RX 9070 XT 16GB — local LLM fit & speed

AMD · 16 GB GDDR6 · 640 GB/s · 304 W · MSRP $599 · Search Amazon (affiliate)

The RX 9070 XT 16GB is a AMD accelerator with 16 GB GDDR6 and about 640 GB/s peak memory bandwidth. AI Hashrate estimates local LLM decode (batch 1) at 8K context: fit counts weights + KV cache + 1 GB runtime overhead — not weights alone. At Q4, about 25 curated models fit fully on this card; at FP16, about 10 fit. Top Q4 speeds: Qwen3-0.6B ≈ 509.1 tok/s (estimated, fits); gemma-3-1b-it ≈ 407.3 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 339.4 tok/s (estimated, fits); Qwen3-1.7B ≈ 203.6 tok/s (estimated, fits); gpt-oss-20b ≈ 113.1 tok/s (estimated, fits). Relative ranking is more reliable than absolute tok/s. See methodology for the bandwidth formula and measured-anchor policy. Methodology.

Table: decode speed estimates (Q4 / FP16). Measured rows override estimates. MSRP is list price, not live retail. Context default 8K.

Model	Params (B)	Quant	tok/s	Fits?
Qwen3-0.6B	0.8	Q4	509.1 est.	Yes
gemma-3-1b-it	1.0	Q4	407.3 est.	Yes
Llama-3.2-1B-Instruct	1.2	Q4	339.4 est.	Yes
Qwen3-30B-A3B	30.5	Q4	293.9 est.	No
Qwen3-Coder-30B-A3B-Instruct	30.5	Q4	293.9 est.	No
Qwen3-30B-A3B-Instruct-2507	30.5	Q4	293.9 est.	No
GLM-4.7-Flash	31.2	Q4	282.1 est.	No
Qwen3-1.7B	2.0	Q4	203.6 est.	Yes
Qwen3-0.6B	0.8	FP16	140.0 est.	Yes
gpt-oss-20b	21.5	Q4	113.1 est.	Yes
gemma-3-1b-it	1.0	FP16	112.0 est.	Yes
Phi-4-mini-instruct	3.8	Q4	107.2 est.	Yes
Phi-4-mini-reasoning	3.8	Q4	107.2 est.	Yes
gemma-3-4b-it	4.3	Q4	104.4 est.	Yes
Mistral-7B-Instruct-v0.3	7.2	Q4	103.3 measured (storagereview-rx9070xt-review)	Yes
Qwen3-4B-Instruct-2507	4.0	Q4	101.8 est.	Yes
Qwen3-4B-Thinking-2507	4.0	Q4	101.8 est.	Yes
Llama-3.2-1B-Instruct	1.2	FP16	93.3 est.	Yes
Llama-3.2-3B-Instruct	3.2	Q4	89.6 measured (geerlingguy-ai-benchmarks)	Yes
Llama-3.1-8B-Instruct	8.0	Q4	89.1 measured (storagereview-rx9070xt-review)	Yes
Qwen3-1.7B	2.0	FP16	56.0 est.	Yes
Qwen2.5-7B-Instruct	7.6	Q4	53.6 est.	Yes
Qwen2.5-Coder-7B-Instruct	7.6	Q4	53.6 est.	Yes
DeepSeek-R1-Distill-Qwen-7B	7.6	Q4	53.6 est.	Yes
Qwen3-8B	8.2	Q4	49.7 est.	Yes
DeepSeek-R1-0528-Qwen3-8B	8.2	Q4	49.7 est.	Yes
Qwen3-Coder-Next	79.7	Q4	48.1 est.	No
Qwen3-Next-80B-A3B-Instruct	81.3	Q4	46.3 est.	No
Mistral-7B-Instruct-v0.3	7.2	FP16	40.1 est.	No
Devstral-Small-2507	23.6	Q4	36.1 est.	No
Llama-3.2-3B-Instruct	3.2	FP16	35.0 est.	Yes
Mistral-Small-3.2-24B-Instruct-2506	24.0	Q4	34.4 est.	No
Magistral-Small-2509	24.0	Q4	34.4 est.	No
