The Asus ROG Zephyrus G14 (2026) is a hardware marvel — but for those of us in MLOps, cloud computing, or deep learning, Windows is often just the "bootloader" for our real work in Linux. Running Ubuntu on a machine with a brand-new Blackwell RTX 5060 and specialized ASUS hardware requires more than a default install. This guide covers the entire journey.
A Windows installation already on the G14, a USB drive (8GB+), and about 2 hours. The final setup gives you a full dual-boot with GPU switching and fan control.
The Windows Pre-Flight Check
Before leaving Windows, document the hardware to ensure your Linux environment matches the silicon's capabilities.
1. Verifying System Architecture
echo $env:PROCESSOR_ARCHITECTURE
# Output: AMD64 → confirms 64-bit x86 architecture
2. Identifying the Blackwell GPU
The RTX 5060 is based on the Blackwell architecture (GB206 die) and supports FP4 precision — a major leap for AI/ML inference efficiency. Before partitioning, use GPU-Z in Windows to verify the hardware ID and ensure the "Standard" NV driver is active.
FP4 (4-bit floating point) is Blackwell's headline AI feature — it roughly doubles throughput over FP8 for quantised inference workloads like LLMs and vision transformers.
Preparing Ubuntu 24.04 "Noble Numbat"
BIOS Adjustments (Crucial)
Restart and press F2 to enter BIOS. Two settings matter:
-
1Disable Secure Boot — community drivers like
asusctloften struggle with Secure Boot signature enforcement. -
2Set Graphics to "Standard" or "Hybrid" — leaving it on "Eco" mode can cause the Ubuntu installer to miss the NVIDIA card entirely.
Partitioning
Use Disk Management in Windows to shrink your C: drive. Allocate at least 200
GB for Ubuntu — enough for heavy Docker images and MLOps datasets without constantly running out of space
mid-training.
The Linux Installation
I chose Ubuntu 24.04.4 LTS for stability over the short-term release cycle.
If your screen freezes during USB boot, press e at the GRUB menu and append nomodeset
to the linux boot line. This forces software rendering past the NVIDIA init hang.
During installation, always check "Install third-party software for graphics and Wi-Fi" — this pulls in the initial NVIDIA proprietary drivers automatically.
Installing ASUS Linux Tools from Source
On a stock Ubuntu install, you lose control over fan curves, keyboard LEDs, and power profiles. The old PPAs are
deprecated for 24.04, so we build asusctl and supergfxctl from source.
1. Install the Rust Environment
asusctl is written in Rust for performance and memory safety.
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source $HOME/.cargo/env
2. Install Build Dependencies
sudo apt update
sudo apt install -y libdbus-1-dev libudev-dev libfontconfig-dev \
libpango1.0-dev libglib2.0-dev libxml2-dev libhandy-1-dev \
libseat-dev git curl build-essential libssl-dev pkg-config
3. Compile and Install
# asusctl — fan curves, keyboard LEDs, charge limit
git clone https://github.com/asus-linux/asusctl.git
cd asusctl && make && sudo make install
# supergfxctl — GPU mode switching
git clone https://github.com/asus-linux/supergfxctl.git
cd ../supergfxctl && make && sudo make install
4. Enable Services
sudo systemctl enable --now asusd
sudo systemctl enable --now supergfxd
Hardware Optimization for MLOps
GPU Mode Switching
With supergfxctl installed, you can toggle how the RTX 5060 behaves — a massive
quality-of-life upgrade depending on context:
| Mode | Behaviour | Best for |
|---|---|---|
| Hybrid | iGPU handles desktop, RTX activates on demand | Daily use — good balance of battery and compute |
| Integrated | RTX 5060 fully disabled | Classes at Northeastern, note-taking, long meetings |
| Dedicated | Everything routes through RTX 5060 | Model training, CUDA workloads, benchmarking |
Battery Health
Staying plugged in at the library for hours at a time accelerates battery degradation. Capping charge at 80% meaningfully extends long-term battery health — especially important given how the G14's slim form factor limits thermal headroom:
# Limit charge to 80% — set once, persists across reboots
asusctl -c 80
Final Verification
# OS architecture — should output x86_64
uname -m
# x86_64
# Confirm Blackwell compute capability
nvidia-smi -q | grep "Compute Capability"
# Compute Capability: 10.0 ← Blackwell confirmed
Setting up a 2026 G14 on Ubuntu 24.04 isn't just "installing an OS" — it's fine-tuning a machine that's
genuinely capable of running production-grade AI workloads anywhere. With Blackwell's FP4
support, precise power management via asusctl, and GPU switching from
supergfxctl, this setup is everything a Master's student in AI needs without the datacenter.