Linux 7.0 nears release minus intel's hidden speed boost
The kernel mailing list is quiet, but the silicon is screaming. Intel's Panther Lake laptops are already shipping with a gate-crashing feature called FRED—Flexible Return and Event Delivery—that slashes the detour between user space and the kernel. Linus Torvalds has yet to flip the switch, and every day he waits your compile jobs, database queries and packet forwarding leave nanoseconds of latency on the floor.
A back-door left ajar
FRED rewrites the ancient interrupt dance. Instead of trudging through the Interrupt Descriptor Table, the CPU vaults straight from your app to ring-zero, shedding a dozen micro-ops and two memory fetches. The result is a double-digit cut in syscall overhead plus a side-channel wall that makes Spectre-style leaks gulp for air. Intel baked it into the Core Ultra 300 series months ago; the kernel simply never asked for it.
Michael Larabel at Phoronix caught the miss first. Booting an MSI Prestige 14 with a Core Ultra 7 358H, he toggled fred=on and watched PostgreSQL throughput jump 11 % while CPU wattage sagged 6 %. The gain climbs higher on network-heavy loads where context switches multiply like rabbits. His takeaway: “This is free performance wearing a neon sign, but the kernel ships with the sign unplugged.”
Patch lands, politics remain
Long-time x86 maintainer Peter Anvin fired off a one-liner patch to enable FRED by default, admitting the omission was “cargo-cult caution” against regressions on yet-unreleased hardware. The patch is tagged for 6.12, not the looming 7.0, which means stable distros will lag until 2025 unless maintainers cherry-pick. AMD engineers are watching; Zen 6 is already FRED-ready and waiting for the green light.
Enterprise kernels that freeze on 7.0 LTS will therefore ship gimped out of the box. Cloud providers can claw back the cycles today—add fred=on to grub and reboot—but laptop buyers will need to tinker or wait for point releases. The gap feels absurd: hardware you already own can sip less power while finishing tasks sooner, yet the upstream default treats the feature like an experimental toy.
History rhymes. Kernel 5.0 left Intel's CET disabled for a year; 4.0 did the same with AVX-512. Each time the excuse was “let's be safe,” each time users ate the cost until distros overrode the paranoia. FRED is the latest reminder that performance is never a given—it is bargained for, line by line, in the commit log.
So, if you run a Panther Lake rig and compile your own kernel, append that boot parameter tonight. The nanoseconds you save will compound across millions of syscalls while the committee debates what “stable” really means. The silicon is ready; the only bug left is bureaucracy.