It’s been nearly a decade since ARM introduced its big.LITTLE chip designs that allow a single processor to use two or more different CPU cores. The name comes from the idea that you could pair “big” high-performance cores with “LITTLE” energy-efficient cores to get the best of both worlds – speed when you need it and long battery life when you don’t.
This year Intel released its first chips that use a similar arrangement, although Intel achieves it a different way thanks to its Foveros 3D stacking technology.
Now it looks like AMD may be developing its own technology to pair two different types of CPU cores on a single chip.
A recently published patent application filed by AMD describes a “heterogenous processor system” that allows a two processors to be used together including a higher-performance processor and a second processor that “is configured to execute an instruction thread by consuming less power and with lower performance than the first processor.”
Keep in mind that companies apply for patents on technology that never sees the light of day all the time, so there’s no guarantee AMD will follow this path. But the patent application suggests that the company is at least considering building heterogenous computers that will balance performance and power consumption by combining different types of CPU technology onto a single chip.
The patent application was originally filed in 2017, but it was not published until June 30, 2020 and recently spotted by @Underfox.
If AMD does follow this path, it wouldn’t be the company’s first foray into heterogenous computing. In 2014 the company was touting the ability of its processors to leverage CPU and GPU functions simultaneously for accelerated performance. when running certain tasks… although the company’s chips at the time still generally lagged behind rival Intel’s processors when it came to raw CPU horsepower.
Over the past few years, AMD has largely caught up to Intel and many of the company’s recent chips actually offer better multi-threaded performance than Intel’s closest equivalents (although Intel still holds the edge in single-threaded performance).
via Tom’s Hardware
Is it possible that they want to add GPU computational power to help CPUs in SoCs as it is done in supercomputers?
If that is the case is Intel – and Nvidia -, and not ARM the target, as Intel GPUs are far from AMD and Nvidia ones, and this GPU+CPU design can beat NVIDIA GPU computing price/performance,
It’s an interesting concept. But I think ARM is still king when it comes low power devices with long battery life. I’m holding out for an ARM powered Macbook next year.
Apple silicon it a custom ARM based SoC. It is not safe to expect the same chipset will be used on laptops that Apple uses on the iPhone and the iPad. It is also not safe to expect similar battery life.
Not sure how AMD is going to do this.
Maybe they will use their InfinityFabric method, by splitting the High-Speed and Low-Energy chillers.
For instance, one 4-core chiplet, No hyperthreading, with the best binned fabrication (eg +5nm). And it will use less voltage, have fewer cache, and run at low frequencies (eg 0.5 – 2.0GHz).
The other 4-core chiplet, Full SMT hyperthreading, with only average binned fabrication (eg 7nm). And it will use higher voltage, have much larger cache, and run at high frequency (eg 2.0GHz – 4.0GHz).
Together they would make an 8-core processor, that would be decent enough for High-end Desktop Users. But dynamically switching they could perhaps have 8-core performance in a 4-core power-drain/battery life.
Though Intel certainly has the headstart on this front, with their very small/efficient Atom cores, and their very performance/large Core-i cores. Too bad they will fail due to being stuck on older wafer lithography.