Lenovo Legion 5 Gen 7 Review: 8 GB VRAM for SDXL and FLUX.1
Who is this laptop for?
The Lenovo Legion 5 Gen 7 AMD is a used gaming laptop that doubles as a serious AI machine. The config we recommend pairs the 8-core AMD Ryzen 7 6800H with the RTX 3070 Ti Laptop GPU and 8 GB of GDDR6 VRAM. That extra 2 GB over a 6 GB card is the difference between “SDXL works” and “SDXL, FLUX.1 and light fine-tuning all work” — 8 GB is the current sweet spot for local generative AI on a budget.
Students (Budget: £700–£950)
It’s a real investment, but no other used laptop gives a student this much GPU AI for the money. SDXL is fast, FLUX.1 runs with quantisation, and Ollama 13B flies. The catch is portability and noise: this is a 2.4 kg gaming laptop with loud fans under load. If you mostly do CPU/API work, the ThinkPad T14 Gen 3 is lighter and cheaper.
ML Engineers & Data Scientists
A capable local-AI workstation at a fraction of new-GPU prices. 8 GB VRAM handles SDXL and FLUX.1 (quantised), ComfyUI pipelines, and 7B QLoRA fine-tuning — the entry point to actually training, not just running, models. The Ryzen 7 6800H (Zen 3+) sustains 45W and the MUX switch lets the dGPU drive the display directly for maximum throughput. 32 GB RAM covers large datasets and CPU fallback for bigger models.
Small Teams & Startups
For prototyping generative-AI features, the Legion 5 Gen 7 is the cheapest reliable 8 GB CUDA box. At £700–£950 used it undercuts any new RTX 4070 laptop while matching its VRAM for inference. It’s a desk machine — loud and heavy for client meetings — but for a shared “AI workstation” in the office it’s outstanding value.
What can it actually run?
| Task | Works? | Notes |
|---|---|---|
| GitHub Copilot / Cursor AI | ✅ Yes | API-based, runs perfectly |
| Whisper transcription (local) | ✅ Yes | ~7× realtime on base model (GPU-accelerated) |
| Ollama 7B (Llama 3, Mistral) | ✅ Yes | ~30–38 tok/s with full GPU offload (estimated) |
| Ollama 13B | ✅ Yes | Q4 fits comfortably in 8 GB. ~16–20 tok/s (estimated) |
| Stable Diffusion 1.5 | ✅ Yes | ~3–5s per 512×512 image at 20 steps |
| Stable Diffusion XL | ✅ Yes | ~12–18s per 1024×1024 image. Comfortable on 8 GB. |
| ComfyUI / FLUX.1 | ✅ Yes | FLUX.1 runs with Q8/NF4 quantisation. SDXL workflows are smooth. |
| LoRA fine-tuning (7B QLoRA) | ✅ Yes | 7B QLoRA and SD LoRA training feasible. 8 GB is the entry point. |
Key:
- ✅ Yes — works well
- ⚠️ Possible but slow — usable with patience
- ❌ No — hardware limitation prevents this
Full Specifications
| Component | Specification |
|---|---|
| CPU | AMD Ryzen 7 6800H (8C/16T, Zen 3+) |
| CPU Generation | AMD Ryzen 6000H (Zen 3+, 2022) |
| RAM | 32 GB DDR5-4800 (2× SO-DIMM, upgradeable to 32 GB) |
| Storage | 512 GB NVMe Gen 4 (M.2 2280, 2 slots) |
| GPU | NVIDIA RTX 3070 Ti Laptop (8 GB GDDR6, 5888 CUDA cores) |
| VRAM | 8 GB GDDR6 (dedicated) |
| Display | 15.6” 1920×1080 IPS, 165 Hz (common config) |
| Battery | 80 Wh |
| Weight | 2.4 kg |
| TDP | 45W CPU + up to 140W GPU (MUX switch) |
| AI Score | 79/100 |
Mobile vs desktop: the 3070 Ti naming trap
The desktop RTX 3070 Ti has 8 GB, and so does the Laptop RTX 3070 Ti — but the laptop chip runs at a far lower power limit (up to ~140W vs 290W) and lower clocks. The VRAM matches, so what fits is the same; what differs is speed. For AI that’s the right trade: VRAM decides whether a model loads at all, and 8 GB loads SDXL, FLUX.1 (quantised) and 13B LLMs. You simply wait a little longer per image or per token than on a desktop. If you want to understand why VRAM, not raw TFLOPS, is the gatekeeper, read what VRAM is and why it matters for AI.
