GPU Picker

Which GPU is right for me?

Five questions, one honest answer. No affiliate links, no invented benchmarks — just the hardware class that fits your workload, with real examples.

Q1 / 5

What will this GPU mostly do?

Local LLM inference Gaming ML training / fine-tuning Creative (3D, video, rendering)

Q2 / 5

What are you putting it in?

Desktop tower Laptop Server / homelab rack

Q3 / 5

Budget band (GPU only)?

Prices move constantly — treat these as bands, and check current listings.

Under ~$400 ~$400–900 ~$900–2,000 $2,000+

Q4 / 5

How big are the models (or scenes) you want to run?

For LLMs: a model needs roughly 1GB of VRAM per billion parameters at 8-bit, about half that at 4-bit.

Small — 7–8B models / 1080p Medium — 13–14B / 1440p Large — 30–70B / 4K & heavy scenes Huge — 70B+ / multiple models at once

Q5 / 5

Power and noise?

Quiet & efficient — it lives near me Balanced Don't care — max performance

Answer all five questions first — every one changes the recommendation.

rule of thumb: ~1GB VRAM per 1B params @ 8-bit · ~0.5GB per 1B @ 4-bit · leave headroom for context

Prices move — check current listings before buying. Recommendations are hardware classes, not endorsements; no affiliate links on this page.

The long-form version — per-model VRAM tables, used-market buying checklist, laptop vs. desktop math — is in the works.

That email didn't look valid — mind checking it?

gpu4.me — part of the gpu4.* family