ThinkPad P15 Gen 2 Review: 16 GB VRAM for Fine-Tuning on a Used Laptop
Who is this laptop for?
The Lenovo ThinkPad P15 Gen 2 is the used-market flagship for laptop AI — and our headline example of why mobile VRAM naming matters. The config we recommend pairs the Intel Core i7-11800H (or Xeon W-11955M) with the RTX A5000 Laptop GPU and a full 16 GB of GDDR6 VRAM. 16 GB is the line where FLUX.1, SDXL at scale, and real fine-tuning stop being compromises. It’s a heavy, expensive desktop-replacement workstation — and the most capable AI laptop we review.
Students (Budget: £900–£1,300)
For most students this is too much laptop. But for a postgrad or researcher doing genuine model training — not just inference — 16 GB VRAM on a portable machine is transformative, and far cheaper used than any new 16 GB laptop. If you only run inference, save hundreds with the Lenovo Legion 5 Gen 7 (8 GB) instead.
ML Engineers & Data Scientists
This is the one. 16 GB VRAM runs FLUX.1 at full precision, SDXL with every extension, large ComfyUI graphs, and 13B-class LLMs fully on GPU. Crucially it enables real fine-tuning: 7B/13B QLoRA, SD/SDXL LoRA and DreamBooth all fit. With 64 GB RAM (expandable to 128 GB across 4 slots) you can stage huge datasets and run 30B+ models on CPU alongside GPU work. The i7-11800H sustains 45W in a chassis built for continuous load.
Small Teams & Startups
As a shared AI workstation, the P15 Gen 2 replaces a desktop for many teams. ISV-certified, ThinkPad-reliable, with enterprise support — it handles fine-tuning experiments and batch generation that an 8 GB laptop can’t touch. At £900–£1,300 used it’s a fraction of a new mobile workstation. It is genuinely heavy (2.9 kg) — a transportable workstation, not a daily carry.
What can it actually run?
| Task | Works? | Notes |
|---|---|---|
| GitHub Copilot / Cursor AI | ✅ Yes | API-based, runs perfectly |
| Whisper transcription (local) | ✅ Yes | large-v3 at ~1.5× realtime (GPU-accelerated) |
| Ollama 7B / 13B | ✅ Yes | Both fully GPU-resident. 13B Q4 ~22–28 tok/s (estimated) |
| Ollama 34B | ⚠️ Tight | Q4 ~16 GB — fits at low context, or split GPU+CPU. Usable. |
| Stable Diffusion XL | ✅ Yes | ~9–14s per 1024×1024 image, with ControlNet + refiner headroom |
| ComfyUI / FLUX.1 | ✅ Yes | FLUX.1 runs at full precision. Large multi-node graphs are fine. |
| Fine-tuning (QLoRA 7B/13B) | ✅ Yes | 16 GB enables real QLoRA, SDXL LoRA and DreamBooth training. |
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) or Xeon W-11955M |
| CPU Generation | Intel 11th Gen (Tiger Lake H, 2021) |
| RAM | 64 GB DDR4-3200 ECC-capable (4× SO-DIMM, max 128 GB) |
| Storage | 1 TB NVMe Gen 4 (2× M.2 2280 slots) |
| GPU | NVIDIA RTX A5000 Laptop (16 GB GDDR6, 6144 CUDA cores) |
| VRAM | 16 GB GDDR6 (dedicated) |
| Display | 15.6” 1920×1080 or 4K UHD IPS (config-dependent) |
| Battery | 94 Wh |
| Weight | 2.9 kg |
| TDP | 45W CPU + up to 110W GPU |
| AI Score | 88/100 |
Mobile vs desktop: our flagship VRAM example
This laptop is the reason we hammer the VRAM point. The desktop RTX A5000 has 24 GB. The RTX A5000 Laptop GPU in the P15 Gen 2 has 16 GB — same name, two-thirds the memory, a different chip and power envelope. A buyer who reads “A5000” and assumes 24 GB will be disappointed. But 16 GB is still enormous for a laptop: it’s double the 8 GB sweet-spot cards and quadruple a 4 GB entry GPU. It runs models that simply will not load on anything else portable. The lesson holds across the range — always buy on the laptop VRAM figure, never the desktop name. The full explanation is in what VRAM is and why it matters for AI.
AI Performance in Practice
The RTX A5000 Laptop is a professional 16 GB GPU with 6144 CUDA cores and ECC-capable memory. It is the most capable GPU in any laptop we review.
