AMD · datacenter

Instinct MI210

Instinct MI210 has 64 GB of VRAM at 1638 GB/s — about 59.52 GiB usable after driver and compositor overhead. 2063 of 2118 indexed models fit at 8K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
64 GB
HBM2e
Bandwidth
1638 GB/s
4096-bit bus
Tensor FP16
181 TF
dense
TDP
300 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1774vision language 185image 2audio asr 39audio tts 21video 16embedding 26

What fits at 8K context

largest quantization that fits, per model · 2063 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
gpt-oss-120bMoEQ2_K_L120B58.54 GiB0.09 GiB59.52 GiB0.00 GiB98±37%
gpt-oss-safeguard-120bMoEQ2_K_L120B58.54 GiB0.09 GiB59.52 GiB0.00 GiB98±37%
llm-surgery-dark-arts-gpt-oss-60b-96a12MoEI1-Q6_K60.9B58.56 GiB0.06 GiB59.50 GiB0.02 GiB54±37%
gpt-oss-120b-uncensored-bf16MoEQ4_K_M117B58.53 GiB0.09 GiB59.50 GiB0.02 GiB98±37%
MiMo-V2-FlashMoEKV unresolvedI1-IQ1_S310B58.20 GiB0.26 GiB59.42 GiB0.10 GiB95±37%
Qwen3.5-REAP-212B-A17BMoEIQ2_XS212B58.30 GiB0.07 GiB59.32 GiB0.20 GiB93±37%
dots.llm1.instMoEQ2_K143B56.10 GiB2.18 GiB59.21 GiB0.31 GiB68±37%
Hypernova-60B-2605MoEQ8_058.7B58.13 GiB0.08 GiB59.10 GiB0.42 GiB84±37%
Qwen3.5-88BMoEI1-Q5_K_M87.7B58.10 GiB0.05 GiB59.09 GiB0.43 GiB88±37%
command-a-plus-05-2026-bf16MoEIQ2_XXS219B57.98 GiB0.19 GiB59.07 GiB0.45 GiB76±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ2_XXS235B57.55 GiB0.41 GiB58.90 GiB0.62 GiB75±37%
Qwen3.5-99BMoEI1-Q4_199.0B57.88 GiB0.05 GiB58.86 GiB0.66 GiB93±37%
gemma-4-31B-itBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma4-Gutenberg-31BBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Wanabi-Gemma4-31BBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma4-Gutenberg-31B-HereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
G4-MeroMero-31B-uncensored-hereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31B-it-Claude-Opus-Distill-v2BF1632.7B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31B-it-uncensored-hereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma4-31B-Fable-5-DistilledF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Glistening-Gem-31B-v1.0BF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma-4-31B-Fable-5-Agent-DistillBF1632.7B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
G4-MeroMero-31BBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
G4-Alice-v1.2-31BBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-hereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Equinox-31BBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma-4-Gembrain-31B-it-uncensored-hereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma-4-31B-Storymaxxed3F1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma-4-Harmonia-31B-uncensored-hereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31B-MergemaxxedF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31B-it-qat-q4_0-unquantized-hereticF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31B-it-abliteratedF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma-4-Garnet-V2-31B-it-ultra-uncensored-hereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma-4-Queen-31B-it-uncensored-hereticBF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Gemma-4-31B-storymaxxed2F1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31B-it-SDFT-Heretic-RPBF1630.7B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31B-anthologyF1631.3B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
gemma-4-31BBF1632.7B57.20 GiB0.68 GiB58.86 GiB0.66 GiB18±26.5%
Step-3.7-FlashIQ2_XS201B56.76 GiB1.13 GiB58.82 GiB0.70 GiB18±26.5%
GLM-4.6VMoEQ4_0108B57.48 GiB0.40 GiB58.81 GiB0.71 GiB75±37%
grok-2MoEIQ1_M270B57.16 GiB0.56 GiB58.77 GiB0.75 GiB33±37%
Mixtral-8x22B-Instruct-v0.1MoEQ3_K_S141B57.28 GiB0.49 GiB58.73 GiB0.79 GiB33±37%
Mixtral-8x22B-v0.1MoEQ3_K_S141B57.28 GiB0.49 GiB58.73 GiB0.79 GiB33±37%
Mixtral-8x22B-v0.1MoEQ3_K_S141B57.27 GiB0.49 GiB58.73 GiB0.79 GiB33±37%
Qwen3.5-122B-A10BMoEUD-IQ4_XS125B57.67 GiB0.05 GiB58.66 GiB0.86 GiB99±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ4_K_S109B57.23 GiB0.42 GiB58.58 GiB0.94 GiB75±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q3_K_M124B57.47 GiB0.19 GiB58.56 GiB0.96 GiB89±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ2_XXS236B57.28 GiB0.15 GiB58.36 GiB1.16 GiB93±37%
DeepSeek-V2.5MoEIQ2_XXS236B57.28 GiB0.15 GiB58.36 GiB1.16 GiB93±37%
DeepSeek-Coder-V2-InstructMoEIQ2_XXS236B57.28 GiB0.15 GiB58.36 GiB1.16 GiB93±37%
Devstral-2-123B-Instruct-2512Q3_K_M125B56.46 GiB0.77 GiB58.29 GiB1.23 GiB18±26.5%
Mistral-Medium-3.5-128BI1-Q3_K_M128B56.46 GiB0.77 GiB58.29 GiB1.23 GiB18±26.5%
XORTRON-NXTXPRTXXLI1-Q3_K_M128B56.46 GiB0.77 GiB58.29 GiB1.23 GiB18±26.5%
MiniMax-M2.1-REAP-139B-A10BMoEI1-IQ3_M139B56.81 GiB0.54 GiB58.24 GiB1.28 GiB80±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-IQ3_M139B56.81 GiB0.54 GiB58.24 GiB1.28 GiB80±37%
GLM-4.7-REAP-218B-A32BMoEUD-IQ1_S218B56.46 GiB0.81 GiB58.20 GiB1.32 GiB64±37%
Assistant_Pepe_70BQ6_K_L70.6B56.43 GiB0.70 GiB58.16 GiB1.36 GiB18±26.5%
MiniMax-M2.7MoEIQ2_XXS229B56.67 GiB0.54 GiB58.10 GiB1.42 GiB93±37%
Qwen3-Coder-30B-A3B-InstructMoEBF1630.5B56.90 GiB0.21 GiB58.00 GiB1.52 GiB77±37%
Salience-1.5-FlashMoEBF1631.1B56.90 GiB0.21 GiB58.00 GiB1.52 GiB77±37%
●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 Instinct MI210 run?
2063 of 2118 indexed open-weight models fit a Instinct MI210 at 8,192 context with q4_0 KV cache, the largest being gpt-oss-120b at Q2_K_L. That covers text, vision-language, image, video and speech models.
How much usable memory does a Instinct MI210 actually have?
Its nameplate is 64 GB, but about 59.52 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Instinct MI210 fast for local AI?
Its memory bandwidth is 1638 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.