NVIDIA · datacenter

Tesla V100 32GB

Tesla V100 32GB has 32 GB of VRAM at 900 GB/s — about 29.76 GiB usable after driver and compositor overhead. 2022 of 2118 indexed models fit at 8K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
HBM2
Bandwidth
900 GB/s
4096-bit bus
Tensor FP16
125 TF
dense
TDP
300 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
video 16text 1737vision language 181image 2embedding 26audio tts 21audio asr 39

What fits at 8K context

largest quantization that fits, per model · 2022 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Bernini-RQ8_014.3B28.71 GiB0.00 GiB29.76 GiB0.00 GiB18±22%
Maenad-70BI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Rombos-LLM-70b-Llama-3.3I1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
L3.3-Electra-R1-70bI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Llama-3.3_70_b_uncensored_continuedI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Llama-3.3-70B-Instruct-abliteratedI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
grok-oss-Revenant-70BI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
L3.3-70B-Euryale-v2.3I1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Hermes-3-Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Hermes-4-70B-hereticI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Llama-3.3-70B-InstructIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Anubis-70B-v1.2IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Hermes-4-70BIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Golem-70B-v1bI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
DeepSeek-R1-Distill-Llama-70B-hereticI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
llama-3-firefunction-v2IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Legion-V2.1-LLaMa-70BI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Assistant_Pepe_70BI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Tess-R1-Limerick-Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Infinity-Instruct-7M-Gen-Llama3_1-70BI1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
New-Dawn-Llama-3-70B-32K-v1.0I1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Meta-Llama-3-70B-Instruct-abliterated-v3.5I1-IQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Athene-70BIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
L3.3-70B-Magnum-DiamondIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Meta-Llama-3-70B-InstructIQ3_XS70.6B27.29 GiB1.33 GiB29.75 GiB0.01 GiB18±22%
Salience-1.5-ProMoEQ6_K_L36.0B28.66 GiB0.08 GiB29.75 GiB0.01 GiB105±37%
Qwable-v1MoEQ6_K_L36.0B28.66 GiB0.08 GiB29.75 GiB0.01 GiB105±37%
T-SearchMoEQ6_K_L36.0B28.66 GiB0.08 GiB29.75 GiB0.01 GiB105±37%
Melody1437-27BQ3_K_M27.8B28.40 GiB0.27 GiB29.73 GiB0.03 GiB18±22%
Nous-Hermes-Llama2-70bQ2_K69.0B27.27 GiB1.33 GiB29.72 GiB0.04 GiB18±22%
llama2_70b_chat_uncensoredQ2_K69.0B27.27 GiB1.33 GiB29.72 GiB0.04 GiB18±22%
Xwin-LM-70b-V0.1Q2_K69.0B27.27 GiB1.33 GiB29.72 GiB0.04 GiB18±22%
Llama-2-70b-chat-hfQ2_K69.0B27.27 GiB1.33 GiB29.72 GiB0.04 GiB18±22%
command-r-35b-writer-v2I1-Q5_K_M35.0B23.29 GiB5.31 GiB29.71 GiB0.05 GiB18±22%
Qwen3-53B-A3B-2507-THINKING-TOTAL-RECALL-v2-MASTER-CODERMoEI1-Q4_053.0B28.02 GiB0.70 GiB29.71 GiB0.05 GiB64±37%
Qwen2.5-7B-Instruct-1MF327.6B28.38 GiB0.23 GiB29.66 GiB0.10 GiB18±22%
DeepSeek-R1-Distill-Qwen-7BF327.6B28.38 GiB0.23 GiB29.66 GiB0.10 GiB18±22%
UI-TARS-7B-DPOF328.3B28.38 GiB0.23 GiB29.66 GiB0.10 GiB18±22%
Qwen2-7B-InstructF327.6B28.38 GiB0.23 GiB29.66 GiB0.10 GiB18±22%
Hercules-5.0-Qwen2-7BF327.6B28.38 GiB0.23 GiB29.66 GiB0.10 GiB18±22%
Kepler-8B-Instruct-v2F167.6B28.37 GiB0.23 GiB29.66 GiB0.10 GiB18±22%
MiniCPM-o-2_6F328.7B28.37 GiB0.23 GiB29.65 GiB0.11 GiB18±22%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-Q5_K_M42.4B28.05 GiB0.56 GiB29.60 GiB0.16 GiB73±37%
Mixtral-8x22B-Instruct-v0.1MoEIQ1_S141B27.61 GiB0.93 GiB29.60 GiB0.16 GiB32±37%
Mixtral-8x22B-v0.1MoEIQ1_S141B27.61 GiB0.93 GiB29.60 GiB0.16 GiB32±37%
CalmeRys-78B-Orpo-v0.1I1-IQ2_XS78.0B27.00 GiB1.43 GiB29.55 GiB0.21 GiB18±22%
calme-2.3-rys-78bIQ2_XS78.0B27.00 GiB1.43 GiB29.55 GiB0.21 GiB18±22%
Huihui-GLM-4.7-Flash-abliterated-57BMoEIQ4_XS57.3B27.93 GiB0.56 GiB29.53 GiB0.23 GiB73±37%
Qwen3.5-88BMoEI1-Q2_K_S87.7B28.30 GiB0.10 GiB29.42 GiB0.34 GiB94±37%
Kimi-Linear-48B-A3B-InstructMoEQ4_K_L49.1B28.26 GiB0.13 GiB29.39 GiB0.37 GiB18±22%
DeepCoder-14B-PreviewBF1614.8B27.52 GiB0.80 GiB29.36 GiB0.40 GiB18±22%
SuperNova-MediusF1614.8B27.52 GiB0.80 GiB29.36 GiB0.40 GiB18±22%
Qwen2.5-14B-Instruct-1MF1614.8B27.52 GiB0.80 GiB29.36 GiB0.40 GiB18±22%
OpenCodeReasoning-Nemotron-14BBF1614.8B27.52 GiB0.80 GiB29.36 GiB0.40 GiB18±22%
Qwen2.5-Coder-14B-Instruct-abliteratedF1614.8B27.52 GiB0.80 GiB29.36 GiB0.40 GiB18±22%
●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 Tesla V100 32GB run?
2022 of 2118 indexed open-weight models fit a Tesla V100 32GB at 8,192 context with q8_0 KV cache, the largest being Bernini-R at Q8_0. That covers text, vision-language, image, video and speech models.
How much usable memory does a Tesla V100 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 Tesla V100 32GB fast for local AI?
Its memory bandwidth is 900 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.