It’s become clear that hybrid cores — big.Little in ARM’s parlance — are going to be a feature of mainstream x86 CPUs as well. Intel’s Lakefield combines one Ice Lake “big core” with four Tremont “little” cores. Its upcoming Alder Lake platform will scale the solution up, with (rumored) up to eight low-power cores (Gracemont) and eight high-performance cores (Golden Cove).
More details on AMD's version of big.LITTLEhttps://t.co/OwsdshPV7G pic.twitter.com/T55kM5Yg1w
— Kepler (@Kepler_L2) June 12, 2021
A few signs have indicated AMD has its own plans to get into this game, and a new patent filing backs up that idea. AMD has applied for a patent describing methods by which one type of CPU would move work over to another type of CPU:
Figure 5 from AMD’s patent application.
According to the patent, CPUs would rely on core utilization metrics to determine when it was appropriate to move a workload from one type of CPU to the other. Proposed metrics include the amount of time the CPU has been working at maximum speed, the amount of time the CPU has been using maximum memory, average utilization over a period of time, and a more general category in which a workload is moved from one CPU to the other based on unspecified metrics related to the execution of the task.
When the CPU determines that a workload should move from CPU A to CPU B, the core currently performing the work (CPU A, in this case), is put into an idle or stalled state. The architecture state of CPU A is saved to memory and loaded by CPU B, which continues the process. AMD’s patent describes these shifts as bi-directional — the small core can shift work to the large, or vice-versa.
big.Little Won This Fight
Ten years ago, I asked Intel’s smartphone SoC designers why they were relying on DVFS — Dynamic Voltage and Frequency Scaling — to keep Atom’s power consumption competitive rather than big.Little. According to Intel’s architects, DVFS was competitive with what big.Little could deliver when one considered the silicon die space requirements and the overall power savings.
That may have been true at the time — Medfield was generally competitive with the midrange Cortex-A9 devices it was intended to compete against — but it doesn’t seem to be true today. At the time, Medfield was built on the 32nm process node, while the Cortex-A9 was generally built on 40nm or 28nm. Even if we count conservatively, TSMC and Intel have been through multiple full node shrinks. Intel’s 10nm technology is considered equivalent to TSMC’s 7nm node. The die size penalty for using a hybrid x86 core design is much smaller than it used to be.
The other reason we suspect big.Little may have won this fight is the changing nature of silicon design. The tuning required to run a 10nm CPU at high clocks may also result in a CPU core that doesn’t save as much power at low clocks as a low-power, purpose-built CPU core would.
The big question where AMD is concerned is whether the company will build a new from the ground-up CPU core, or if it will return to a design like Jaguar and use that as a basis for future products. While the CPU would need a refresh or update, it was a solid, balanced design when new. It typically takes a company such as AMD, Intel, or Apple 3-5 years to build a new core from scratch, so which avenue AMD goes may depend on when it started the project. We know it applied for this patent back in 2019, so the company has clearly been working on this idea for a little while.
We don’t expect to see hybrid cores from AMD in 2021, but 2022-2023 wouldn’t be unreasonable. Now that we know some of the power savings the Apple M1 delivers come courtesy of its OS scheduling, we expect both Intel and AMD to work with Microsoft in order to adopt similar tech. Hybrid cores won’t do much for x86 CPU power consumption at full load/clock, but they could reduce power consumption significantly.
When Joe Macri told folks in late 2020 that AMD had no near-term plans to introduce hybrid cores, he indicated one major stumbling block was a lack of scheduler support in Windows. We now know that Windows 11 is on the horizon and Alder Lake is prepping for launch. No one has officially said that this kind of hybrid scheduler will feature in Windows 11, but it’s virtually guaranteed to show up.
Now Read:
- AMD’s Steam CPU Share Skyrockets Over 12 Months
- GPU Shipments Soared in Q1 2021, to Nvidia’s Benefit
- CPU Manufacturers Are Pushing the Boundaries of CMOS and Starting to Pay For It
No comments:
Post a Comment