ThinkPad T14 Gen 3 AMD Review: Best Budget Entry into Local AI
Who is this laptop for?
This is a used/refurbished business laptop that handles lightweight local AI tasks without breaking the bank. No discrete GPU means no Stable Diffusion, but if you need Ollama 7B on CPU, Whisper transcription, or API-based AI tools — this is the cheapest reliable entry point.
Students (Budget: £320–£480)
The T14 Gen 3 is arguably the best value entry into local AI for students. At £320–£380 for a 16 GB config, you get a machine that runs Llama 3.1 7B locally (slowly, but it works). GitHub Copilot and Cursor AI run perfectly since they’re API-based. The build quality survives a backpack — these are ex-corporate machines built to last.
ML Engineers & Data Scientists
This is a travel/secondary machine, not your daily driver. Use it for SSH into your GPU server, running Jupyter notebooks with small datasets, and testing code locally before pushing to a training cluster. CPU-only inference at 4–5 tok/s is too slow for production work but fine for prototyping prompts.
Small Teams & Startups
Solid choice for equipping a team on a budget. Business lease returns mean you can buy 5 of these for the price of one MacBook Pro. The ThinkPad keyboard and trackpoint are genuinely good for productivity. Back Market offers 12-month warranty on refurbished units.
What can it actually run?
| Task | Works? | Notes |
|---|---|---|
| GitHub Copilot / Cursor AI | ✅ Yes | API-based, runs perfectly |
| Whisper transcription (local) | ✅ Yes | ~3× realtime on base model |
| Ollama 7B (Llama 3, Mistral) | ⚠️ Slow | ~4–5 tok/s on CPU with Q4_K_M quantization |
| Ollama 13B | ⚠️ Tight | Needs 32 GB RAM config. ~2 tok/s. Barely usable. |
| Stable Diffusion 1.5 | ❌ No | Integrated GPU, no dedicated VRAM |
| Stable Diffusion XL | ❌ No | Not possible without discrete GPU |
| ComfyUI / FLUX.1 | ❌ No | Requires dedicated GPU with 6+ GB VRAM |
| LoRA fine-tuning | ❌ No | Not practical on integrated GPU |
Key:
- ✅ Yes — works well
- ⚠️ Possible but slow — usable with patience
- ❌ No — hardware limitation prevents this
Full Specifications
| Component | Specification |
|---|---|
| CPU | AMD Ryzen 5 Pro 6650U (6C/12T, Zen 3+) |
| CPU Generation | AMD Zen 3+ (Rembrandt, 6000 series) |
| RAM | 16 GB DDR5-4800 (2× SO-DIMM, upgradeable to 32 GB) |
| Storage | 512 GB NVMe Gen 3 (M.2 2280) |
| GPU | AMD Radeon 660M (integrated, RDNA 2) |
| VRAM | Shared system RAM (no dedicated VRAM) |
| Display | 14” 1920×1080 IPS, 300 nits |
| Battery | 52.5 Wh |
| Weight | 1.41 kg |
| TDP | 28W (sustained) |
| AI Score | 42/100 |
AI Performance in Practice
The Ryzen 5 Pro 6650U is a solid 6-core/12-thread chip running at 28W sustained TDP. For CPU-only inference, it delivers approximately 4–5 tokens per second running Llama 3.1 7B with Q4_K_M quantization through Ollama. That’s slow for interactive chat but workable for testing prompts and short generations.
The integrated Radeon 660M (RDNA 2) technically supports ROCm, but in practice the shared VRAM situation makes GPU offloading unreliable on this chip. Stick to CPU inference — it’s more predictable.
Whisper (base model) runs at approximately 3× realtime — a 10-minute audio file transcribes in about 3 minutes. The medium model drops to roughly 1× realtime. The large-v3 model is impractical at ~0.3× realtime.
Thermal behaviour
ThinkPad T14 Gen 3 has excellent thermal management for its class. Under sustained all-core CPU load (Ollama inference), the fan ramps up noticeably but the chip maintains 25–28W without significant throttling. Surface temperature stays comfortable — you can actually use it on your lap during inference, unlike gaming laptops.
