HP ZBook Studio G8 Review: A Quiet 6 GB CUDA Workstation for AI
Who is this laptop for?
The HP ZBook Studio G8 is a premium 15.6” mobile workstation — thinner, quieter and better built than most workstation laptops, with a real NVIDIA CUDA GPU inside. The version we recommend pairs the 8-core Intel Core i7-11800H with the RTX A3000 Laptop GPU and 6 GB of dedicated GDDR6 VRAM. That 6 GB is the threshold where Stable Diffusion XL and Ollama 13B become comfortable rather than a fight.
Students (Budget: £580–£800)
This is the upper end of a student budget, but if you need a quiet machine that also handles GPU AI, the ZBook earns it. SDXL runs at usable speeds, Ollama 7B flies with full GPU offload, and the chassis stays library-quiet under light load. If you only ever run CPU inference or API tools, save money with a ThinkPad T14 Gen 3 instead — the ZBook is overkill without the dGPU workloads.
ML Engineers & Data Scientists
A genuinely portable CUDA workstation. The RTX A3000 has 6 GB VRAM and 4096 CUDA cores — enough for SDXL, ComfyUI, GPU-accelerated notebooks and small LoRA fine-tuning runs. The i7-11800H sustains 45W, and 32 GB RAM (expandable to 64 GB) lets you load larger models on CPU alongside GPU work. The headline feature is acoustics: the ZBook is dramatically quieter than a Legion or Precision under the same load.
Small Teams & Startups
For a team that takes the laptop to meetings and client sites, the ZBook’s premium build, quiet cooling and HP enterprise support make it a better daily-carry than a gaming chassis. At £580–£800 used it costs far less than a new RTX workstation, while still giving prototyping-grade CUDA. It is a desk-and-travel machine, not a render farm.
What can it actually run?
| Task | Works? | Notes |
|---|---|---|
| GitHub Copilot / Cursor AI | ✅ Yes | API-based, runs perfectly |
| Whisper transcription (local) | ✅ Yes | ~6× realtime on base model (GPU-accelerated) |
| Ollama 7B (Llama 3, Mistral) | ✅ Yes | ~22–28 tok/s with full GPU offload (estimated) |
| Ollama 13B | ✅ Yes | Q4 fits in 6 GB. ~10–14 tok/s with GPU offload (estimated) |
| Stable Diffusion 1.5 | ✅ Yes | ~4–6s per 512×512 image at 20 steps |
| Stable Diffusion XL | ✅ Yes | ~18–28s per 1024×1024 image. 6 GB is the comfortable minimum. |
| ComfyUI / FLUX.1 | ⚠️ Tight | FLUX.1 needs aggressive quantisation at 6 GB. SDXL workflows are fine. |
| LoRA fine-tuning (small) | ✅ Yes | Small SD LoRAs and 7B QLoRA feasible with gradient checkpointing. |
Key:
- ✅ Yes — works well
- ⚠️ Possible but slow — usable with patience
- ❌ No — hardware limitation prevents this
Full Specifications
| Component | Specification |
|---|---|
| CPU | Intel Core i7-11800H (8C/16T, Tiger Lake H) |
| CPU Generation | Intel 11th Gen (Tiger Lake H, 2021) |
| RAM | 32 GB DDR4-3200 (2× SO-DIMM, upgradeable to 64 GB) |
| Storage | 512 GB NVMe Gen 4 (M.2 2280) |
| GPU | NVIDIA RTX A3000 Laptop (6 GB GDDR6, 4096 CUDA cores) |
| VRAM | 6 GB GDDR6 (dedicated) |
| Display | 15.6” 1920×1080 IPS or 4K DreamColor (config-dependent) |
| Battery | 83 Wh |
| Weight | 1.79 kg |
| TDP | 45W CPU + 80W GPU (configurable) |
| AI Score | 69/100 |
Mobile vs desktop: what “RTX A3000” actually means
This is where buyers get caught. The desktop RTX A4000 has 16 GB and the desktop A5000 has 24 GB, but the RTX A3000 Laptop GPU in the ZBook has 6 GB — a completely different chip with a lower power limit. NVIDIA reuses the “A-series” branding across wildly different memory tiers, so the name alone tells you little. For AI, only the VRAM figure decides what loads: 6 GB runs SDXL and 13B-class LLMs; it does not run FLUX.1 at full precision or fine-tune mid-size models. If you need 16 GB for serious fine-tuning, look at the ThinkPad P15 Gen 2 instead. To understand why this number dominates everything, read what VRAM is and why it matters for AI.
