Processor designers come up with new tricks to get the most out of their new architectures, one of which is to increase the clock speed of a single core above the common limit of the entire CPU. This is called a single-core boost and we are going to explain in this post what it consists of and what advantages it brings to your PCs.
The most veteran will remember the years in which processors were single-core and could not run more than one thread, it was the time when the race was for the largest amount of MHz first GHz later, until they could not go up more for physical limitations and had to move to multi-core.
A feature that we are going to see in the new CPUs is the single-core boost, which consists in that a single core of a multicore processor reaches a higher clock speed than that of the boost for several cores, all thanks to being able to disconnect the rest of the CPU cores, lower their clock speed or cause them to not be able to reach the boost speed.
Single core boost, redistributing CPU power
The clock speed that a processor can reach depends on the amount of energy available for it, but CPUs have traps to better optimize energy consumption although most are based on being able to disconnect unused parts of the processor always so that they do not consume energy while doing nothing.
At the kernel level, many designs that can work with multiple threads tend to execute certain instructions taking advantage of the duplication of the unit of certain parts of the control unit to avoid contention and increase performance, but in other designs this is not the case and makes the disable SMT or Hyperthreading in order to increase the clock speed.
Under the same principles it is possible to completely deactivate all the cores except one and even remove the SMT or Hypethreading from it so that the clock speed of a single core is as high as possible.
What does it add to the total processor performance?
There are many benchmarks that measure the performance of a single CPU core, these benchmarks are not used to evaluate the absolute performance of the CPU and are not based on realistic scenarios either. They simply serve us in a comparative way to know how one architecture has evolved with respect to another, comparing core by core.
In the programs that exist in the market, because for years that consumers have available several cores, most of the programs are designed to work in parallel most of the time, but it must be taken into account that the programs have a part that can be parallelized and another that cannot.
If we pay attention to Amdahl’s Law, the execution time of each of the parallelizable parts of the code will decrease by adding more cores, but there is a part of the code that is serial and therefore cannot be executed in parallel, that part of the programs depends on the speed of a single core.
If we turn off all the cores except one, then the code that works in parallel will be negatively affected, but if we maintain the base speed in all the cores and apply a boost or a slight advantage in one of them then the part of the code that runs serially within the program will be sped up, thereby increasing system performance.
Is single-core boost the same as big.LITTLE?
Although it may remind us of the big.LITTLE concept of certain low-power cores, it is not the same, since in this concept the idea is to use a low-power core to perform certain instructions, which are simple enough that it is impossible to optimize it. more in terms of consumption within a core, so the use of a simpler core is recommended to reduce energy consumption.
In “big.LITTLE” the clock speed of the cores is not increased or decreased but rather the core that executes said program or part of the program is changed, so in a processor with this type of design we have a heterogeneous formation of cores , while the single-core-centered speed boost can occur in both homogeneous and heterogeneous sets of nuclei.
However, since the clock speed is directly tied to the voltage in the power consumption, we can safely conclude that raising the clock speed of a core as opposed to the “bit.LITTLE” is not a way of saving processor power consumption.