After 30+ minutes of sustained Ollama use, expect ~5% performance degradation as the system settles to thermal equilibrium. This is normal and minimal compared to thinner ultrabooks.
Battery life under AI load
Under normal use (browsing, coding, API calls), expect 6–8 hours. Under sustained CPU inference (Ollama), battery drains fast — roughly 90–120 minutes before needing to plug in. The 52.5 Wh battery is decent but AI inference is power-hungry. Always carry the 65W USB-C charger for AI work.
What to Check Before Buying (Used)
Battery health
T14 Gen 3 batteries hold up reasonably well. On a 2-year-old unit, expect 60–80% of original capacity. Run powercfg /batteryreport in PowerShell and check Full Charge Capacity vs Design Capacity. Below 35 Wh (67% health) means you’ll want a replacement battery (~£40–60).
GPU throttling test Not applicable — integrated GPU only. Focus on CPU thermals instead. Run Cinebench R23 multi-core for 10 minutes. A healthy unit scores 8,500–9,500 sustained. Significantly lower scores suggest thermal paste degradation or blocked vents.
Storage health Check CrystalDiskInfo. Reallocated Sectors Count must be 0. Power On Hours above 15,000 on NVMe drives indicates heavy use — not a deal-breaker but worth negotiating price down. Check remaining SSD life percentage (should be >90% for a 2-year-old unit).
RAM slots T14 Gen 3 AMD has 2× SO-DIMM DDR5 slots. This is a major advantage over the T14s Gen 4 (soldered RAM). Verify both slots work and that RAM runs at DDR5-4800. Open Task Manager → Performance → Memory to check speed. If you buy a 16 GB unit, you can upgrade to 32 GB later (~£50–70 for a 16 GB DDR5 SO-DIMM).
Model-specific issues to watch for
- Some early units had trackpad rattle — press firmly on all corners of the trackpad to check
- BIOS version should be at least 1.30 — older BIOS versions had memory stability issues
- Verify it’s the AMD variant (6650U/6850U), not the Intel version (different performance profile)
Where to Buy in the UK
Back Market UK — The most reliable option. Expect £350–£480 depending on condition grade. “Excellent” grade units look nearly new. All listings include 12-month warranty and 30-day returns. The 16 GB / 512 GB config is the most common.
Laptops Direct — Refurbished stock varies. Slightly cheaper than Back Market (£320–£420) but warranty is typically 6 months. Check the exact spec carefully — some listings are the Intel variant.
eBay UK — Can find deals at £300–£380, but verify seller rating (99%+ feedback, 100+ sales). Insist on photos of the actual unit. Check the serial number against Lenovo’s warranty checker to verify age and original spec.
What to avoid: Don’t confuse with ThinkPad T14 Gen 1 or Gen 2 — older AMD Ryzen 4000/5000 CPUs are significantly slower. Always verify “Gen 3” and “6650U” or “6850U” in the listing. The Gen 3 Intel variant (Core i5-1245U) exists too — it’s fine but has slightly worse battery life.
Verdict
AI Score: 42/100 — LLM Ready
The ThinkPad T14 Gen 3 AMD is the best way to start experimenting with local AI on a budget. At £320–£480, you get a well-built machine that runs Ollama 7B models, handles Whisper transcription, and works perfectly with API-based tools like Copilot and Claude.
The limitation is clear: no discrete GPU means no image generation and no GPU-accelerated inference. You’re stuck with CPU-only speeds of 4–5 tok/s. For many use cases — prompt prototyping, lightweight inference, learning — that’s enough.
The upgradeable RAM (2× SO-DIMM slots) is a genuine advantage. Buy a 16 GB unit now, add another 16 GB stick later when you need 13B models. The build quality is business-grade — expect it to last 4–5 more years of daily use.
Buy if: You need to run Ollama 7B locally, your budget is under £500, and you’re comfortable with CPU-only inference speed. Also ideal as a secondary machine for remote development.
Don’t buy if: You need Stable Diffusion, SDXL, or any image generation — the integrated GPU can’t handle it. Look at the Dell Precision 5560 (4 GB VRAM, ~£480–£680) or Lenovo Legion 5 Gen 6 (6 GB VRAM, ~£550–£750) instead.