Qwen2.5-7B-Instruct	7.6	FP16	34.3 est.	No
Qwen2.5-Coder-7B-Instruct	7.6	FP16	34.3 est.	No
DeepSeek-R1-Distill-Qwen-7B	7.6	FP16	34.3 est.	No
gemma-3-12b-it	12.2	Q4	33.9 est.	Yes
Llama-3.1-8B-Instruct	8.0	FP16	29.6 est.	No
Phi-4-mini-instruct	3.8	FP16	29.5 est.	Yes
Phi-4-mini-reasoning	3.8	FP16	29.5 est.	Yes
gemma-3-4b-it	4.3	FP16	28.7 est.	Yes
Qwen3-4B-Instruct-2507	4.0	FP16	28.0 est.	Yes
Qwen3-4B-Thinking-2507	4.0	FP16	28.0 est.	Yes
Phi-4	14.7	Q4	27.7 est.	Yes
Phi-4-reasoning	14.7	Q4	27.7 est.	Yes
Qwen3-8B	8.2	FP16	27.6 est.	No
DeepSeek-R1-0528-Qwen3-8B	8.2	FP16	27.6 est.	No
Qwen3-14B	14.8	Q4	27.5 est.	Yes
Qwen2.5-14B-Instruct	14.8	Q4	27.5 est.	Yes
Qwen2.5-Coder-14B-Instruct	14.8	Q4	27.5 est.	Yes
DeepSeek-R1-Distill-Qwen-14B	14.8	Q4	27.5 est.	Yes
Mixtral-8x7B-Instruct-v0.1	46.7	Q4	27.0 est.	No
gemma-3-27b-it	27.4	Q4	23.7 est.	No
Qwen3-32B	32.8	Q4	13.9 est.	No
Qwen2.5-32B-Instruct	32.8	Q4	13.9 est.	No
Qwen2.5-Coder-32B-Instruct	32.8	Q4	13.9 est.	No
DeepSeek-R1-Distill-Qwen-32B	32.8	Q4	13.9 est.	No
gpt-oss-20b	21.5	FP16	11.4 est.	No
gemma-3-12b-it	12.2	FP16	8.8 est.	No
Qwen3-30B-A3B	30.5	FP16	7.0 est.	No
Qwen3-Coder-30B-A3B-Instruct	30.5	FP16	7.0 est.	No
Qwen3-30B-A3B-Instruct-2507	30.5	FP16	7.0 est.	No
GLM-4.7-Flash	31.2	FP16	6.7 est.	No
Phi-4	14.7	FP16	5.0 est.	No
Phi-4-reasoning	14.7	FP16	5.0 est.	No
Qwen3-14B	14.8	FP16	4.9 est.	No
Qwen2.5-14B-Instruct	14.8	FP16	4.9 est.	No
Qwen2.5-Coder-14B-Instruct	14.8	FP16	4.9 est.	No
DeepSeek-R1-Distill-Qwen-14B	14.8	FP16	4.9 est.	No
Llama-3.1-70B-Instruct	70.6	Q4	1.5 est.	No
Llama-3.3-70B-Instruct	70.6	Q4	1.5 est.	No
DeepSeek-R1-Distill-Llama-70B	70.6	Q4	1.4 est.	No
Qwen2.5-72B-Instruct	72.7	Q4	1.3 est.	No
Mistral-Small-3.2-24B-Instruct-2506	24.0	FP16	1.2 est.	No
Magistral-Small-2509	24.0	FP16	1.2 est.	No
Devstral-Small-2507	23.6	FP16	1.2 est.	No
gemma-3-27b-it	27.4	FP16	0.8 est.	No

Fit checks

Qwen3-0.6B — fits · ≈509.1 tok/s Q4 est.
Qwen3-8B — fits · ≈49.7 tok/s Q4 est.
Qwen2.5-7B-Instruct — fits · ≈53.6 tok/s Q4 est.
Llama-3.2-1B-Instruct — fits · ≈339.4 tok/s Q4 est.
Llama-3.1-8B-Instruct — fits · ≈89.1 tok/s Q4 measured
gpt-oss-20b — fits · ≈113.1 tok/s Q4 est.
Qwen3-32B — does not fit · ≈13.9 tok/s Q4 est.
Qwen3-1.7B — fits · ≈203.6 tok/s Q4 est.
Mistral-7B-Instruct-v0.3 — fits · ≈103.3 tok/s Q4 measured
Qwen3-14B — fits · ≈27.5 tok/s Q4 est.
Qwen3-4B-Instruct-2507 — fits · ≈101.8 tok/s Q4 est.

All 61 model fit checks — see the linked Fits? column in the table above