I spent a week switching between linux and linux-cachyos on my Ryzen 7 7730U laptop. Same hardware, same desktop, same apps — only the kernel changed. The differences were subtle but real: slightly smoother window dragging, faster app launches, and one unexpected finding about battery life that I didn’t anticipate.

Architecture Diagram

The Compilation Pipeline Difference

This is where the real divergence happens. Both kernels start from the same Linus Torvalds source tree, but CachyOS applies a fundamentally different build process.

Architecture Diagram

Three Compiler Optimizations Explained

Standard Arch compiles each kernel source file independently. ThinLTO defers optimization to link time, letting the compiler see the entire kernel as one program.

What it does: Cross-module inlining — a function in kernel/sched/ can be inlined into drivers/gpu/drm/ if it’s called frequently enough. The compiler eliminates dead code paths across module boundaries.

Typical gain: 1–3% throughput improvement. Build time increases by ~20%, RAM usage by ~2GB during linking.

2. PGO (Profile-Guided Optimization / AutoFDO)

CachyOS uses AutoFDO — a variant of PGO that collects hardware performance counter data from real kernel execution.

What it does: The compiler knows which code paths are hot (frequently executed) and which are cold. It inlines hot functions aggressively and reorders basic blocks for branch prediction.

Typical gain: 2–5% on workloads that stress kernel paths (I/O, scheduling, networking).

3. BOLT (Binary Optimization and Layout Tool)

BOLT operates after the kernel is compiled and linked. It reads a runtime profile and rearranges functions in vmlinux for better instruction cache locality.

What it does: Hot functions are placed close together in memory. Cold functions move to separate pages. This reduces I-cache misses.

Typical gain: 2–5% on I-cache-sensitive workloads. Meta reports significant QPS improvements in production.

Architecture Diagram

The Scheduler Menu

This is the biggest practical difference for desktop users. Standard Arch ships EEVDF (Earliest Eligible Virtual Deadline First) — the upstream default. CachyOS offers four choices:

Scheduler Best For How It Works
BORE Gaming, desktop Burst-Oriented Response Enhancer — prioritizes interactive tasks, minimizes latency for UI threads
EEVDF General purpose Upstream default — fair time-sharing with deadline-based prioritization
BMQ Throughput workloads BitMap Queue — optimized for batch processing, compiles, data pipelines
sched-ext (SCX) Custom/experimental BPF-based — load user-space schedulers at runtime without kernel recompile
Architecture Diagram

sched-ext: The Game Changer

sched-ext (merged in Linux 6.12) lets you write CPU schedulers as BPF programs. Load them, unload them, switch at runtime — no kernel recompile needed.

CachyOS integrates this with GUI tools (scxctl and SCX Manager) and pre-built profiles:

Profile What It Does
Gaming Prioritize game threads, reduce stuttering, lock to performance cores
Power Saving Aggressive idle states, park idle cores, reduce frequency
Low Latency Audio production, minimize scheduling jitter
Server Throughput-first, maximize parallelism

The tradeoff: sched-ext schedulers are specialized. Using the wrong one for your workload can hurt performance. A gaming scheduler on a compilation-heavy system will starve build threads.

What About Battery Life?

This surprised me. On my Ryzen 7 7730U:

Metric Arch linux CachyOS linux-cachyos
Idle power ~4.2W ~3.8W
Light web browsing ~6.1W ~5.7W
Video playback ~5.8W ~5.4W
Full CPU stress ~28W ~28W

CachyOS’s BOLT-optimized kernel uses less power at idle — the instruction cache is more efficient.

The CPU wastes fewer cycles fetching instructions. Under full load, both hit the same thermal wall.

Kernel Config Differences

Beyond compiler flags, CachyOS tweaks kernel configuration:

Same config:

Config Both
PREEMPT PREEMPT_DYNAMIC
HZ 300
MGLRU Enabled

CachyOS additions:

Config Arch CachyOS
ZSTD Level 1 Level 3 (smaller image)
NVIDIA Upstream Pre-patched
NTFS ntfs3 driver ntfs3 + perf fixes
BBRv3 Available Enabled by default

The Compounding Effect

Each optimization alone is small. Together, they compound:

Optimization Isolated Gain In CachyOS Stack
ThinLTO 1–3% Cross-module inlining across entire kernel
PGO (AutoFDO) 2–5% Hot paths inlined, cold paths deprioritized
BOLT 2–5% I-cache miss reduction on optimized code
x86-64-v3 targeting 5–15% (SIMD) AVX2 instructions for crypto, compression
BORE scheduler 5–10% (latency) Interactive tasks get priority

Combined: Gains don’t stack linearly — real-world improvement is typically 5–15% on desktop, 5–10% on server throughput. Most visible in I/O-heavy operations and scheduling-sensitive tasks.

When to Use Which

Use CachyOS for:

Scenario Why
Desktop gaming BORE scheduler, lowest latency
Audio production sched-ext low-latency profile
Compiles / dev work BMQ scheduler, max throughput
Battery life BOLT saves idle power

Use Arch linux for:

Scenario Why
Server / production Proven, minimal variables
Maximum stability Fewer patches, fewer bugs
Hardware compatibility Most tested

How to Switch

# Check which kernel you're running
uname -r

# Install CachyOS kernel (on Arch)
sudo pacman -S linux-cachyos

# Or with BORE scheduler
sudo pacman -S linux-cachyos-bore

# Or with sched-ext support
sudo pacman -S linux-cachyos-scx

# Update bootloader (GRUB)
sudo grub-mkconfig -o /boot/grub/grub.cfg

# For Limine (CachyOS default)
sudo limine-snapper-sync

To switch back:

sudo pacman -Rns linux-cachyos
sudo pacman -S linux
sudo grub-mkconfig -o /boot/grub/grub.cfg

The Bottom Line

CachyOS’s kernel isn’t just “Arch with patches.” It’s a multi-pass optimized build of the same source code, with scheduler choices that standard Arch doesn’t offer. The 1.4MB size difference buys you ThinLTO cross-module inlining, PGO-guided branch prediction, BOLT code layout, and AVX2 instruction targeting.

For most users, the difference is subtle — slightly snappier UI, slightly faster app launches. For power users pushing audio latency, gaming frame times, or server throughput, the compounding optimizations are measurable and meaningful.

Tested on AMD Ryzen 7 7730U (Barcelo, Vega 7 iGPU), 16GB DDR5. CachyOS July 2026. linux-cachyos 7.1.4-1 vs linux 7.1.4-arch1-1. Mesa 24.1.4. KDE Plasma 6.7.3. Power measurements via powertop --time=60 on battery.