AMD · workstation

Radeon AI Pro R9700S

Radeon AI Pro R9700S has 32 GB of VRAM at 640 GB/s — about 29.76 GiB usable after driver and compositor overhead. 2022 of 2118 indexed models fit at 16K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
GDDR6
Bandwidth
640 GB/s
256-bit bus
Tensor FP16
—
dense
TDP
300 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1737video 16vision language 181image 2embedding 26audio tts 21audio asr 39

What fits at 16K context

largest quantization that fits, per model · 2022 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Hypernova-60B-2605MoEI1-Q3_K_S58.7B28.72 GiB0.15 GiB29.76 GiB0.00 GiB66±37%
Seed-OSS-36B-Instruct-biprojected-norm-preserving-abliteratedI1-Q6_K36.2B27.63 GiB1.13 GiB29.75 GiB0.01 GiB14±26.5%
Seed-OSS-36B-InstructQ6_K36.2B27.63 GiB1.13 GiB29.75 GiB0.01 GiB14±26.5%
Hermes-4.3-36B-hereticI1-Q6_K36.2B27.63 GiB1.13 GiB29.75 GiB0.01 GiB14±26.5%
Hermes-4.3-36BQ6_K36.2B27.63 GiB1.13 GiB29.75 GiB0.01 GiB14±26.5%
Seed-OSS-36B-BaseQ6_K36.2B27.63 GiB1.13 GiB29.75 GiB0.01 GiB14±26.5%
Llama-3_3-Nemotron-Super-49B-v1_5Q2_K49.9B17.45 GiB11.25 GiB29.74 GiB0.02 GiB14±26.5%
Valkyrie-49B-v2.1I1-Q2_K49.9B17.45 GiB11.25 GiB29.74 GiB0.02 GiB14±26.5%
Llama-3_3-Nemotron-Super-49B-v1Q2_K49.9B17.45 GiB11.25 GiB29.74 GiB0.02 GiB14±26.5%
Maenad-70BI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Rombos-LLM-70b-Llama-3.3I1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
L3.3-Electra-R1-70bI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Llama-3.3_70_b_uncensored_continuedI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Llama-3.3-70B-Instruct-abliteratedI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
grok-oss-Revenant-70BI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
L3.3-70B-Euryale-v2.3I1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Hermes-3-Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Hermes-4-70B-hereticI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Llama-3.3-70B-InstructIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Anubis-70B-v1.2IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Hermes-4-70BIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Golem-70B-v1bI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
DeepSeek-R1-Distill-Llama-70B-hereticI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
llama-3-firefunction-v2IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Legion-V2.1-LLaMa-70BI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Assistant_Pepe_70BI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Tess-R1-Limerick-Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Infinity-Instruct-7M-Gen-Llama3_1-70BI1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
New-Dawn-Llama-3-70B-32K-v1.0I1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Meta-Llama-3-70B-Instruct-abliterated-v3.5I1-IQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Athene-70BIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
L3.3-70B-Magnum-DiamondIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Meta-Llama-3-70B-InstructIQ3_XS70.6B27.29 GiB1.41 GiB29.73 GiB0.03 GiB14±26.5%
Snowpiercer-15B-v4BF1615.0B27.90 GiB0.88 GiB29.73 GiB0.03 GiB14±26.5%
Nous-Hermes-Llama2-70bQ2_K69.0B27.27 GiB1.41 GiB29.70 GiB0.06 GiB14±26.5%
llama2_70b_chat_uncensoredQ2_K69.0B27.27 GiB1.41 GiB29.70 GiB0.06 GiB14±26.5%
Xwin-LM-70b-V0.1Q2_K69.0B27.27 GiB1.41 GiB29.70 GiB0.06 GiB14±26.5%
Llama-2-70b-chat-hfQ2_K69.0B27.27 GiB1.41 GiB29.70 GiB0.06 GiB14±26.5%
Huihui-Qwen3-Coder-Next-abliteratedMoEI1-IQ3_XXS79.7B28.68 GiB0.11 GiB29.68 GiB0.08 GiB88±37%
Bernini-RQ8_014.3B28.71 GiB0.00 GiB29.66 GiB0.10 GiB14±26.5%
Salience-1.5-ProMoEQ6_K_L36.0B28.66 GiB0.09 GiB29.65 GiB0.11 GiB79±37%
Qwable-v1MoEQ6_K_L36.0B28.66 GiB0.09 GiB29.65 GiB0.11 GiB79±37%
T-SearchMoEQ6_K_L36.0B28.66 GiB0.09 GiB29.65 GiB0.11 GiB79±37%
Melody1437-27BQ3_K_M27.8B28.40 GiB0.28 GiB29.65 GiB0.11 GiB14±26.5%
Qwen3-53B-A3B-2507-THINKING-TOTAL-RECALL-v2-MASTER-CODERMoEI1-Q4_053.0B28.02 GiB0.74 GiB29.65 GiB0.11 GiB49±37%
Gemma-3-27B-MeditronFOQ8_028.8B28.13 GiB0.52 GiB29.63 GiB0.13 GiB14±26.5%
Qwen2.5-7B-Instruct-1MF327.6B28.38 GiB0.25 GiB29.58 GiB0.18 GiB14±26.5%
DeepSeek-R1-Distill-Qwen-7BF327.6B28.38 GiB0.25 GiB29.58 GiB0.18 GiB14±26.5%
UI-TARS-7B-DPOF328.3B28.38 GiB0.25 GiB29.58 GiB0.18 GiB14±26.5%
Qwen2-7B-InstructF327.6B28.38 GiB0.25 GiB29.58 GiB0.18 GiB14±26.5%
Hercules-5.0-Qwen2-7BF327.6B28.38 GiB0.25 GiB29.58 GiB0.18 GiB14±26.5%
Kepler-8B-Instruct-v2F167.6B28.37 GiB0.25 GiB29.57 GiB0.19 GiB14±26.5%
MiniCPM-o-2_6F328.7B28.37 GiB0.25 GiB29.57 GiB0.19 GiB14±26.5%
Mixtral-8x22B-Instruct-v0.1MoEIQ1_S141B27.61 GiB0.98 GiB29.56 GiB0.20 GiB25±37%
Mixtral-8x22B-v0.1MoEIQ1_S141B27.61 GiB0.98 GiB29.55 GiB0.21 GiB25±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 Radeon AI Pro R9700S run?
2022 of 2118 indexed open-weight models fit a Radeon AI Pro R9700S at 16,384 context with q4_0 KV cache, the largest being Hypernova-60B-2605 at I1-Q3_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon AI Pro R9700S 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 Radeon AI Pro R9700S fast for local AI?
Its memory bandwidth is 640 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.