Intel · consumer

Arc B570 10GB

Arc B570 10GB has 10 GB of VRAM at 380 GB/s — about 9.30 GiB usable after driver and compositor overhead. 1211 of 2118 indexed models fit at 32K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
10 GB
GDDR6
Bandwidth
380 GB/s
160-bit bus
Tensor FP16
—
dense
TDP
150 W
$219 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1014vision language 100embedding 26image 1video 12audio asr 38audio tts 20

What fits at 32K context

largest quantization that fits, per model · 1211 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Nexa-AI-4x4B-InstructMoEI1-Q2_K_S12.1B3.99 GiB4.50 GiB9.30 GiB0.00 GiB18±37%
OmniAtlas-Qwen3-30B-A3BI1-IQ2_XS31.7B8.45 GiB0.00 GiB9.30 GiB0.00 GiB24±30%
Qwen3-Omni-30B-A3B-CaptionerI1-IQ2_XS31.7B8.45 GiB0.00 GiB9.30 GiB0.00 GiB24±30%
Qwen3-VL-4B-ThinkingQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-VL-4B-Instruct-Uncensored-abliteratedQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
PopiT-Qwen3-4B-Medical-SFT-1128Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-VL-4B-Instruct-Unredacted-MAXQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Zubr1.0-VL-4BQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-VL-4B-Thinking-Unredacted-MAXQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Huihui-Qwen3-VL-4B-Instruct-abliteratedQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-VL-4B-Instruct-UncensoredQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-VL-4B-InstructQ8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Logics-Parsing-v2Q8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Jan-v1-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
OpenCaption-4B-VL-SFT-v1.0Q8_04.4B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Parable-Qwen3-4B-Claude-Fable-5Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Z-Image-Engineer-V6Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4b-Z-Image-Turbo-AbliteratedV1Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Huihui-Qwen3-4B-Instruct-2507-abliteratedQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Josiefied-Qwen3-4B-abliterated-v2Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-Thinking-2507Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Jan-nano-128kQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-Instruct-2507Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Jan-nanoQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-abliteratedQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-Qwen3.6-plus-Reasoning-DistilledQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-Kimi2.5-Reasoning-DistilledQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Neuron-4B-InstructQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
ChineseErrorCorrector4-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-Instruct_NSFW-V2.1Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
FastContext-1.0-4B-SFT-abliteratedQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
FastContext-1.0-4B-SFTQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-abliterated-v2Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
LocoOperator-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
fable-traces-abliteratedQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Nexa-AI-4B-InstructQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
fable-tracesQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-Instruct-2507-hereticQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Lumen-4B-InstructQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
CyberSecQwen-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-HereticLM-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Huihui-Qwen3-4B-abliterated-v2Q8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-4B-BaseQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-Reranker-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Ternary-Bonsai-4B-unpackedQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Octen-Embedding-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Qwen3-Embedding-4BQ8_04.0B3.99 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Luna-7B-A4BMoEI1-Q4_16.7B3.98 GiB4.50 GiB9.30 GiB0.00 GiB18±37%
MiMo-VL-7B-RLI1-IQ4_XS8.3B3.97 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
Kuwutu-7B-CYOA-v2I1-IQ4_XS7.6B3.97 GiB4.50 GiB9.30 GiB0.00 GiB24±30%
internlm3-8b-instructQ6_K_L8.8B6.97 GiB1.50 GiB9.29 GiB0.01 GiB24±30%
Bonsai-8B-unpackedQ3_K_M8.2B3.96 GiB4.50 GiB9.29 GiB0.01 GiB24±30%
INTELLECT-1-InstructI1-IQ2_S10.2B3.20 GiB5.25 GiB9.29 GiB0.01 GiB24±30%
Hubble-4B-v1Q8_04.5B4.47 GiB4.00 GiB9.29 GiB0.01 GiB24±30%
Aura-4BQ8_04.5B4.47 GiB4.00 GiB9.29 GiB0.01 GiB24±30%
magnum-v2-4bQ8_04.5B4.47 GiB4.00 GiB9.29 GiB0.01 GiB24±30%
Impish_LLAMA_4BQ8_04.5B4.47 GiB4.00 GiB9.29 GiB0.01 GiB24±30%
Llama-3.1-Minitron-4B-Width-BaseQ8_04.5B4.47 GiB4.00 GiB9.29 GiB0.01 GiB24±30%
Falcon3-10B-InstructI1-Q2_K_S10.3B3.42 GiB5.00 GiB9.28 GiB0.02 GiB24±30%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Questions people ask

What AI models can a Arc B570 10GB run?
1211 of 2118 indexed open-weight models fit a Arc B570 10GB at 32,768 context with f16 KV cache, the largest being Nexa-AI-4x4B-Instruct at I1-Q2_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc B570 10GB actually have?
Its nameplate is 10 GB, but about 9.30 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Arc B570 10GB fast for local AI?
Its memory bandwidth is 380 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.