AI Performance in Practice
The RTX A3000 is a capable 6 GB GPU. With 4096 CUDA cores it sits between the consumer RTX 3060 Laptop (6 GB) and RTX 3070 Laptop in raw throughput, with workstation-grade drivers and ECC-capable memory.
For Ollama, GPU offload transforms the experience. Llama 3.1 7B with Q4_K_M loads entirely into 6 GB and delivers an estimated 22–28 tok/s — faster than reading speed. 13B models at Q4 also fit, running around 10–14 tok/s. These are estimates extrapolated from RTX 3060 Laptop benchmarks with comparable core counts. For a full model-by-model breakdown, see our Ollama laptop requirements guide.
Stable Diffusion XL is the real unlock at 6 GB. At 1024×1024 with 20 steps, expect roughly 18–28 seconds per image — slower than a desktop 3060 but perfectly usable for iterating. SD 1.5 is near-instant at 4–6 seconds. ComfyUI runs SDXL pipelines comfortably; FLUX.1 is tight and needs heavy quantisation at this VRAM tier.
Whisper benefits strongly from CUDA: roughly 6× realtime on the base model and ~2× on medium.
Thermal behaviour
This is the ZBook’s signature strength. HP’s vapour-chamber cooling keeps the machine noticeably quieter than a Legion or Precision under identical load. Under combined CPU+GPU work the fans are present but never gaming-laptop loud. The trade-off is a slightly lower sustained GPU power limit (around 80W) than a thick gaming chassis, so peak throughput is a touch lower — but for a laptop you can actually run in a shared office, the acoustics win.
Battery life under AI load
The 83 Wh battery delivers 6–8 hours of light work (browsing, coding, API calls). Under sustained GPU load it drops fast — roughly 70–100 minutes before you reach for the charger. CPU-only inference lands around 2.5–3 hours. Carry the charger for any real AI session.
What to Check Before Buying (Used)
Confirm the GPU is the A3000 6 GB variant ZBook Studio G8 shipped with several GPUs, including the T1200 (4 GB, far weaker) and the A3000 (6 GB). Always verify “RTX A3000” in GPU-Z and in the listing. A T1200 unit is worth meaningfully less for AI.
Thermal paste and fans Run FurMark for 10 minutes and watch GPU temperature in GPU-Z. A healthy unit holds below 87°C. The vapour chamber is reliable, but 3-year-old paste can dry out (~£10 DIY).
RAM configuration The G8 has 2× SO-DIMM slots. Many used units ship with 16 GB — budget £40–60 to reach 32 GB, which you want for 13B CPU fallback and large datasets.
Display variant Configs range from 1080p IPS to a 4K DreamColor panel. The 4K panel is gorgeous but drains battery faster; the 1080p panel is the pragmatic choice for AI work.
Battery health
Run powercfg /batteryreport. On a 2–3 year unit expect 70–85% of the 83 Wh design capacity. Replacements are pricier than mainstream laptops.
Where to Buy in the UK
Back Market UK — The most reliable source for graded ZBook Studio G8 units. Expect £620–£800 for the i7-11800H / RTX A3000 / 32 GB config with a 12-month warranty. Always confirm the GPU is A3000 (not T1200).
Laptops Direct — Occasionally lists ZBook refurbs at £580–£760. Stock is inconsistent; verify the exact GPU and RAM.
eBay UK — Best prices (£560–£740) but more risk. Many are ex-corporate CAD machines. Ask for a GPU-Z screenshot to confirm the A3000 and VRAM before buying.
What to avoid: Any ZBook Studio G8 listing that doesn’t name the GPU, or that shows the T1200 — it has only 4 GB and won’t run SDXL comfortably.
Verdict
AI Score: 69/100 — SD Ready
The HP ZBook Studio G8 is the quiet, premium way into 6 GB GPU AI. That 6 GB of VRAM is the practical floor for Stable Diffusion XL and comfortable Ollama 13B inference, and the ZBook delivers it in a chassis you can actually use in a shared room without sounding like a jet. The vapour-chamber cooling, solid build and HP enterprise support set it apart from gaming laptops at the same price.
The limits are honest: 6 GB won’t run FLUX.1 at full precision or fine-tune mid-size models, and the lower sustained GPU power means a desktop or a thick gaming laptop will out-throughput it. But as a portable CUDA workstation at £580–£800 used, the balance of capability, build and quiet operation is hard to beat.
Buy if: You want SDXL, Ollama 13B and CUDA development in a quiet, premium, genuinely portable workstation.
Don’t buy if: You need 16 GB VRAM for fine-tuning — see the ThinkPad P15 Gen 2 — or you want maximum throughput per pound, where the Lenovo Legion 5 Gen 7 with 8 GB is the stronger value. Comparing options? See our best used laptops for local LLMs roundup.