Best used graphics cards for local AI

Ranked by what each one runs fully in memory at 32k context, with prices in the UK today.

In short

The top pick in the UK right now is the NVIDIA RTX PRO 6000 Blackwell, £14,499 at eBay (used, asking price): it runs DeepSeek V4 Flash fully in memory at 32k context at about 69 tok/s (estimated). Ranked by the most capable open model each one runs fully in memory at 32k context and a usable speed, then by how many models it runs, then by speed on Qwen 3.8 27B, then by price.

Updated . Speeds are estimates; how they are worked out.

The list

Best first. Used prices are eBay asking prices. Speeds are estimates est.

#CardMemoryPriceBest model it runs at 32kModels in memoryQwen 3.8 27Btok/s per £1,000
1NVIDIA RTX PRO 6000 Blackwell96 GB£14,499eBay, askingDeepSeek V4 FlashUD-Q2_K_XL69 tok/s4035 tok/s2.4
2NVIDIA GeForce RTX 509032 GB£5,668eBay, askingQwen 3.8 27BQ8_035 tok/s2935 tok/s6.2
3NVIDIA GeForce RTX 3090 Ti24 GB£1,250eBay, askingQwen 3.8 27BUD-Q5_K_M28 tok/s2628 tok/s22.6
4NVIDIA GeForce RTX 409024 GB£2,812eBay, askingQwen 3.8 27BUD-Q5_K_M28 tok/s2628 tok/s10.1
5NVIDIA GeForce RTX 309024 GB£991eBay, askingQwen 3.8 27BUD-Q5_K_M26 tok/s2626 tok/s26.5
6AMD Radeon RX 7900 XTX24 GB£780eBay, askingQwen 3.8 27BUD-Q5_K_M24 tok/s2624 tok/s30.6
7NVIDIA Tesla P4024 GB£261eBay, askingQwen 3.6 35B-A3BUD-Q4_K_M50 tok/s1510 tok/s37.2
8Intel Arc A770 16GB16 GB£400eBay, askinggpt-oss 20BMXFP448 tok/s13offload–
9AMD Radeon RX 7800 XT16 GB£460eBay, askinggpt-oss 20BMXFP464 tok/s13offload–
10NVIDIA GeForce RTX 5060 Ti 16GB16 GB£650eBay, askinggpt-oss 20BMXFP452 tok/s13offload–
11AMD Radeon RX 9070 XT16 GB£680eBay, askinggpt-oss 20BMXFP465 tok/s13offload–
12NVIDIA GeForce RTX 4070 Ti Super16 GB£802eBay, askinggpt-oss 20BMXFP477 tok/s13offload–

Questions

What is the best used graphics card for local AI in the UK?
The top pick in the UK right now is the NVIDIA RTX PRO 6000 Blackwell, £14,499 at eBay (used, asking price): it runs DeepSeek V4 Flash fully in memory at 32k context at about 69 tok/s (estimated).
Which gives the most speed for the money?
The NVIDIA Tesla P40: about 10 tok/s on Qwen 3.8 27B for £261 (used, asking price), the most tokens per second per unit of price on this list (estimated).
What is the cheapest option here that runs Qwen 3.8 27B fully in memory?
The NVIDIA Tesla P40, £261 at eBay (used, asking price), at about 10 tok/s at 32k context (estimated).

Other lists