Intel · consumer

Arc A770 16GB

Arc A770 16GB has 16 GB of VRAM at 560 GB/s — about 14.88 GiB usable after driver and compositor overhead. 1856 of 2118 indexed models fit at 8K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
16 GB
GDDR6
Bandwidth
560 GB/s
256-bit bus
Tensor FP16
—
dense
TDP
225 W
$349 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1591video 15vision language 162embedding 26audio asr 39audio tts 21image 2

What fits at 8K context

largest quantization that fits, per model · 1856 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
GLM-Z1-Rumination-32B-0414IQ3_XS33.1B12.98 GiB1.01 GiB14.88 GiB0.00 GiB21±30%
MN-GRAND-23.5B-Gutenberg-UNCENSORED-V2-GLM4.7-ThinkingQ4_K_S23.4B12.68 GiB1.34 GiB14.87 GiB0.01 GiB21±30%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-Q2_K_S42.4B13.52 GiB0.56 GiB14.87 GiB0.01 GiB71±37%
dolphin-2.9.3-mistral-7B-32kF167.2B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
Mistral-7B-Instruct-v0.3-ParasiteF167.2B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
Mistral-7B-Instruct-v0.3-JbliteratedF167.2B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
Mistral-7B-Instruct-v0.3F167.2B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
Mistral-7B-v0.3F167.2B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
Mistral-7B-v0.3-Chinese-ChatF167.2B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
mistral-7b-v0.3-bnb-4bitBF167.5B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
Mathstral-7B-v0.1F167.2B13.50 GiB0.53 GiB14.87 GiB0.01 GiB21±30%
Fallen-Gemma3-27B-v1Q3_K_L27.4B13.54 GiB0.48 GiB14.87 GiB0.01 GiB21±30%
Qwen3.6-27B-A3B-CoderMoEI1-Q4_026.7B13.97 GiB0.08 GiB14.86 GiB0.02 GiB90±37%
Teuken-7B-instruct-research-v0.4F167.5B13.89 GiB0.13 GiB14.86 GiB0.02 GiB21±30%
Nous-Hermes-2-Yi-34BI1-IQ3_XXS34.4B12.98 GiB1.00 GiB14.86 GiB0.02 GiB21±30%
OpenChat-3.5-7B-Qwen-v2.0KV unresolvedF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
ContextualKunoichi_KTO-7BF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
mistral-7b-uncensoredKV unresolvedF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
xLAM-7b-rBF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
Yarn-Mistral-7b-128kKV unresolvedF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
MegaBeam-Mistral-7B-512kF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
Mistral-7B-Instruct-v0.2BF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
Ninja-v1-RP-WIPKV unresolvedF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
Silicon-Maid-7BKV unresolvedF167.2B13.49 GiB0.53 GiB14.86 GiB0.02 GiB21±30%
Qwen3-Coder-REAP-25B-A3BMoEQ4_K_S24.9B13.66 GiB0.40 GiB14.85 GiB0.03 GiB69±37%
Trinity-MiniMoEIQ4_NL26.1B13.92 GiB0.13 GiB14.84 GiB0.04 GiB83±37%
granite-4.0-h-smallMoEIQ3_M32.2B13.99 GiB0.07 GiB14.84 GiB0.04 GiB58±37%
Qwen-AgentWorld-35B-A3BMoEUD-IQ3_S34.7B13.96 GiB0.08 GiB14.84 GiB0.04 GiB106±37%
Ornith-1.0-35BMoEUD-IQ3_S34.7B13.96 GiB0.08 GiB14.84 GiB0.04 GiB106±37%
