ASUS ROG Zephyrus G14 (2022) Review: 8 GB RTX 3080 in a 1.7 kg AI Laptop
Who is this laptop for?
The ASUS ROG Zephyrus G14 (2022) is the most portable 8 GB AI laptop on the used market. Where every other 8 GB machine we recommend is a 2.3–2.9 kg desktop replacement, the G14 squeezes an RTX 3080 Laptop GPU with 8 GB of GDDR6 and a Ryzen 9 6900HS into a 1.7 kg, 14-inch chassis you can actually carry every day. For anyone who wants real CUDA compute — SDXL, FLUX, 13B LLMs on GPU — without lugging a workstation, this is the one to look at.
Students (Budget: £650–£900)
This is an excellent capstone or research machine. 8 GB VRAM runs Stable Diffusion XL comfortably and most 7B/13B LLMs on GPU, so coursework in computer vision or generative AI is genuinely practical. The 14” form factor fits in a backpack and lasts on battery for lecture-hall work in a way no gaming brick does. It costs more used than an integrated-graphics ThinkPad, but you get a real GPU — see the ThinkPad T14 Gen 3 if your budget can’t stretch and you only need CPU inference.
ML Engineers & Data Scientists
A strong portable inference box. The RTX 3080 Laptop is a fast 8 GB consumer GPU — quicker than the workstation A4000 8 GB at raw diffusion throughput thanks to higher clocks and more aggressive power. SDXL, ComfyUI graphs, FLUX.1 with quantisation, and 13B LLMs in Q4 all run well. The limit is the same 8 GB ceiling every consumer card hits: full-precision FLUX and serious fine-tuning need 16 GB — for that, the ThinkPad P15 Gen 2 is the used-market answer.
Small Teams & Startups
If your team needs a machine that does demos in the meeting room and generates images at the desk, the G14 is uniquely flexible. It is not ISV-certified like a ZBook or ThinkPad P, and the consumer GPU lacks ECC, but for prototyping and inference it punches far above its weight and size. Carry the 240W charger for sustained GPU work.
What can it actually run?
| Task | Works? | Notes |
|---|---|---|
| GitHub Copilot / Cursor AI | ✅ Yes | API-based, runs perfectly |
| Whisper transcription (local) | ✅ Yes | large-v3 at ~1.3× realtime (GPU-accelerated) |
| Ollama 7B | ✅ Yes | Fully GPU-resident. ~45–55 tok/s (estimated) |
| Ollama 13B | ✅ Yes | Q4 fits in 8 GB at moderate context. ~18–24 tok/s (estimated) |
| Stable Diffusion XL | ✅ Yes | ~10–16s per 1024×1024 image at 20 steps |
| ComfyUI / FLUX.1 | ⚠️ Tight | FLUX runs quantised (Q8/Q4) on 8 GB; full precision will not |
| Fine-tuning (QLoRA 7B) | ⚠️ Tight | Small-batch 7B QLoRA only; 13B and SDXL LoRA want 16 GB |
Key:
- ✅ Yes — works well
- ⚠️ Possible but slow — usable with patience
- ❌ No — hardware limitation prevents this
Full Specifications
| Component | Specification |
|---|---|
| CPU | AMD Ryzen 9 6900HS (8C/16T) |
| CPU Generation | AMD Zen 3+ (Rembrandt, 6 nm, 2022) |
| RAM | 32 GB DDR5-4800 (16 GB soldered + 1× SO-DIMM) |
| Storage | 1 TB NVMe Gen 4 (1× M.2 2280) |
| GPU | NVIDIA GeForce RTX 3080 Laptop (8 GB GDDR6, 105W) |
| VRAM | 8 GB GDDR6 (dedicated) |
| Display | 14” 2560×1600 IPS, 120 Hz (config-dependent) |
| Battery | 76 Wh |
| Weight | 1.72 kg |
| TDP | 35W CPU + up to 105W GPU (with Dynamic Boost) |
| AI Score | 79/100 |
Mobile vs desktop: the 8 GB you actually get
The badge says “RTX 3080” — the same name as a desktop card that shipped with 10 GB or 12 GB. The RTX 3080 Laptop GPU in the G14 has 8 GB, a different GA104 die, and a 105W power limit versus 320W+ on the desktop. For AI this distinction is everything: VRAM is the hard wall that decides whether a model loads at all. 8 GB is a genuinely useful figure — it doubles a 4 GB entry card and clears the bar for SDXL and quantised FLUX — but it is not the 10–12 GB a “3080” buyer might assume. Always read the laptop VRAM spec, never the desktop name. The full reasoning is in our guide to what VRAM is and why it matters for AI, and the quantisation maths is in GGUF quantization explained.
