RTX 5090 32GB for local AI in the UK

🧾 Key facts

Cheapest UK price

£4,240 used

Memory

32 GB

Memory bandwidth

1792 GB/s

Largest model it can run

~50B parameters at 4-bit (estimate: ~0.6 GB per billion parameters + ~2 GB overhead)

Shop

eBay UK (used)

• £4,240 used: median of 131 UK eBay sold listings, 90 days, checked date unknown (source)
• £4,400 new: 'New from' column, checked date unknown (source)

Disclosure: eBay "Shop" links are affiliate links (marked rel="sponsored"): if you buy, we may earn a small commission at no extra cost to you. Rankings are never affected by commission.

📏 Speed and value by model size

Speed is the median single-user decode speed running the like-for-like reference model for that class, in tokens per second (*a token is roughly ¾ of a word, so words/sec is a rough guide). Ranks use the cheapest UK price and need at least 3 test runs; ⚠ = low sample, shown but not ranked. "Fits" means the model fits fully in memory at ~4-bit. How we measure.

🧪 Other models measured

• deepseek-coder-v2-lite (size not verified): 301.3 tok/s only 2 tests ⚠ · run
• phi-4 (size not verified): 140.0 tok/s only 2 tests ⚠ · run
• stable-code-instruct-3b (3.0B): 336.5 tok/s only 2 tests ⚠ · run
• Qwen 2.5 7B (7.0B, 7-9B class): 239.7 tok/s only 2 tests ⚠ · run
• qwen2.5-coder-7b (7.0B, 7-9B class): 250.1 tok/s only 2 tests ⚠ · run
• llama-3.1-8b (8.0B, 7-9B class): 230.8 tok/s only 2 tests ⚠ · run
• Gemma 2 9B (9.0B, 7-9B class): 111.1 tok/s only 2 tests ⚠ · run
• yi-coder-9b (9.0B, 7-9B class): 133.1 tok/s only 2 tests ⚠ · run
• Qwen 2.5 14B (14.0B, 12-14B class): 131.8 tok/s only 2 tests ⚠ · run
• qwen2.5-coder-14b (14.0B, 12-14B class): 138.0 tok/s only 2 tests ⚠ · run
• qwen2.5-coder-14b Q4_K_M (14.0B, 12-14B class): 137.2 tok/s only 1 test ⚠ · run
• gpt-oss-20b (20.0B): 193.9 tok/s only 2 tests ⚠ · run
• gpt-oss-20b Q4_K_M (20.0B): 324.2 tok/s only 1 test ⚠ · run
• codestral-22b-v0.1 (22.0B): 81.8 tok/s only 2 tests ⚠ · run
• lfm2.5-24b-a2b Q4_K_M (24.0B, 27-35B class): 308.5 tok/s only 1 test ⚠ · run
• Qwen 3.8 27B Q4_K_M (27.0B, 27-35B class): 65.7 tok/s from 4 tests · run
• qwen3-coder-30b-a3b (30.0B, 27-35B class): 164.2 tok/s only 2 tests ⚠ · run
• qwen3-coder-30b-a3b Q4_K_M (30.0B, 27-35B class): 304.9 tok/s only 1 test ⚠ · run
• Gemma 4 31B Q4_K_M (31.0B, 27-35B class): 67.3 tok/s from 33 tests · run
• Qwen 2.5 32B (32.0B, 27-35B class): 70.3 tok/s only 2 tests ⚠ · run
• qwen2.5-coder-32b (32.0B, 27-35B class): 69.7 tok/s only 2 tests ⚠ · run
• Qwen 3 32B (32.0B, 27-35B class): 68.0 tok/s only 2 tests ⚠ · run
• Qwen 3 32B Q4_K_M (32.0B, 27-35B class): 69.8 tok/s only 1 test ⚠ · run
• qwen3.6-35b-a3b Q4_K_M (35.0B, 27-35B class): 220.4 tok/s from 26 tests · run
• nvidia-nemotron-3.5-bolt-74b-a9b-24gb (74.0B, 70B+ class): 154.7 tok/s only 1 test ⚠ · run
• nvidia-nemotron-3.5-bolt-74b-a9b-24gb-ctx32k (74.0B, 70B+ class): 157.3 tok/s only 1 test ⚠ · run

Shown for information only: these don't affect rank. Mixture-of-experts models (names with "a3b", "moe" and similar) run much faster than their size suggests.

⚖️ Similar price alternatives

• RTX 4090 24GB £2,200 used · 24 GB · 167.7 tok/s on 7-9B · #19 of 23 for 7-9B
• AMD Strix Halo (Ryzen AI Max+ 395) 128GB £2,830 new · 128 GB · 50.6 tok/s on 7-9B · #23 of 23 for 7-9B
• Mac Studio M5 Max 40-core GPU (base config) £3,099 new · 48 GB · 111.6 tok/s on 7-9B · #22 of 23 for 7-9B
• Mac Studio M5 Ultra (base config) £5,499 new · 96 GB · 179.1 tok/s on 7-9B