ThinkPad X1 Extreme Gen 4 Review: Premium 4 GB VRAM for Portable AI
Who is this laptop for?
The Lenovo ThinkPad X1 Extreme Gen 4 is the premium thin entry into GPU AI. It puts an RTX 3050 Ti Laptop GPU with 4 GB of GDDR6 and a Core i7-11800H into a 16-inch, ~1.8 kg ThinkPad with a first-class keyboard and a stunning display. It is not the fastest AI machine here — 4 GB VRAM is the entry tier — but it is by far the nicest to live with day to day, and it crosses the line that integrated-graphics laptops cannot: it has CUDA.
Students (Budget: £620–£880)
A superb all-round student laptop that also does real GPU AI. The 4 GB RTX 3050 Ti runs Stable Diffusion 1.5 and accelerates Whisper and 7B LLMs — enough to learn the full CUDA workflow without a desk-bound machine. You pay a premium for the X1’s build and screen versus a cheaper Dell Precision; if image-generation capability matters more than polish, weigh the Dell Precision 5560 (also 4 GB) against it.
ML Engineers & Data Scientists
Think of this as a beautiful daily-driver with a CUDA safety net, not a primary training box. 4 GB VRAM handles SD 1.5, 7B LLMs partly on GPU, and CUDA development and debugging, but it cannot load SDXL comfortably or 13B on GPU. If you mostly write code, call APIs and occasionally test a small model locally — and you want the best ThinkPad keyboard and a 16” display for it — the X1 Extreme is a joy. For sustained generative work, an 8 GB machine is the real tool.
Small Teams & Startups
As a client-facing, premium laptop that still runs local AI demos, the X1 Extreme looks and feels the part. It is portable enough to carry daily and capable enough to show a Stable Diffusion 1.5 or 7B LLM demo on-device. For a shared generation or fine-tuning workstation, though, point the budget at 8 GB or 16 GB instead.
What can it actually run?
| Task | Works? | Notes |
|---|---|---|
| GitHub Copilot / Cursor AI | ✅ Yes | API-based, runs perfectly |
| Whisper transcription (local) | ✅ Yes | medium model GPU-accelerated; large-v3 better on CPU+RAM |
| Ollama 7B | ✅ Yes | Q4 splits GPU+CPU (4 GB won’t hold it all). ~12–18 tok/s |
| Ollama 13B | ⚠️ Tight | CPU + 16 GB RAM only; slow. 32 GB strongly recommended |
| Stable Diffusion 1.5 | ✅ Yes | ~12–20s per 512×512 image at 20 steps |
| Stable Diffusion XL | ⚠️ Tight | Loads with --medvram/tiling but slow; 4 GB is below comfort |
| ComfyUI / FLUX.1 | ❌ No | FLUX needs far more than 4 GB even quantised |
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) |
| CPU Generation | Intel 11th Gen (Tiger Lake H, 2021) |
| RAM | 16 GB DDR4-3200 (2× SO-DIMM, max 64 GB) |
| Storage | 512 GB NVMe Gen 4 (2× M.2 2280 slots) |
| GPU | NVIDIA GeForce RTX 3050 Ti Laptop (4 GB GDDR6, ~60W) |
| VRAM | 4 GB GDDR6 (dedicated) |
| Display | 16” 2560×1600 IPS or 4K UHD+ (config-dependent) |
| Battery | 90 Wh |
| Weight | 1.81 kg |
| TDP | 45W CPU + up to 60W GPU |
| AI Score | 60/100 |
Mobile vs desktop: why 4 GB is the floor
There is no desktop “RTX 3050 Ti” — this is a laptop-only GPU — but the VRAM lesson still holds, because 4 GB is the floor for serious AI. Stable Diffusion 1.5 fits and runs well; SDXL needs 6–8 GB to be comfortable and FLUX needs far more. VRAM is the hard wall: a model either fits in 4 GB or it does not, and no amount of CPU or RAM changes that for GPU work. The X1 Extreme is the most pleasant way to own that 4 GB floor — but if your work is image generation, budget for 6–8 GB instead. Our what VRAM is and why it matters for AI guide explains the tiers, and GGUF quantization explained shows how quantisation stretches small VRAM for LLMs.
