Intel describes a new approach in a patent: combining individual cores into so-called “super cores” using software. This should significantly increase single-core performance without sacrificing efficiency too much.
While the performance of graphics cards can be increased comparatively easily by installing more computing units, the situation with processors is more difficult. Although more cores help for multicore tasks, many applications require a high single-core performance. In turn, this can only be achieved by disproportionately larger derailleurs or higher clock rates. This is at the expense of costs and energy consumption. However, Intel now wants to have found a solution for this dilemma – so-called “super cores”.
Simply Merge Intel patent aims to combine cores into “super cores” using software
In a patent filed by Intel in 2023 and published this July, the company describes so-called “software defined supercores” (SDC) – i.e. “software-defined supercores”. With this approach, two or more cores are to be combined as a super core that processes the instructions of a single thread. Intel wants to make the individual computing units look like a single core to the outside – and thus significantly increase the single-core performance.
Intel wants to combine individual cores into "super cores" via software - and thus significantly increase single-thread performance.
According to the patent specification, the collaboration should take place quickly and efficiently and work with a wide variety of programs – for example with pre-compiled legacy software, newly compiled applications or tools interpreted in real time. All that is required for this is an “SDC interface“, which supplements each core, for example, with improved memory communication. As a result, neighboring cores should probably exchange data quickly and thus enable cooperation.
As usual for such patents, Intel’s explanations of the new super cores are unfortunately very vague. Therefore, it remains unclear whether the company is actually planning a corresponding product and whether the approach is profitable at all. In particular, the computational and energy efficiency of the approach would be interesting. Because here the prediction of upcoming decisions in the program code carried out by the processor, which is only usually true, should limit. This is already one of the most complex topics of CPU development today. And with the cooperation of several cores, this task would probably have to be done much more reliably, so that the cores do not have to discard their work afterwards and start all over again.
What might Intel’s Superkern approach look like in detail – and what problems do you expect? Use the comment function and let us know what you think. However, please observe the forum rules when commenting.
