In a processor, all the instructions do not have the same energy cost, but if we compare one architecture to another, we see that the most complex ones execute very simple instructions with a much higher energy cost than those which are simpler. They don’t have the computing power of complex cores, but a comparison based on power versus power consumption turns it around.
There are times when our PC is not running extremely complex programs, it is at these times that processor cores with a better ratio of power to consumption are needed. A more powerful core will not make them go excessively faster and it will consume more, hence the need for a hybrid system.
The patent for the AMD big-LITTLE system
A few hours ago a new patent issued to AMD appeared, titled TASK TRANSITION METHOD BETWEEN HETEROGENEOUS PROCESSORS that in Spanish we could title like: Method of transition of tasks between heterogeneous processors. The patent refers to the implementation of a BIG-LITTLE system by AMD, like ARM processors and their recent DynamicQ system, since it is a system capable of changing the type of core to be used on the fly.
We had already seen the concept of big-LITTLE on PC in the case of Intel Lakefield and it will be used in Alder Lake from now on. In exchange for AMD we know that it will be released in Zen 5, we will therefore have to wait for the deployment of the TSMC 3 nm node on a PC. Considering the fact that we will see the 5nm jump next year, it can be said that for this patent to become a reality there is still some time.
And what exactly does the patent say?
The patent does not explain how at a descriptive level, but it does specify what it is. The idea is to have a series of high performance and high power cores and other low performance and low power cores that can move tasks in real time.
That is, the state of one processor is transferred to another processor. In this way, the two cores distribute the work on the code, the lightest processor can execute the instructions of the code, which in the more complex processor is not an advantage in terms of speed, but a disadvantage in terms of consumption.
In AMD’s patent system, the division of labor is not done at a thread level for each core, but rather that the two kernel types are in tandem to execute the code for each thread. execution in the fastest and most efficient manner. We are therefore not dealing with a system that uses one type of processor or another in a switched manner.