Way back when, you bought a CPU, and it had a marked clock rate. You ran it at that rate. The end.
Later you got a turbo button, but that was more for application compatibility. We were spinning for a fixed number of cycles to mark time!
Around the Athlon time, we started throttling CPUs. It turned out that they could be damaged if run too hot for too long, and laptops were having trouble getting the heat out. So Intel (and later AMD) parts started to slow themselves down if they got to a dangerous temperature.
Later still – within the last five years – we got ‘Turbo Boost’. Originally, this was to reflect that the CPU could run faster for a very brief time, but eventually we would be unable to remove the heat fast enough and the CPU could reach dangerous temperatures again. In some ways, this reflected the thermal mass of the CPU, its heatspreader and the immediate heatsink. Heatpipes were now in common use, and while they could remove a lot of heat from a small area, they couldn’t change the rate of heat conductance rapidly. While desktops were usually designed to remove all of the heat that the CPU could produce at maximum power, laptops couldn’t afford this – the space and weight required was just too great.
Recently, “a brief time” has become “a really long time”. My wife’s Macbook Air, for instance, runs at a ‘base clock’ of 1.6GHz. If you watch the actual CPU speed, however, it never runs at 1.6GHz. If it’s idle, it will run at less than 1GHz (and it’ll actually be asleep for much of that). If you work it hard, it’ll increase to 2.4GHz. For as long as the workload lasts. So there’s no thermal mass effect here – it’s just 1GHz/sleeping for low load, 2.4GHz for high load, and somewhere in the middle for a mixed load.
Under high load, the clock rate is determined by the cooling capacity of the laptop. But – importantly! – there are no circumstances under which the laptop will ‘prefer’ to run at its rated ‘base clock’ of 1.6GHz. There’s no point. The CPU can adjust its clock rate anywhere from about 1GHz to 2.4GHz in fine-grained steps, and it chooses the exact clock rate that it needs to balance performance and energy efficiency.
Intel Ark has this to say about “Processor Base Frequency”:
The processor base frequency is the operating point where TDP is defined.
Nowhere does it say “this is the preferred frequency” or “this is the maximum” or “this is the most efficient point”. It’s just where the processor runs at its TDP. The TDP is chosen by Intel! The exact same CPU can be sold at two different TDPs, at two different clock rates, to two different markets (e.g. laptop and desktop).
TDP is ‘thermal design power’ – typically between 5W and 150W for modern Intel chips. Importantly, though, it’s an arbitrarily chosen number. For a laptop part, TDP is chosen to be smaller – say, 15W. For a desktop or server part, TDP is larger – 35-135W. TDP is important for manufacturers because it dictates how big a cooling solution is needed. If they have to move a ‘nominal’ 15W from a laptop CPU instead of 135W for a many-core server CPU, they can use a smaller and lighter cooler.
Higher clock speeds and core counts require higher output power. TDP is arbitrarily selected to suit the end-user, but it doesn’t imply that the CPU is more or less capable than another. We know that our ‘1.6GHz’ CPU can run over 2.4GHz! It’s just that at the TDP, this is how fast we can run in steady state. The same CPU could run faster forever if you have a big enough cooler!
So, ‘base clock’ is a pointless figure now. Intel and the machine manufacturers publish it, but it’s more like “under these circumstances (workload, ambient temperature and heatsink efficiency), we can run this CPU at this clock rate indefinitely”.
The computer manufacturer thus has a big impact in how fast the CPU will run, because they design the cooling system. A too-small cooling system (e.g. Macbook Air 11” or 2015 Macbook) will constrain CPU performance simply because under load, the CPU will heat up and the clock speed will need to be reduced. A too-small cooling system is great for the manufacturer (less weight and volume leads to a smaller, lighter laptop) but you’re trading off CPU performance. Cooling efficiency is never reported!
For CPUs in the same series and with the same nominal TDP, there might be advantages to the faster ones. They’re sold as faster for the same rated TDP, and conversely they might run slightly cooler at the same clock rate. Given that the difference in clock rate is usually tiny (10%) and the price difference can be huge (hundreds of dollars) there’s rarely any point in buying the faster parts.
All of this is wrapped up in the GHz figure – the one the consumer looks at – but it’s no guarantee that performance is actually better. A laptop with a high clock rate, high TDP CPU might perform worse than one with a lower clock rate if the cooling is inadequate.
There’s an interesting comparison to be made between the 2016 Macbook Pros. The ‘Escape Edition’ has a 2.0GHz CPU, while the ‘Touch Bar’ model has a 2.9GHz CPU. On the outside, the machines look identical (except for the Touch Bar). Inside, the differences are tremendous. It’s a completely different design. Notably, the Escape Edition has a single CPU fan, while Touch Bar has two fans and bigger heatsinks.
At the time of writing, the Geekbench single-core benchmarks show:
The Touch Bar model has a 45% faster base clock. We’re testing a CPU-bound workload. We would expect it to get close to a 45% increase in performance. In reality, it only gets a 4.5% increase. The clock rate does not tell the full story!
The Escape Edition CPU has a maximum Turbo speed of 3.1GHz, while the Touch Bar CPU has a maximum Turbo speed of 3.3GHz – a 6.5% increase. This more closely explains the difference in benchmark results!
Better yet, the 1.2GHz Macbook scores 3003. That’s 80% of the performance of the Escape Edition with 41% of the base clock rate.