Best GPU for 7B to 9B AI models
The "B" in a model's name is how many billions of settings it has. Bigger models give better answers but need more memory: to run 7B to 9B models fully on the card (or a Mac's shared memory) you need at least 6 GB. Below is every option we track with that much memory and a verified UK price, ranked by value where there are enough real speed tests. Prices refresh every morning.
Best value for 7B to 9B models
RTX 3060 Ti 8GB: £215 used, 92.2 tokens per second on 7B to 9B models (from 8 tests), 8 GB of memory.
A token is roughly ¾ of a word, so that is about 69 words per second.
Cheapest way in
The cheapest option we track with enough memory for 7B to 9B models is the RTX 3060 Ti 8GB at £215 used (8 GB).
Ranked by value
Price divided by speed on 7B to 9B models. Lower £ per speed is better.
| Rank | Option | UK price | Speed | £ per speed | Memory |
|---|---|---|---|---|---|
| #1 | RTX 3060 Ti 8GB | £215 used | 92.2 tok/s from 8 tests | £2.34 | 8 GB |
| #2 | RTX 3080 10GB | £341 used | 131.0 tok/s from 5 tests | £2.60 | 10 GB |
| #3 | RTX 3060 12GB | £238 used | 71.1 tok/s from 19 tests | £3.35 | 12 GB |
| #4 | RX 6800 XT 16GB | £357 used | 95.6 tok/s from 5 tests | £3.73 | 16 GB |
| #5 | RX 6800 16GB | £316 used | 83.9 tok/s from 19 tests | £3.76 | 16 GB |
| #6 | RX 6900 XT 16GB | £380 used | 98.2 tok/s from 14 tests | £3.87 | 16 GB |
| #7 | Intel Arc B580 12GB | £287 used | 70.1 tok/s from 3 tests | £4.09 | 12 GB |
| #8 | RX 7900 XT 20GB | £565 used | 117.5 tok/s from 22 tests | £4.81 | 20 GB |
| #9 | RTX 4070 12GB | £468 used | 92.3 tok/s from 3 tests | £5.08 | 12 GB |
| #10 | RX 7900 XTX 24GB | £772 used | 141.2 tok/s from 176 tests | £5.47 | 24 GB |
| #11 | RTX 5070 Ti 16GB | £921 used | 168.2 tok/s from 13 tests | £5.47 | 16 GB |
| #12 | RX 9070 XT 16GB | £656 used | 116.5 tok/s from 79 tests | £5.63 | 16 GB |
| #13 | RTX 4080 Super 16GB | £850 used | 147.9 tok/s from 4 tests | £5.75 | 16 GB |
| #14 | RTX 4070 Ti Super 16GB | £788 used | 134.3 tok/s from 6 tests | £5.87 | 16 GB |
| #15 | RTX 3090 Ti 24GB | £1,038 used | 155.6 tok/s from 4 tests | £6.67 | 24 GB |
| #16 | RTX 3090 24GB | £986 used | 143.5 tok/s from 35 tests | £6.87 | 24 GB |
| #17 | RTX 5080 16GB | £1,274 used | 183.8 tok/s from 18 tests | £6.93 | 16 GB |
| #18 | RTX 4090 24GB | £2,201 used | 167.7 tok/s from 22 tests | £13.12 | 24 GB |
| #19 | RTX 5090 32GB | £4,301 used | 292.5 tok/s from 26 tests | £14.70 | 32 GB |
| #20 | Mac mini M5 Pro (base config) | £1,699 new | 63.9 tok/s from 5 tests | £26.59 | 24 GB |
| #21 | Mac Studio M5 Max 40-core GPU (base config) | £3,099 new | 111.6 tok/s from 4 tests | £27.77 | 48 GB |
| #22 | AMD Strix Halo (Ryzen AI Max+ 395) 128GB | £2,830 new | 50.6 tok/s from 48 tests | £55.93 | 128 GB |
Also fits, not ranked yet
- RTX 4060 Ti 16GB £422 used, 16 GB, no speed tests at this size yet
- RTX 5060 Ti 16GB £526 used, 16 GB, 82.4 tok/s from only 2 tests ⚠
- Mac Studio M5 Ultra (base config) £5,499 new, 96 GB, 179.1 tok/s from only 1 test ⚠
Other model sizes: 12B to 14B models · 27B to 35B models · 70B and bigger models · By budget: under £300 · under £500 · under £1,000 · under £2,000 · All cards · Beginner's guide · How we rank
Weekly UK value moves, by email
Price drops, new benchmark results, best buys under £500 / £1,000 / £2,000.
Check your inbox to confirm: we use double opt-in, so you're only added after clicking the link in the confirmation email. Free; unsubscribe any time. Sent via Mailchimp (see their privacy notice).