gemma-4-E4B-uncensoredF167.9B13.92 GiB0.09 GiB14.83 GiB0.05 GiB21±30%
gemma-4-E4B-it-qat-heretic_decensoredF167.9B13.92 GiB0.09 GiB14.83 GiB0.05 GiB21±30%
gemma-4-E4B-it-QAT-SOMPOA-heresyF167.9B13.92 GiB0.09 GiB14.83 GiB0.05 GiB21±30%
gemma-4-E4B-it-hereticBF168.0B13.92 GiB0.09 GiB14.83 GiB0.05 GiB21±30%
Tinman-gemma4-companion-mergedBF167.9B13.92 GiB0.09 GiB14.83 GiB0.05 GiB21±30%
LFM2-24B-A2BMoEQ4_123.8B13.93 GiB0.08 GiB14.83 GiB0.05 GiB85±37%
Gemma-4-Novelist-Eclipse-31BIQ3_XXS32.7B12.65 GiB1.29 GiB14.82 GiB0.06 GiB21±30%
Gemma-4-31B-StyleTuneIQ3_XXS32.7B12.65 GiB1.29 GiB14.82 GiB0.06 GiB21±30%
Skyfall-31B-v4.2Q3_K_S31.4B13.00 GiB0.90 GiB14.81 GiB0.07 GiB22±30%
Qwen3.5-27B-Engineer-Deckard-GeminiI1-IQ4_XS27.7B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredI1-IQ4_XS27.4B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-ThinkingI1-IQ4_XS27.4B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Huihui-Qwen3.5-27B-abliteratedI1-IQ4_XS27.8B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Qwen3.5-27B-Unredacted-MAXI1-IQ4_XS27.4B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Qwen3.5-27B-hereticI1-IQ4_XS27.4B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Qwen3.5-27B-DerestrictedI1-IQ4_XS27.8B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-DistilledI1-IQ4_XS27.8B13.68 GiB0.27 GiB14.81 GiB0.07 GiB21±30%
Wan2.2-S2V-14BQ5_K_M16.3B13.97 GiB0.00 GiB14.81 GiB0.07 GiB21±30%
Rocinante-XL-16B-v1Q6_K_L16.1B13.06 GiB0.90 GiB14.80 GiB0.08 GiB21±30%
gemma-4-26B-A4B-itMoEIQ4_NL26.5B13.69 GiB0.32 GiB14.80 GiB0.08 GiB21±30%
North-Mini-Code-1.0MoEQ3_K_L30.5B13.74 GiB0.28 GiB14.80 GiB0.08 GiB79±37%
v6-Finch-14B-HFQ5_K_L14.1B9.90 GiB4.05 GiB14.80 GiB0.08 GiB21±30%
DA3-BASEF32—13.94 GiB0.00 GiB14.79 GiB0.09 GiB21±30%
Qwen3-Coder-30B-A3B-InstructMoEQ3_K_L30.5B13.58 GiB0.40 GiB14.77 GiB0.11 GiB75±37%
Qwen3-VL-30B-A3B-ThinkingMoEQ3_K_L31.1B13.58 GiB0.40 GiB14.77 GiB0.11 GiB75±37%
MiroThinker-v1.0-30BMoEQ3_K_L30.5B13.58 GiB0.40 GiB14.77 GiB0.11 GiB75±37%
Qwen3-30B-A3BMoEQ3_K_L30.5B13.58 GiB0.40 GiB14.77 GiB0.11 GiB75±37%
Pantheon-Proto-RP-1.8-30B-A3BMoEQ3_K_L30.5B13.58 GiB0.40 GiB14.77 GiB0.11 GiB75±37%
Tongyi-DeepResearch-30B-A3BMoEQ3_K_L30.5B13.58 GiB0.40 GiB14.77 GiB0.11 GiB75±37%
internlm2-math-plus-20bI1-Q5_K_M19.9B13.11 GiB0.80 GiB14.77 GiB0.11 GiB22±30%
Muse-Glimmer-30BQ3_K_L29.8B13.77 GiB0.10 GiB14.75 GiB0.13 GiB22±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 A770 16GB run?
1856 of 2118 indexed open-weight models fit a Arc A770 16GB at 8,192 context with q8_0 KV cache, the largest being GLM-Z1-Rumination-32B-0414 at IQ3_XS. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc A770 16GB actually have?
Its nameplate is 16 GB, but about 14.88 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Arc A770 16GB fast for local AI?
Its memory bandwidth is 560 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.