For LLMs, both 7B and 13B run fully GPU-resident with long context: 13B Q4 at an estimated 22–28 tok/s. Even 34B Q4 squeezes in at reduced context, or splits across GPU+CPU using the 64 GB system RAM. See the model-by-model breakdown in our Ollama laptop requirements guide.
For diffusion, 16 GB removes the usual ceilings: SDXL at 9–14 seconds per image with ControlNet and a refiner loaded simultaneously, and FLUX.1 at full precision — the model that needs quantisation on 8 GB and won’t load on 6 GB. Figures estimated from RTX A5000 Laptop benchmarks.
Most importantly, 16 GB makes fine-tuning practical on a laptop: 7B and 13B QLoRA, SDXL LoRA and DreamBooth all fit with room for reasonable batch sizes. This is the capability that separates the P15 from every other machine here.
Thermal behaviour
The P15 is a full workstation with a large dual-fan cooler. It sustains the GPU near its 110W limit for long training runs without throttling. It’s audible under load but lower-pitched and less shrill than a gaming laptop. The 2.9 kg chassis exists precisely to dissipate this heat — it’s the price of sustained 16 GB GPU compute.
Battery life under AI load
The 94 Wh battery is large but the hardware is hungry: light use gives 4–6 hours, while sustained GPU training drains it in under an hour. This is a mains-powered workstation; the 230W charger is heavy. Plan to work plugged in.
What to Check Before Buying (Used)
Confirm the A5000 16 GB — not the A2000/A3000/T1200 The P15 Gen 2 shipped with GPUs from the T1200 (4 GB) up to the A5000 (16 GB). The 16 GB A5000 is the whole reason to buy this machine. Verify “RTX A5000 Laptop” and 16 GB in GPU-Z, and insist on a screenshot. A lower-tier GPU variant is worth far less.
RAM configuration and slots The 4× SO-DIMM layout is rare and valuable. Confirm capacity (32/64 GB common) and that all four slots work — 128 GB is achievable for large-dataset CPU work.
GPU thermals Run FurMark for 15 minutes; a healthy A5000 holds below 87°C near 110W. Workstation GPUs are robust, but verify no throttling on a 3-year-old unit.
Storage Two M.2 Gen 4 slots. Verify SSD health (CrystalDiskInfo) and consider a second drive for datasets and model weights.
Battery and weight expectation
Run powercfg /batteryreport. Don’t over-pay for battery health on a machine you’ll run plugged in. Be realistic about the 2.9 kg weight before buying.
Where to Buy in the UK
Back Market UK — The most reliable graded source, but A5000 16 GB units are less common (limited stock) — expect £950–£1,300 with a 12-month warranty. Confirm the 16 GB A5000 explicitly.
Laptops Direct — Occasionally lists P15 Gen 2 workstations from £900. Verify the GPU tier carefully; lower-VRAM configs are common.
eBay UK — Best prices (£850–£1,200) and where most A5000 units surface, from corporate CAD/engineering fleets. Always demand a GPU-Z screenshot confirming the A5000 and 16 GB.
What to avoid: Any P15 Gen 2 listing that names “Quadro/RTX” without the model and VRAM. The whole value proposition is the 16 GB A5000 — a T1200 or A2000 unit is a different, far cheaper laptop.
Verdict
AI Score: 88/100 — Pro AI
The ThinkPad P15 Gen 2 with the 16 GB RTX A5000 is the most capable AI laptop on the used market and the clearest illustration of our core message: buy on laptop VRAM, not the desktop name. 16 GB runs FLUX.1 at full precision, SDXL with every extension, 13B LLMs on GPU and — uniquely here — real QLoRA fine-tuning. With 64 GB RAM and ISV-grade reliability, it genuinely replaces a desktop workstation.
The price is weight and money: 2.9 kg, a 230W brick, and £900–£1,300. It’s a transportable workstation, not a laptop you carry daily, and it must run plugged in for AI. But if you need 16 GB of mobile VRAM, nothing else used comes close.
Buy if: You fine-tune models, need FLUX.1 at full precision, or run 13B-plus LLMs on GPU — and you accept a heavy, mains-powered workstation.
Don’t buy if: You only run inference (the Lenovo Legion 5 Gen 7 at 8 GB is far cheaper) or you need portability (the HP ZBook Studio G8 is lighter and quieter). Compare the whole dGPU range in our best used laptops for local LLMs roundup.