AI Performance in Practice
The RTX 3080 Laptop is one of the faster 8 GB GPUs you can buy used. With 6144 CUDA cores running at high clocks and Dynamic Boost pushing the GPU to 105W, it out-throughputs the workstation-class A4000 8 GB on diffusion: expect SDXL at roughly 10–16 seconds per 1024×1024 image at 20 steps. ComfyUI graphs run smoothly until you stack ControlNet plus a refiner plus high resolution, at which point 8 GB becomes the limit.
For LLMs, 7B models are fully GPU-resident and fast (an estimated 45–55 tok/s in Q4), and 13B fits in Q4 at moderate context — see the model-by-model breakdown in our Ollama laptop requirements guide. FLUX.1 runs in quantised form (Q8 or Q4 via GGUF); full-precision FLUX needs the 16 GB you’ll only find on the P15 Gen 2.
Thermal behaviour
The G14’s biggest compromise is thermal. ASUS engineered a remarkable cooler for a 14” chassis — liquid metal on the CPU, a vapour chamber — but physics still applies: under sustained combined CPU+GPU load the GPU runs warm and clocks settle below the 105W peak. For burst generation (a batch of images, a transcription job) it holds up well; for hour-long training runs a thicker chassis like the HP ZBook Fury 15 G8 sustains more. Fan noise at full tilt is loud, as on any thin gaming laptop.
Battery life under AI load
The 76 Wh battery is large for the size and gives genuinely good general-use longevity (6–9 hours light work) — a rarity among GPU laptops. But AI load is brutal: sustained GPU inference drains it in about 60–80 minutes, and the GPU throttles hard on battery anyway. Run AI work on the 240W adapter; use battery for coding and API-based tools.
What to Check Before Buying (Used)
Confirm the RTX 3080 8 GB — not the 3060 6 GB variant The 2022 G14 (GA402) shipped with RTX 3060 (6 GB), 3070 Ti (8 GB) and 3080 (8 GB) options. For AI you want 8 GB. Verify “RTX 3080 Laptop” or “3070 Ti Laptop” and 8 GB in GPU-Z and insist on a screenshot — a 6 GB unit is a meaningfully weaker AI machine for a similar asking price.
RAM configuration (partly soldered) The G14 has 16 GB soldered plus one SO-DIMM slot. Confirm the total (32 GB is the useful target) and remember you can only upgrade the single slot — there is no path past 32 GB.
Thermal paste and fan health These run hot and are 3+ years old. Ask whether the paste/liquid metal has been serviced, and listen for fan rattle. Run a 15-minute FurMark stress test if you can and watch for thermal throttling below ~90°C.
Battery wear and charger
Confirm the original 240W barrel charger is included (USB-C PD will not sustain GPU load). Run powercfg /batteryreport — a well-used G14 may be at 80% design capacity.
Where to Buy in the UK
Back Market UK — The most reliable graded source for G14 units, typically £700–£900 with a 12-month warranty. Filter for the RTX 3080 / 8 GB config explicitly.
Laptops Direct — Occasionally lists refurbished G14s and open-box units from £680. Verify the GPU tier — 6 GB 3060 models are common and cheaper for a reason.
eBay UK — Best prices (£650–£850) and the widest selection, often from creators upgrading. Always demand a GPU-Z screenshot confirming the 8 GB GPU and check battery health photos.
What to avoid: Any G14 listing that says only “RTX 3080” without “Laptop” and a VRAM figure, or that omits the GPU tier entirely. A 6 GB 3060 G14 is a different, cheaper laptop.
Verdict
AI Score: 79/100 — Pro AI
The ASUS ROG Zephyrus G14 (2022) is the portable 8 GB option in our line-up — the only machine that gives you a fast CUDA GPU, SDXL, quantised FLUX and 13B LLMs in a chassis you can carry to a café. The RTX 3080 Laptop is quick, the 32 GB DDR5 and 1 TB Gen 4 SSD are well matched, and at £650–£900 used it is far cheaper than any new 8 GB ultraportable.
The trade-offs are thermal and VRAM. The thin chassis throttles on long sustained loads, and 8 GB still can’t do full-precision FLUX or comfortable fine-tuning. But for portable inference and image generation, nothing else used is this capable at this weight.
Buy if: You want real GPU AI — SDXL, FLUX, 13B LLMs — in a laptop you carry daily, and you accept thermal limits on long runs.
Don’t buy if: You need full-precision FLUX or fine-tuning (get the ThinkPad P15 Gen 2, 16 GB) or you want sustained workstation thermals (the HP ZBook Fury 15 G8 holds clocks longer). Compare the whole dGPU range in our best used laptops for local LLMs and best used laptops for Stable Diffusion roundups.