AMD brings Capacity Aware Scheduling to Linux for its hybrid processors
On 15 November 2026 AMD posted a patch series so that the Linux scheduler can measure each core’s real capacity and place tasks correctly on its hybrid processors, which mix Zen 5 and compact Zen 5c cores. Understand what this changes for your workloads before the support lands in a future kernel.
15 November 2026. Phoronix reports that AMD has posted a new set of Linux kernel patches to enable Capacity Aware Scheduling on its heterogeneous and hybrid processors. Linux 7.0. The release that already introduced, earlier this year, a P-core/E-core hybrid routing overhaul co-designed with Intel. Linux 4.19. The kernel that first gained Energy Aware Scheduling for Arm’s big.LITTLE chips. Why it matters: the Linux scheduler assumes by default that every core is equal — an assumption that hybrid processors, now on the AMD side too, have made false.
The problem: a scheduler that believes all cores are equal
The Completely Fair Scheduler (CFS), Linux’s default scheduler, was built for symmetric SMP machines: every core is identical, and every task deserves an equal share of power. Under that model, placing a task on core 0 or core 7 changes nothing — they share the same frequency, the same microarchitecture, the same capacity.
The hybrid processor breaks that assumption. A chip can carry two distinct populations of cores: “big” cores, fast and hungry, and “small” or compact cores, slower but cheaper and denser. At Intel, these are the P-cores and E-cores of Alder Lake and its successors. At AMD, they are the classic Zen 5 cores and the Zen 5c cores — identical at the ISA level but more compact and lower-clocked. A scheduler that treats them as equivalent places a latency-sensitive job on a slow core, or wastes a fast core on background work — both outcomes are bad.
The central notion is capacity. In the kernel, each core is assigned a normalized capacity — 1024 by default on a symmetric system. A “big” core receives a higher capacity, a compact core a lower one. That metric is what lets the scheduler decide where to place a task, and when to migrate it.
The history: Arm led the way, Intel followed
The concept does not start with AMD. The first to industrialize it was Arm, whose mobile chips have combined big.LITTLE cores for years — one or two powerful cores surrounded by frugal ones. The kernel answered with Energy Aware Scheduling (EAS), merged around Linux 4.19, which integrates a capacity and energy model to place tasks on the right core while optimizing battery life.
Intel then imposed hybrid on x86 with Alder Lake in late 2021. Its answer was twofold: Thread Director, on-die hardware that tells the operating system which core type suits each task, and a progressive kernel-side adoption. More recently, Linux 7.0 shipped a hybrid routing overhaul co-designed with Intel for the Nova Lake platform, which we covered here — foreground tasks pinned to P-cores, background work relegated to E-cores, with measured battery gains.
It is worth separating two ideas that often blur together. Energy Aware Scheduling is an energy policy: it places work to minimize power draw, which on a phone means preferring the small cores whenever the big ones are not needed. Capacity Aware Scheduling is a throughput and latency policy: it makes sure the scheduler knows each core’s true capacity so that a latency-sensitive task never lands on a slow core and a fast core never idles under a queue of cheap work. A hybrid x86 laptop or a dense EPYC server cares mostly about the second — you do not tune a server for battery life — which is why AMD’s patch series targets capacity specifically, not energy.
AMD remained the missing link. Its hybrid chips already exist — the Ryzen AI 300 “Strix Point” mix Zen 5 and Zen 5c cores, and dense server EPYC parts rest on Zen 5c cores — but the capacity-aware scheduling support on the AMD side was not yet aligned with Arm’s and Intel’s.
What AMD’s patches bring
The series posted on 15 November 2026 aims precisely to close that gap. It gives the kernel the capacity model specific to AMD’s hybrid processors, so the scheduler can tell a Zen 5 core from a Zen 5c core and place each task accordingly.
Three behaviors follow from this. First, initial placement: a latency-sensitive task — audio processing, UI rendering, a game server — should land on a fast core, while background work — an indexer, a backup, a package upgrade — can settle for a compact core. Second, migration: as load shifts, the scheduler must know how to move a task off a saturated compact core onto a free fast core, and back, guided by real capacity rather than assumed equality. Third, load balancing across core populations: without a capacity model, the kernel can spread work blindly and create an imbalance where fast cores idle while compact cores saturate.
This is still a patch series posted to the mailing list — not code merged into a reference kernel. No merge timeline has been announced, and the support must pass community review before landing, at best, in a future merge window.
What this changes for you
For the owner of a Ryzen AI 300 or a future hybrid AMD chip, the stakes are direct. Today, without a capacity model, Linux can place interactive work on a Zen 5c core while leaving a Zen 5 core nearly idle — a latency degradation that averages hide but usage reveals. The coming support should make that placement predictable, on the model Intel achieved with the hybrid routing in Linux 7.0.
For operators of dense EPYC servers, the question is subtler. Zen 5c cores are not “worse” — they are more numerous per socket and optimized for throughput at a given energy envelope. A correct capacity model should therefore place parallel, latency-tolerant workloads on the compact cores and reserve the fast cores for latency-sensitive ones. That is exactly the logic Capacity Aware Scheduling formalizes.
Until then, the symptom is visible if you know where to look. On a hybrid AMD chip, interactive work can land on a Zen 5c core and surface as jitter — a lag spike that averages out in monitoring but is felt by the user, or a fast core idling while a compact core queues. There is no tunable that fixes it today; what is missing is the capacity model itself. That is why this patch series matters more than its size suggests: a modest change that alters where the kernel believes a task belongs.
The verdict
If you run a hybrid AMD chip on Linux today, change nothing for now — the patches are not merged yet, and forcing manual placement with taskset is a fragile stopgap that breaks at the first load change. Watch the upcoming merge windows instead to learn when AMD support becomes effective.
If you are sizing a fleet of dense EPYC servers, keep the underlying logic: a compact core’s capacity is not measured in frequency alone but in throughput per watt. Capacity Aware Scheduling is one instance of a broader rule — align scheduling with the real geometry of the silicon, not with the assumption of a universal core. When AMD support lands, distributions will ship it with the kernel, and your workloads will improve without you changing a single line of configuration.