- Intel fuses the physical CPU nuclei in a single virtual design of super nuclei
- The fused nuclei execute parallel instructions before reorder to improve performance
- The approach is directed to a greater efficiency of a single subprocess without expanding the size of the nucleus
Intel has presented a patent for what calls Software Defined Super Nores, a technology that fuses two or more CPU physical nuclei in a single virtual “super core”.
For the operating system, the fused nuclei appear as a unit, but the instructions are divided and executed in parallel before being rearranged, with the aim of improving the performance of a single thread without the high costs of building larger processors.
This approach resembles the oldest concepts of “inverse hyperingido” of the era of Pentium 4, which suggests that Intel is reviewing past experiments with modern refinements.
Efficiency and scale balance
The idea behind this approach is to improve the performance of a single thread avoiding the greatest energy demands associated with faster clock speeds or broader nuclei.
Intel design distributes workloads in multiple nuclei through shared memory and synchronization modules.
If the mechanism works, the company expects profits in the performance per watt, which allows processors to alternate between normal modes and super nucleus.
The observers have compared Intel’s idea with the old multiple subprocesses of AMD, although the methods differ.
AMD divides the nuclei into modules, while Intel’s proposal combines whole nuclei under software control.
Some also link the patent with the Royall Core canceled of Intel, which, according to the reports, pursued high watch instructions, but felt little practical for manufacturing.
By reliving such strategies, Intel seems to be looking for alternatives to gross force design expansions.
However, the lack of measured data makes it impossible to know if this could rival the fastest CPU designs in the market.
The patent describes a small synchronization module within each nucleus, supported by a reserved memory region called worm hole address space.
These handle registration transfers, order and data flow to guarantee the integrity of the instruction.
On the software side, compilers or binary instrumentation divide the code into manageable blocks when inserting flow control commands.
Operating systems must decide when a workload benefits from super core mode, a requirement that could complicate programming and compatibility.
Without a wide support of both hardware and software, the design runs the risk of becoming an unused feature.
Intel’s documentation does not stimulate clear performance profits, it only suggests that two narrower nuclei could address the capacity of a broader nucleus under certain conditions.
Technology might interest researchers who explore specialized workloads, including scenarios in which a mining CPU could seek improved efficiency in single -thread tasks.
However, for general computer science, the lack of proven reference points leaves the uncertain promise, and if this really creates the best CPU for demanding workloads it remains an open question.
Via Hardware Toms
