) of cores according to their demand,
• by managing cores in clusters, so that a whole cluster of P cores can be left in low-power mode until they’re required,
• by tight integration into a single SoC.

The importance of core frequency and voltage are apparent in the formula for estimating dynamic power use:
P = C × f × V^2
where P is dynamic power, C is a constant normally considered to be a switched load capacitance, f is core frequency, and V is voltage. Thus, if voltage is constant, a core will use twice as much power when its frequency is doubled from idle of about 700 MHz to 1,400 MHz. For M3 P cores, you’d thus expect their power consumption to increase nearly 6 times when their frequency is increased from idle at just under 700 MHz to its maximum at just over 4,000 MHz.

What uses the power?

According to estimates given by the powermetrics command tool, relatively little of the total power measured in Apple’s tests is used by the CPU cores. When at idle, M1 Pro (8P+2E) and M3 Pro (6P+6E) CPU cores typically use less than 0.2 W, and GPUs use almost zero, but the whole Mac excluding display is likely to be using 5-10 W of power in total. That leaves well over 90% of idle power use to be accounted for by the rest of the chip, including its attached memory, and other modules such as the SSD, power supply, fans and network interfaces.

Another part of Apple’s low power strategy is to move specialist high-power processing into separate units. That includes the NEON vector processor built into each CPU core, the neural engine (ANE), an undocumented matrix co-processor (AMX), and the GPU running in Compute mode. When running maximum conventional CPU loads, maximum power used by all 12 cores in an M3 Pro remains less than 7 W, while the 10 cores in an M1 Pro should remain below 9 W. However, reported power consumption when those units are heavily loaded can rise greatly: the peak seen in tests on the M3 Pro is just under 45 W when it’s presumed the AMX was in heavy use.

Total power figures given for the Mac mini also take no account of that used by the GPU. During a heavy Compute workload, that can amount to as much as 24 W in the modest GPU of an M3 Pro, and would be expected to be significantly higher in the larger and faster GPU of an M3 Max. Those possibilities are allowed for in the “maximum continuous power” given by Apple for the Mac mini M2 Pro model as 185 W, which is likely to be the sum of maximum sustained power requirements for all chips and modules, and not something that should ever be encountered in normal use.

Why is power important?

Power, or rather energy, use is of most critical importance to notebook computers, where it determines endurance between recharging the battery. It’s also of obvious importance when large numbers of computers are operated in a confined space, such as in server accommodation and data centres, which may use more power cooling their ambient air than running their computers.

Totalled over a year, you’d have to run many Apple silicon Macs to see any significant return in the cost of their power or energy, so for desktop Macs the most immediate benefit is likely to be seen in their cooling requirements. A Mac that uses between 7 and 100 W of power has to dissipate almost all of that as heat, while one that uses 20-122 W, like the 2018 6-core Mac mini, has to dissipate nearly 3 times as much heat when idle, and 122% more when running with a high CPU workload.

In most situations, that means Apple silicon Macs run cooler than their Intel predecessors, although measuring temperature isn’t a good way to monitor this, as it’s a variable controlled by system management, which is well-known for running kernel_task processes and pulling other tricks to prevent chip temperatures from reaching high levels. Instead, chip manufacturers including both Intel and Apple recommend using metrics such as thermal pressure as the best indicators of thermal strain.

Ultimately, when thermal pressure becomes too great for system management, processing can be first throttled, then shut down altogether, to protect the chip from thermal damage. This is particularly easy in Apple silicon cores and the GPU because they already run at variable frequency (and possibly voltage). As dynamic power is proportional to core frequency, 10% reduction in frequency may be sufficient to reduce heat output and restore a steady state with low thermal pressure, a technique known as dynamic frequency scaling.

Thus the biggest benefits to all Apple silicon Mac users, whether in notebooks or desktops, is their ability to sustain high performance without being constrained by rising thermal pressure.

What I haven’t considered yet is whether they achieve any net energy savings. It’s all very well running background processes slowly on Efficiency cores, but does that extend battery endurance? Some would argue that it’s more efficient to ‘race to idle’ by running every process as quickly as possible so the CPU cores can be put back into low-power idle.

Concepts

• Overall, Apple silicon Macs deliver better performance with lower power consumption.
• Core frequency control is central to performance, power use and thermal management.
• Moving specialist high-power processing into separate units is another strategy for power efficiency.
• Both performance and power use can be significantly higher when running computational tasks in the GPU, AMX co-processor or neural engine.
• Thermal pressure is the best indicator of thermal strain.
• Power efficiency enables high performance to be sustained at low thermal pressure.

Previously in this series

Apple silicon: 1 Cores, clusters and performance

Further reading

Evaluating M3 Pro CPU cores: 1 General performance
Evaluating M3 Pro CPU cores: 2 Power and energy
Evaluating M3 Pro CPU cores: 3 Special CPU modes
Evaluating M3 Pro CPU cores: 4 Vector processing in NEON
Evaluating M3 Pro CPU cores: 5 Quest for the AMX
Evaluating the M3 Pro: Summary
Finding and evaluating AMX co-processors in Apple silicon chips
Comparing Accelerate performance on Apple silicon and Intel cores
M3 CPU cores have become more versatile