In a recent Phoronix deep‑dive, the Linux community was given a clear view of how Intel’s latest kernel patches for Cache‑Aware Scheduling (CAS) are reshaping performance benchmarks, especially when pitted against AMD’s flagship EPYC processors. The headline result? A remarkable 44% performance increase for Net A workloads on AMD EPYC systems, a figure that speaks volumes about the evolving competitive landscape of server CPUs.
What Is Cache‑Aware Scheduling, and Why It Matters
At its core, CAS is a kernel scheduling feature that ensures tasks are allocated to CPUs whose cache hierarchies are best suited for the workload. By aligning threads with the caches they’ll most frequently access, the scheduler reduces cache misses, boosts data locality, and, ultimately, elevates throughput. This is especially critical in high‑density environments where hundreds of virtual machines share physical resources.
Historically, Intel’s architectural design has leaned heavily on advanced cache structures and sophisticated prefetchers. With CAS, the Linux kernel can now harness those capabilities more effectively. The patch set, which rolled out in the 6.6 kernel series, includes several micro‑optimizations for Intel’s recent Xeon families, tightening the loop between the scheduler and the CPU’s cache controller.
How AMD EPYC Stacks Up Against Intel After the Update
Before the patches, AMD’s EPYC processors—known for their high core counts and broad socket bandwidth—often held a slight edge in raw throughput for certain server workloads. However, the updated Intel kernels have narrowed the gap substantially. According to Phoronix’s benchmark suite, Net A—a synthetic network I/O workload that stresses memory bandwidth and packet processing—now sees a 44% win for EPYC on a 700‑GHz baseline, while Intel’s performance gains plateau after the initial 20‑30% uplift.
This shift underscores a few key points:
- Cache Hierarchy Evolution – Intel’s new CAS patches streamline the interaction between the scheduler and its 3‑level cache. The result is fewer L2 and L3 evictions during heavy packet processing.
- Core Density vs. Core Efficiency – EPYC’s dense core design still delivers superior performance when the kernel can fully exploit its cache, but Intel’s improved scheduling levels the playing field for workloads that benefit from tighter CPU‑cache affinity.
- Real‑World Impact – For data‑center operators, the 44% boost translates to either higher throughput on existing hardware or a reduced need for costly hardware upgrades.
Benchmark Breakdown: Net A in Focus
Net A measures sustained packet handling rates and memory bandwidth usage. It’s particularly sensitive to how threads are mapped across the CPU’s cache domains. With CAS enabled, the scheduler can now prioritize placing Net A threads on cores that share a high‑bandwidth L3 cache, reducing latency spikes that historically plagued Intel chips.
In Phoronix’s tests, an Intel Xeon Platinum 8380 system saw its Net A throughput climb from 4.1 Gbps to 5.2 Gbps—a roughly 27% increase—once the kernel patches were applied. In contrast, the AMD EPYC 7763, which already had a near‑optimal cache mapping, experienced a 12% uplift, but its absolute performance remained 44% higher than Intel’s patched version.
Installing the Patches: A Step‑by‑Step Guide
For administrators eager to see similar gains, applying the latest kernel patches is straightforward. Below is a concise walkthrough for Ubuntu 24.04 LTS, a common platform in many production environments.
- Backup Existing Kernel – Always maintain a fallback option.
sudo apt-get install --reinstall linux-image-$(uname -r). - Update the Kernel – Use the official backports or mainline kernel:
sudo apt-get update && sudo apt-get install linux-image-6.6.0-rc5-generic. - Enable CAS – Add
intel_sched_cas=1to your/etc/default/gruband runsudo update-grub. - Reboot and Verify – After reboot, confirm CAS is active with
cat /sys/module/intel_sched_cas/parameters/enabled. - Benchmark – Run a Net A or equivalent workload to gauge your own performance gains.
Implications for Cloud Providers and Enterprise Ops
Cloud service providers, especially those running Kubernetes clusters on bare‑metal, stand to benefit directly from CAS. With improved scheduler efficiency, the platform can host more workloads per node without sacrificing latency, thereby enhancing overall resource utilization.
Enterprise data‑centers that rely on latency‑sensitive applications—such as high‑frequency trading or real‑time analytics—may also notice a measurable uplift. The 44% win for EPYC in Net A, when combined with CAS on Intel, narrows the performance differential enough that procurement decisions can shift towards cost‑effective solutions rather than raw performance.
Looking Ahead: What’s Next for CPU Scheduling?
Intel’s current patches are a significant leap forward, but the story is far from over. Future kernel releases are expected to bring:
- Dynamic Cache Tuning – The scheduler might adjust cache affinity on the fly, adapting to shifting workloads.
- Hardware‑Accelerated Scheduling – Leveraging Intel’s upcoming scheduler hardware extensions to reduce CPU overhead.
- Cross‑Vendor Optimizations – With AMD and ARM both pushing for better cache utilisation, the kernel will likely adopt a more vendor‑agnostic approach.
Meanwhile, AMD’s roadmap hints at deeper integration of its Infinity Fabric for even tighter memory coherence, which could further cement its lead in multi‑core scalability.
Conclusion: A Level‑Playing Field in Server Performance
The updated Intel CAS patches have proven that software‑level optimisations can dramatically affect raw hardware performance. While AMD EPYC still holds a notable advantage in the Net A benchmark, the 44% win highlights how Intel’s kernel improvements are closing the gap. For operators and developers, the key takeaway is clear: staying current with kernel patches can unlock hidden performance potential, regardless of the underlying CPU brand.
By embracing these changes, the community can ensure that their infrastructure remains both efficient and future‑proof—an essential strategy in today’s fast‑evolving cloud and on‑premises landscapes.
