While little is known about AMD’s Zen 6 CPUs, quite a lot has been leaking about Intel’s competing Nova Lake CPUs. Nova Lake could be an interesting and innovative generation thanks to new cores bringing back AVX‑512 and introducing APX for the first time. And they will be boosted by another new technology: FRED, which is already present in Panther Lake CPUs, and as it turns out, it brings notable benefits in certain applications.
The acronym (or perhaps backronym) FRED stands for Flexible Return and Event Delivery. It is a technology that has been in development for quite a long time, aiming—similarly to APX or the cancelled x86S—to streamline and modernize certain outdated and limiting parts of the x86 CPU architecture. In the case of FRED, the goal is to make switching between privilege levels more efficient and simpler (especially between the OS kernel and user space), which likely also applies to interrupt handling.
Thanks to the FRED extension, handling these security‑ and reliability‑critical situations can become more robust if software replaces older routines with the new code using instructions provided by FRED. Beyond just “cleaning up” legacy mechanisms, the goal was also to improve performance by reducing latency and overhead.
Unlike APX, FRED will not debut in Nova Lake—it is already included in Panther Lake processors, i.e., Core Ultra 300, which began shipping in laptops this year (and the low‑cost Wildcat Lake models may support it as well due to sharing the same architecture). Support has also already landed in the Linux kernel. Users of various distributions could therefore benefit from it relatively soon.
The Phoronix website has now examined the potential performance improvements and benchmarked them. In current Linux builds, FRED can be enabled (using the fred=on kernel parameter), but it is not active by default, so it is not normally used. Patches already exist to enable it by default.
Benchmark results suggest that FRED may typically help with disk I/O performance, especially with high‑performance NVMe SSDs capable of handling large numbers of IOPS. This is a situation where kernel calls cause frequent privilege switches between kernel and user space. Phoronix recorded noticeable performance gains in databases (SQLite is improved by only 3–4%, but PostgreSQL by up to 48%) and in the Flexible IO Tester benchmark (improvements up to 79%). Server workloads dependent on the networking subsystem may also see gains.
Conversely, web browsers (JavaScript), games, and graphics drivers are unlikely to be affected much. Put simply, improvements can be expected in similar areas as where performance drops were previously observed due to mitigations for security vulnerabilities revealed during the Meltdown, Spectre, and similar bug disclosures, as those were also largely tied to kernel system calls being used heavily.
A more niche but interesting area where the technology could be welcome is audio and music production workstations (in Phoronix tests, the Stargate Digital Audio Workstation 22.11.5 benchmark improved by 4–7%). These workloads could benefit from FRED reducing system‑call latency, which can shorten audio lag—an important factor for this type of software.
So while FRED does not improve performance across all software, the partial gains—combined with other enhancements such as APX—should add up and will help AMD and Intel CPUs keep pace with competing Arm‑based processors.
All processors are expected to support FRED in the future
The FRED extension will also be supported by future AMD processors, as both companies agreed on standardising it within the x86 System Advisory Board harmonization consortium. The technology therefore will not remain an Intel‑exclusive feature, though we do not yet know when it will appear in Ryzen and Epyc CPUs. Thanks to support from both major CPU vendors, FRED should hopefully also be utilized by the Windows operating system eventually.
English translation and edit by Jozef Dudáš
⠀
