Arc B580 12GB

Intel · 12 GB GDDR6 · 456 GB/s · 190 W · MSRP $249 · Search Amazon (affiliate)

The Arc B580 12GB is a Intel accelerator with 12 GB GDDR6 and about 456 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 18 curated models fit fully on this card; at FP16, about 10 fit. Top Q4 speeds: Qwen3-0.6B ≈ 362.7 tok/s (estimated, fits); gemma-3-1b-it ≈ 290.2 tok/s (estimated, fits); Llama-3.2-1B-Instruct ≈ 241.8 tok/s (estimated, fits); Qwen3-1.7B ≈ 145.1 tok/s (estimated, fits); Phi-4-mini-instruct ≈ 76.4 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	362.7 est.	Yes
gemma-3-1b-it	1.0	Q4	290.2 est.	Yes
Llama-3.2-1B-Instruct	1.2	Q4	241.8 est.	Yes
gpt-oss-20b	21.5	Q4	180.6 est.	No
Qwen3-1.7B	2.0	Q4	145.1 est.	Yes
Qwen3-30B-A3B	30.5	Q4	117.8 est.	No
Qwen3-Coder-30B-A3B-Instruct	30.5	Q4	117.8 est.	No
Qwen3-30B-A3B-Instruct-2507	30.5	Q4	117.8 est.	No
GLM-4.7-Flash	31.2	Q4	113.1 est.	No
Qwen3-0.6B	0.8	FP16	99.8 est.	Yes
gemma-3-1b-it	1.0	FP16	79.8 est.	Yes
Phi-4-mini-instruct	3.8	Q4	76.4 est.	Yes
Phi-4-mini-reasoning	3.8	Q4	76.4 est.	Yes
gemma-3-4b-it	4.3	Q4	74.4 est.	Yes
Qwen3-4B-Instruct-2507	4.0	Q4	72.5 est.	Yes
Qwen3-4B-Thinking-2507	4.0	Q4	72.5 est.	Yes
Llama-3.2-1B-Instruct	1.2	FP16	66.5 est.	Yes
Phi-4	14.7	Q4	56.1 est.	No
Phi-4-reasoning	14.7	Q4	56.1 est.	No
Qwen3-14B	14.8	Q4	55.1 est.	No
Qwen2.5-14B-Instruct	14.8	Q4	55.1 est.	No
Qwen2.5-Coder-14B-Instruct	14.8	Q4	55.1 est.	No
DeepSeek-R1-Distill-Qwen-14B	14.8	Q4	55.1 est.	No
Llama-3.2-3B-Instruct	3.2	Q4	47.4 measured (geerlingguy-ai-benchmarks)	Yes
Mistral-7B-Instruct-v0.3	7.2	Q4	40.3 est.	Yes
Qwen3-1.7B	2.0	FP16	39.9 est.	Yes
Qwen2.5-7B-Instruct	7.6	Q4	38.2 est.	Yes
Qwen2.5-Coder-7B-Instruct	7.6	Q4	38.2 est.	Yes
DeepSeek-R1-Distill-Qwen-7B	7.6	Q4	38.2 est.	Yes
Llama-3.1-8B-Instruct	8.0	Q4	36.3 est.	Yes
Qwen3-8B	8.2	Q4	35.4 est.	Yes
DeepSeek-R1-0528-Qwen3-8B	8.2	Q4	35.4 est.	Yes
Llama-3.2-3B-Instruct	3.2	FP16	24.9 est.	Yes
gemma-3-12b-it	12.2	Q4	24.2 est.	Yes
Phi-4-mini-instruct	3.8	FP16	21.0 est.	Yes
Phi-4-mini-reasoning	3.8	FP16	21.0 est.	Yes
gemma-3-4b-it	4.3	FP16	20.5 est.	Yes
Qwen3-4B-Instruct-2507	4.0	FP16	19.9 est.	Yes
Qwen3-4B-Thinking-2507	4.0	FP16	19.9 est.	Yes
Qwen3-Coder-Next	79.7	Q4	19.3 est.	No
Qwen3-Next-80B-A3B-Instruct	81.3	Q4	18.5 est.	No
Mistral-7B-Instruct-v0.3	7.2	FP16	16.1 est.	No
Devstral-Small-2507	23.6	Q4	14.5 est.	No
Qwen2.5-7B-Instruct	7.6	FP16	13.8 est.	No
Qwen2.5-Coder-7B-Instruct	7.6	FP16	13.8 est.	No
DeepSeek-R1-Distill-Qwen-7B	7.6	FP16	13.8 est.	No
Mistral-Small-3.2-24B-Instruct-2506	24.0	Q4	13.8 est.	No
Magistral-Small-2509	24.0	Q4	13.8 est.	No
Llama-3.1-8B-Instruct	8.0	FP16	11.9 est.	No
Qwen3-8B	8.2	FP16	11.0 est.	No
DeepSeek-R1-0528-Qwen3-8B	8.2	FP16	11.0 est.	No
Mixtral-8x7B-Instruct-v0.1	46.7	Q4	10.8 est.	No
gemma-3-27b-it	27.4	Q4	9.5 est.	No
Qwen3-32B	32.8	Q4	5.6 est.	No
Qwen2.5-32B-Instruct	32.8	Q4	5.6 est.	No
Qwen2.5-Coder-32B-Instruct	32.8	Q4	5.6 est.	No
DeepSeek-R1-Distill-Qwen-32B	32.8	Q4	5.6 est.	No
gpt-oss-20b	21.5	FP16	4.6 est.	No
gemma-3-12b-it	12.2	FP16	3.5 est.	No
Qwen3-14B	14.8	FP16	2.0 est.	No
Qwen2.5-14B-Instruct	14.8	FP16	2.0 est.	No
Qwen2.5-Coder-14B-Instruct	14.8	FP16	2.0 est.	No
DeepSeek-R1-Distill-Qwen-14B	14.8	FP16	2.0 est.	No
Phi-4	14.7	FP16	2.0 est.	No
Phi-4-reasoning	14.7	FP16	2.0 est.	No

Fit checks

• Qwen3-0.6B — fits · ≈362.7 tok/s Q4 est.
• Qwen3-8B — fits · ≈35.4 tok/s Q4 est.
• Qwen2.5-7B-Instruct — fits · ≈38.2 tok/s Q4 est.
• Llama-3.2-1B-Instruct — fits · ≈241.8 tok/s Q4 est.
• Llama-3.1-8B-Instruct — fits · ≈36.3 tok/s Q4 est.
• gpt-oss-20b — does not fit · ≈180.6 tok/s Q4 est.
• Qwen3-32B — does not fit · ≈5.6 tok/s Q4 est.
• Qwen3-1.7B — fits · ≈145.1 tok/s Q4 est.
• Mistral-7B-Instruct-v0.3 — fits · ≈40.3 tok/s Q4 est.
• Qwen3-14B — does not fit · ≈55.1 tok/s Q4 est.
• Qwen3-4B-Instruct-2507 — fits · ≈72.5 tok/s Q4 est.

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