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. 2062 of 2118 indexed models fit at 8K context with q8_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 1773vision language 185image 2audio asr 39audio tts 21video 16embedding 26

What fits at 8K context

largest quantization that fits, per model · 2062 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
gpt-oss-120bMoEQ4_K_M120B58.46 GiB0.16 GiB59.51 GiB0.01 GiB97±37%
gpt-oss-safeguard-120bMoEQ4_K_M120B58.46 GiB0.16 GiB59.51 GiB0.01 GiB97±37%
gemma-4-31B-itBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma4-Gutenberg-31BBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Wanabi-Gemma4-31BBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma4-Gutenberg-31B-HereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
G4-MeroMero-31B-uncensored-hereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31B-it-Claude-Opus-Distill-v2BF1632.7B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31B-it-uncensored-hereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma4-31B-Fable-5-DistilledF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Glistening-Gem-31B-v1.0BF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma-4-31B-Fable-5-Agent-DistillBF1632.7B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
G4-MeroMero-31BBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
G4-Alice-v1.2-31BBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-hereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Equinox-31BBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma-4-Gembrain-31B-it-uncensored-hereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma-4-31B-Storymaxxed3F1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma-4-Harmonia-31B-uncensored-hereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31B-MergemaxxedF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31B-it-qat-q4_0-unquantized-hereticF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31B-it-abliteratedF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma-4-Garnet-V2-31B-it-ultra-uncensored-hereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma-4-Queen-31B-it-uncensored-hereticBF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
Gemma-4-31B-storymaxxed2F1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31B-it-SDFT-Heretic-RPBF1630.7B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31B-anthologyF1631.3B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gemma-4-31BBF1632.7B57.20 GiB1.29 GiB59.46 GiB0.06 GiB18±26.5%
gpt-oss-120b-uncensored-bf16MoEIQ4_NL117B58.40 GiB0.16 GiB59.46 GiB0.06 GiB97±37%
Qwen3.5-REAP-212B-A17BMoEIQ2_XS212B58.30 GiB0.12 GiB59.37 GiB0.15 GiB92±37%
grok-2MoEIQ1_M270B57.16 GiB1.06 GiB59.27 GiB0.25 GiB32±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ2_XXS235B57.55 GiB0.78 GiB59.27 GiB0.25 GiB71±37%
command-a-plus-05-2026-bf16MoEIQ2_XXS219B57.98 GiB0.36 GiB59.24 GiB0.28 GiB75±37%
GLM-4.6VMoEQ4_0108B57.48 GiB0.76 GiB59.17 GiB0.35 GiB72±37%
Mixtral-8x22B-Instruct-v0.1MoEQ3_K_S141B57.28 GiB0.93 GiB59.17 GiB0.35 GiB32±37%
Mixtral-8x22B-v0.1MoEQ3_K_S141B57.28 GiB0.93 GiB59.17 GiB0.35 GiB32±37%
Hypernova-60B-2605MoEQ8_058.7B58.13 GiB0.15 GiB59.17 GiB0.35 GiB84±37%
Mixtral-8x22B-v0.1MoEQ3_K_S141B57.27 GiB0.93 GiB59.16 GiB0.36 GiB32±37%
Qwen3.5-88BMoEI1-Q5_K_M87.7B58.10 GiB0.10 GiB59.13 GiB0.39 GiB88±37%
Devstral-2-123B-Instruct-2512Q3_K_M125B56.46 GiB1.46 GiB58.98 GiB0.54 GiB18±26.5%
Mistral-Medium-3.5-128BI1-Q3_K_M128B56.46 GiB1.46 GiB58.97 GiB0.55 GiB18±26.5%
XORTRON-NXTXPRTXXLI1-Q3_K_M128B56.46 GiB1.46 GiB58.97 GiB0.55 GiB18±26.5%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ4_K_S109B57.23 GiB0.80 GiB58.96 GiB0.56 GiB72±37%
GLM-4.7-REAP-218B-A32BMoEUD-IQ1_S218B56.46 GiB1.53 GiB58.92 GiB0.60 GiB59±37%
Qwen3.5-99BMoEI1-Q4_199.0B57.88 GiB0.10 GiB58.91 GiB0.61 GiB92±37%
Assistant_Pepe_70BQ6_K_L70.6B56.43 GiB1.33 GiB58.78 GiB0.74 GiB18±26.5%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q3_K_M124B57.47 GiB0.37 GiB58.73 GiB0.79 GiB86±37%
dots.llm1.instMoEIQ2_M143B53.68 GiB4.12 GiB58.73 GiB0.79 GiB57±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-IQ3_M139B56.81 GiB1.03 GiB58.73 GiB0.79 GiB75±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-IQ3_M139B56.81 GiB1.03 GiB58.73 GiB0.79 GiB75±37%
Qwen3.5-122B-A10BMoEUD-IQ4_XS125B57.67 GiB0.10 GiB58.70 GiB0.82 GiB99±37%
MiniMax-M2.7MoEIQ2_XXS229B56.67 GiB1.03 GiB58.58 GiB0.94 GiB87±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ2_XXS236B57.28 GiB0.28 GiB58.50 GiB1.02 GiB92±37%
DeepSeek-V2.5MoEIQ2_XXS236B57.28 GiB0.28 GiB58.50 GiB1.02 GiB92±37%
DeepSeek-Coder-V2-InstructMoEIQ2_XXS236B57.28 GiB0.28 GiB58.50 GiB1.02 GiB92±37%
GLM-4.5-Air-DerestrictedMoEIQ4_XS110B56.63 GiB0.76 GiB58.32 GiB1.20 GiB72±37%
GLM-4.5-AirMoEIQ4_XS110B56.63 GiB0.76 GiB58.32 GiB1.20 GiB72±37%
Qwen3-Coder-30B-A3B-InstructMoEBF1630.5B56.90 GiB0.40 GiB58.19 GiB1.33 GiB75±37%
Salience-1.5-FlashMoEBF1631.1B56.90 GiB0.40 GiB58.19 GiB1.33 GiB75±37%
Qwen3-VL-30B-A3B-InstructMoEBF1631.1B56.90 GiB0.40 GiB58.19 GiB1.33 GiB75±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?
2062 of 2118 indexed open-weight models fit a Instinct MI210 at 8,192 context with q8_0 KV cache, the largest being gpt-oss-120b at Q4_K_M. 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.