What can a AMD Mi50 32GB run?

32GB · 1000 GB/s · 240W · $495

Runs great30

Comfortable headroom — daily-driver territory.

Model	Quant	Est. speed	Max context
deepseek moe 16b chat	FP16	Infinity tok/s	—
qwen2 7b instruct	FP16	Infinity tok/s	—
Qwen3-0.6B	FP16	499 tok/s	33k
Qwen3-30B-A3B	Q5_K_M	263 tok/s	33k
Qwen3-Coder-30B-A3B	Q5_K_M	263 tok/s	49k
Gemma-3n-E4B	FP16	221 tok/s	131k
SmolLM2-1.7B	FP16	182 tok/s	8k
GPT-OSS-20B	Q8_0	162 tok/s	50k
Llama-3.2-3B	FP16	96 tok/s	131k
Phi-4-mini	FP16	81 tok/s	131k
Gemma-3-4B	FP16	80 tok/s	131k
Qwen3-4B	FP16	79 tok/s	33k
Gemma-3-12B	Q8_0	45 tok/s	122k
Mistral-7B-Instruct	FP16	43 tok/s	33k
DeepSeek-R1-Distill-Qwen-7B	FP16	41 tok/s	126k
Qwen2.5-Coder-7B	FP16	41 tok/s	126k
Phi-4-14B	Q8_0	40 tok/s	16k
DeepSeek-R1-Distill-Qwen-14B	Q8_0	39 tok/s	104k
Llama-3.1-8B	FP16	39 tok/s	119k
Qwen3-8B	FP16	39 tok/s	33k
Qwen3-14B	Q8_0	35 tok/s	33k
Codestral-22B	Q6_K	34 tok/s	33k
Qwen3-32B	Q4_K_M	31 tok/s	33k
DeepSeek-R1-Distill-Qwen-32B	Q4_K_M	31 tok/s	61k
GLM-4-32B	Q4_K_M	31 tok/s	61k
Mistral-Small-24B	Q6_K	31 tok/s	64k
Qwen3.8	Q6_K	28 tok/s	47k
Qwen3.8 27B	Q6_K	28 tok/s	47k
Gemma-3-27B	Q6_K	28 tok/s	47k
qwen35 27B	Q6_K	28 tok/s	45k

Tight squeeze2

It fits, but close the browser tabs and mind the context.

Model	Quant	Est. speed	Max context
Llama-3.3-70B	IQ2_M	26 tok/s	31k
Qwen2.5-72B	IQ2_M	25 tok/s	27k

Measured on this card2

Everything here was benchmarked on this exact card by someone who owns one — the tables above are our arithmetic; these numbers came off real machines.

Qwen3.8 Flash Next: fastest recorded generation is 31.67 tok/s — IQ1_S (UD) on 2× AMD Mi50 32GB in llama.cpp, measured from an empty context.

Try your own machine →← All hardware