AI Performance in Practice
The RTX 3050 Ti Laptop is an entry Ampere GPU with 2560 CUDA cores and a modest ~60W power limit. For Stable Diffusion 1.5 it is genuinely usable — roughly 12–20 seconds per 512×512 image at 20 steps, with ControlNet possible at low resolution. SDXL technically loads with memory-saving flags (--medvram, tiling) but the 4 GB ceiling makes it slow and fiddly; treat SDXL as “possible, not pleasant” here.
For LLMs, 4 GB cannot hold a 7B model alone, so Ollama splits layers across GPU and CPU — expect an estimated 12–18 tok/s on 7B Q4, faster than a pure-CPU laptop but well behind an 8 GB card. 13B is a CPU-and-RAM job; fit at least 32 GB if you plan to run it. See our Ollama laptop requirements guide for the model-by-model picture.
Thermal behaviour
The X1 Extreme is a thin-and-light, and its cooling is tuned for quiet daily use rather than sustained GPU load. The 3050 Ti’s modest ~60W ceiling helps — there is less heat to shed than an RTX 3080 — but under combined CPU+GPU load the chassis warms and fans spin up audibly. For the bursty SD 1.5 and small-LLM workloads this machine is suited to, thermals are fine; it is not built for hour-long GPU training.
Battery life under AI load
The 90 Wh battery is large and gives genuinely good general-use longevity (6–9 hours of coding and browsing) — one of the X1 Extreme’s real strengths. As with every dGPU laptop, sustained GPU inference cuts that to around 90 minutes and the GPU throttles on battery. For SD 1.5 and GPU work, stay on the 135–170W charger.
What to Check Before Buying (Used)
Confirm it has the RTX 3050 Ti — not the integrated-only config Some X1 Extreme Gen 4 SKUs shipped with weaker GPU options or were ordered GPU-light. Verify “RTX 3050 Ti Laptop” and 4 GB in GPU-Z and ask for a screenshot. Without the dGPU this is just an expensive integrated laptop.
RAM and the path to 32 GB Base units often ship 16 GB. Two SO-DIMM slots allow up to 64 GB — budget an upgrade to 32 GB if you intend to run 13B LLMs on CPU. Confirm both slots are accessible.
Display variant and burn-in The 4K UHD+ and OLED-adjacent panels are gorgeous but hammer battery; the 2560×1600 IPS is the pragmatic AI-dev choice. Inspect any high-res panel for uniformity and wear.
Thermals and fan health
Run FurMark for 10–15 minutes and watch for throttling; thin chassis age less gracefully than workstations. Check powercfg /batteryreport for wear on the 90 Wh cell.
Where to Buy in the UK
Back Market UK — Reliable graded X1 Extreme Gen 4 units appear regularly, typically £680–£880 with a 12-month warranty. Confirm the RTX 3050 Ti / 4 GB config.
Laptops Direct — Lists ex-corporate and open-box X1 Extreme units from around £620. Verify the GPU and RAM, as integrated-leaning configs surface here.
eBay UK — Best prices (£620–£820) and the widest choice, often from premium corporate fleets. Demand a GPU-Z screenshot confirming the 3050 Ti and 4 GB, and check the display variant.
What to avoid: Listings that omit the GPU, or that pair a high-res 4K panel with poor battery-health photos — you’ll pay a premium for the screen and inherit a worn battery.
Verdict
AI Score: 60/100 — SD Ready
The ThinkPad X1 Extreme Gen 4 is the premium thin gateway to GPU AI. 4 GB of CUDA VRAM is the entry tier — enough for Stable Diffusion 1.5, accelerated Whisper and partial-GPU 7B LLMs — wrapped in the best keyboard, a beautiful 16” display and a ~1.8 kg chassis you’ll happily carry daily. As an all-round laptop that also does real GPU AI, nothing here is nicer to use.
The honest limit is the 4 GB ceiling: no comfortable SDXL, no FLUX, no 13B on GPU. If image generation is your main job, the money is better spent on 6–8 GB. But if you want a superb daily ThinkPad with a genuine CUDA safety net, the X1 Extreme earns its premium.
Buy if: You want a top-tier portable ThinkPad for coding and daily work, with enough GPU for SD 1.5 and small-model experiments.
Don’t buy if: Your focus is SDXL/FLUX or 13B LLMs (get an 8 GB machine like the ASUS ROG Zephyrus G14) or you want maximum GPU value over polish (the Dell Precision 5560 gives similar 4 GB capability for less). Compare the full range in our best used laptops for local LLMs roundup.