AI Performance in Practice
The RTX 3070 Ti Laptop is a strong 8 GB GPU with 5888 CUDA cores. With the MUX switch set to discrete mode it drives workloads near its full ~140W envelope.
Ollama is effortless: Llama 3.1 7B Q4 runs at an estimated 30–38 tok/s, and 13B Q4 at 16–20 tok/s, both fully GPU-resident. For the complete VRAM/RAM requirements table across model sizes, see our Ollama laptop requirements guide.
Stable Diffusion XL is where 8 GB shines: roughly 12–18 seconds per 1024×1024 image at 20 steps, with room for ControlNet and refiners. FLUX.1 — the model that defeats 6 GB cards — runs here with Q8 or NF4 quantisation. SD 1.5 is essentially instant. These figures are estimated from RTX 3070 Ti Laptop benchmarks.
The headline capability is fine-tuning: 8 GB is enough for 7B QLoRA and Stable Diffusion LoRA training with gradient checkpointing. That moves the Legion from “runs models” to “trains models”, which no 6 GB laptop does comfortably.
Thermal behaviour
This is a gaming laptop and it cools like one — effectively, but loudly. Under combined CPU+GPU load the fans are clearly audible and not office-friendly. The upside is sustained performance: the chassis holds the GPU near its power limit for long generation runs without throttling. Use a cooling pad for marathon training sessions.
Battery life under AI load
The 80 Wh battery manages 4–6 hours of light use, but GPU AI drains it in 45–70 minutes, and the dGPU barely runs at full power on battery anyway. This is a plugged-in machine for AI. The 230W charger is large and heavy.
What to Check Before Buying (Used)
Confirm the 8 GB 3070 Ti, not a 6 GB 3060 Legion 5 Gen 7 shipped with several GPUs. Verify “RTX 3070 Ti Laptop” and 8 GB in GPU-Z — a 3060 (6 GB) variant is cheaper and weaker. The VRAM is the whole point here.
MUX switch and BIOS mode Confirm the MUX switch works (Lenovo Vantage → Hybrid/Discrete). Discrete mode gives maximum AI throughput. Ex-gaming units occasionally have BIOS quirks — check it boots cleanly.
Fan wear and dust Gaming laptops live hot. Run a 15-minute FurMark test; healthy GPU temps stay below 87°C. Listen for fan bearing rattle. A blow-out and repaste (~£15 DIY) is worth doing on any used gaming laptop.
RAM and SSD Two SO-DIMM slots and often a spare M.2 slot. Verify 32 GB DDR5 and check SSD health with CrystalDiskInfo — gaming units sometimes have heavily-written drives.
Battery and cosmetic wear
Run powercfg /batteryreport. Gaming laptops are often used plugged in, so batteries can be healthy — but check for keyboard shine and hinge play.
Where to Buy in the UK
Back Market UK — Graded Legion 5 Gen 7 units appear regularly. Expect £740–£950 for the Ryzen 7 6800H / RTX 3070 Ti / 32 GB config with a 12-month warranty. Confirm the 8 GB GPU.
Laptops Direct — Lists Legion refurbs at £700–£900. Verify the exact GPU — 3060 and 3070 Ti configs look similar in listings.
eBay UK — Best prices (£680–£880) and the widest stock, since gamers upgrade often. Ask for a GPU-Z screenshot confirming the 3070 Ti and 8 GB before buying.
What to avoid: Listings that say “Legion 5 Gen 7” without naming the GPU. The 6 GB RTX 3060 variant won’t run FLUX.1 or 7B fine-tuning — pay only the 8 GB premium for an 8 GB card.
Verdict
AI Score: 79/100 — Pro AI
The Lenovo Legion 5 Gen 7 with the 8 GB RTX 3070 Ti is the best value path to “real” generative AI on the used market. 8 GB VRAM is the threshold where SDXL is comfortable, FLUX.1 becomes possible, and 7B QLoRA fine-tuning is on the table — a meaningful step beyond every 6 GB laptop. For under £1,000 used, nothing else offers this much CUDA capability.
The compromises are pure gaming-laptop: 2.4 kg, loud fans, poor battery under load, and a heavy charger. This is a plugged-in desk workstation, not a café machine. If you can live at a desk, it’s the strongest AI-per-pound laptop we review.
Buy if: You want comfortable SDXL, FLUX.1 and entry-level fine-tuning, and you’ll run it plugged in at a desk.
Don’t buy if: You need portability or quiet — the HP ZBook Studio G8 (6 GB, quiet, premium) is the civilised alternative — or you need 16 GB VRAM for bigger fine-tunes, where the ThinkPad P15 Gen 2 is the answer. See all options in our best used laptops for local LLMs roundup.