Table of Contents
The "cheapest GPU" question depends on what you actually need to run. This page lists five price tiers and what each unlocks.
TL;DR
For Whisper / embeddings only: RTX 3050 6 GB at £79/mo. For 7B chatbots: RTX 3090 24 GB at £159/mo (FP16) or RTX 5060 Ti at £119/mo (FP8). For 13B+ FP16: RTX 5090 at £399/mo.
Five price tiers
| Tier | GPU | Monthly | What it runs |
|---|---|---|---|
| Entry | RTX 3050 6 GB | £79 | Whisper, embeddings, INT4 7B |
| Cheap | RTX 3060 12 GB | £99 | INT4 7B, Phi-3 Mini FP16, embeddings + reranker |
| Sweet spot (cheap) | RTX 3090 24 GB | £159 | FP16 7B/8B, INT4 13B, FLUX FP16 |
| Sweet spot (modern) | RTX 5060 Ti 16 GB | £119 | FP8 7B/8B, FLUX FP8, smaller voice agent |
| Production | RTX 5090 32 GB | £399 | FP8 14B, INT4 70B, voice agent, multi-model |
Cheapest by workload
| Workload | Cheapest credible GPU | Monthly |
|---|---|---|
| Whisper-only transcription | RTX 3050 6 GB | £79 |
| Embeddings only | RTX 3060 12 GB | £159 |
| Mistral 7B FP16 chatbot | RTX 3090 24 GB | £159 |
| Mistral 7B FP8 chatbot | RTX 5060 Ti 16 GB | £119 |
| Llama 3.1 8B FP16 | RTX 3090 24 GB | £159 |
| Qwen 2.5 14B FP16 | RTX 5090 32 GB | £399 |
| SDXL image gen | RTX 3060 12 GB | £159 |
| FLUX.1 dev FP16 | RTX 3090 24 GB | £159 |
| Voice agent stack | RTX 5090 32 GB | £399 |
Verdict
The cheapest GPU isn't always the best value. The RTX 5060 Ti at £119/mo (FP8 modern) and the RTX 3090 at £159/mo (FP16 mature) are essentially tied for "cheapest credible production card".
Bottom line
Don't over-buy. Match the GPU to the workload. See full GPU catalogue.
