Intel · workstation

Arc Pro B70 32GB

Arc Pro B70 32GB has 32 GB of VRAM at 608 GB/s — about 29.76 GiB usable after driver and compositor overhead. 1949 of 2118 indexed models fit at 128K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
GDDR6
Bandwidth
608 GB/s
256-bit bus
Tensor FP16
184 TF
dense
TDP
230 W
$949 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1668vision language 177audio tts 21video 16embedding 26image 2audio asr 39

What fits at 128K context

largest quantization that fits, per model · 1949 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-ThinkingQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.6-27B-Heretic2-ThinkingQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.6-27B-abliteratedQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Webcoda-AI-27BQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
KoQweopus-3.5-27B-experimentalQ8_027.8B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Huihui-Qwen3.6-27B-abliteratedQ8_027.8B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-27B-hereticQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Huihui-Qwen3.5-27B-abliteratedQ8_027.8B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-Queen-27BQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-27B-abliteratedQ8_026.9B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
ThinkingCap-Qwen3.6-27B-hereticQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
MusaCoder-27BQ8_0—26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen-Image-BenchQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-27B-uncensored-heretic-v1Q8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Bonsai-27B-unpackedQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Ternary-Bonsai-27B-unpackedQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-hereticQ8_027.4B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Darwin-28B-REASONQ8_026.9B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliteratedQ8_027.8B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-DistilledQ8_027.8B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-27BQ8_027.8B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Qwen3.5-27B-WebNovel-Writer-zhQ8_026.9B26.63 GiB2.25 GiB29.74 GiB0.02 GiB11±30%
Hunyuan-A13B-InstructMoEIQ2_M80.4B24.45 GiB4.50 GiB29.74 GiB0.02 GiB11±30%
Qwen3.6-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingQ5_K_S39.5B25.49 GiB3.38 GiB29.73 GiB0.03 GiB12±30%
Darwin-35B-A3B-OpusMoEQ6_K_L36.0B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
Aurora-Code-1MoEQ6_K_L34.7B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
grug-35b-v2MoEQ6_K_L35.1B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
grug-35bMoEQ6_K_L35.1B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
WorldSim-Opus-3.6-35B-A3BMoEQ6_K_L35.1B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
Qwen3.6-35B-A3B-AnkoMoEQ6_K_L35.1B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
KAT-Coder-V2.5-DevMoEQ6_K_L34.7B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
Ornith-1.0-35BMoEQ6_K_L34.7B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
Nex-N2-miniMoEQ6_K_L35.1B28.22 GiB0.70 GiB29.73 GiB0.03 GiB54±37%
Qwen3-TTS-12Hz-0.6B-BaseF32915M28.88 GiB0.00 GiB29.72 GiB0.04 GiB11±30%
Phi-3.5-MoE-instructMoEKV unresolvedQ4_141.9B24.41 GiB4.50 GiB29.71 GiB0.05 GiB21±37%
Nemotron-Mini-4B-InstructF164.2B24.38 GiB4.50 GiB29.70 GiB0.06 GiB11±30%
HomunculusBF1612.5B23.21 GiB5.63 GiB29.69 GiB0.07 GiB12±30%
glm-4-9b-chat-1mQ4_K_L9.5B6.30 GiB22.50 GiB29.65 GiB0.11 GiB12±30%
Gemma-4-Novelist-Eclipse-31BQ5_K_L32.7B22.77 GiB5.95 GiB29.60 GiB0.16 GiB12±30%
Gemma-4-31B-StyleTuneQ5_K_L32.7B22.77 GiB5.95 GiB29.60 GiB0.16 GiB12±30%
glm-4-9b-chat-abliteratedQ4_K_L9.4B6.25 GiB22.50 GiB29.60 GiB0.16 GiB12±30%
glm-4-9b-chatQ4_K_L9.4B6.25 GiB22.50 GiB29.60 GiB0.16 GiB12±30%
codegeex4-all-9bQ5_K_S9.4B6.23 GiB22.50 GiB29.58 GiB0.18 GiB12±30%
TildeOpen-30B-Instruct-LVI1-Q5_K_M30.7B20.26 GiB8.44 GiB29.58 GiB0.18 GiB12±30%
Bernini-RQ8_014.3B28.71 GiB0.00 GiB29.56 GiB0.20 GiB12±30%
Qwen3.6-28BMoEQ8_028.2B28.00 GiB0.70 GiB29.51 GiB0.25 GiB51±37%
ALIA-40b-fc-2606I1-Q4_K_S40.4B21.84 GiB6.75 GiB29.50 GiB0.26 GiB12±30%
ALIA-40b-instruct-2606I1-Q4_K_S40.4B21.84 GiB6.75 GiB29.50 GiB0.26 GiB12±30%
Skyfall-31B-v4.2Q5_K_M31.4B20.97 GiB7.59 GiB29.48 GiB0.28 GiB12±30%
Qwen3.5-99BMoEI1-IQ2_S99.0B27.80 GiB0.84 GiB29.47 GiB0.29 GiB49±37%
Qwen3.6-35B-A3BMoEUD-Q6_K36.0B27.95 GiB0.70 GiB29.46 GiB0.30 GiB54±37%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-Q4_K_M42.4B23.94 GiB4.71 GiB29.44 GiB0.32 GiB24±37%
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-Q5_K_S39.5B25.20 GiB3.38 GiB29.44 GiB0.32 GiB12±30%
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-Q5_K_S39.5B25.20 GiB3.38 GiB29.44 GiB0.32 GiB12±30%
Qwen3.5-40B-Claude-4.5-Opus-High-Reasoning-Thinking-uncensored-hereticQ5_K_S39.5B25.20 GiB3.38 GiB29.44 GiB0.32 GiB12±30%
Qwen3.6-27B-NVFP4NVFP421.2B26.29 GiB2.25 GiB29.40 GiB0.36 GiB12±30%
Laguna-XS-2.1MoEQ6_K_L33.4B27.14 GiB1.44 GiB29.37 GiB0.39 GiB46±37%
Noromaid-v0.4-Mixtral-Instruct-8x7b-ZlossMoEIQ4_XS46.7B24.02 GiB4.50 GiB29.36 GiB0.40 GiB16±37%
gemma-4-E4B-it-Uncensored-MAXF328.0B28.02 GiB0.51 GiB29.35 GiB0.41 GiB12±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 Pro B70 32GB run?
1949 of 2118 indexed open-weight models fit a Arc Pro B70 32GB at 131,072 context with q4_0 KV cache, the largest being Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensored at Q8_0. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc Pro B70 32GB actually have?
Its nameplate is 32 GB, but about 29.76 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Arc Pro B70 32GB fast for local AI?
Its memory bandwidth is 608 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.