NVIDIA · datacenter

Tesla V100 16GB

Tesla V100 16GB has 16 GB of VRAM at 900 GB/s — about 14.88 GiB usable after driver and compositor overhead. 1865 of 2118 indexed models fit at 4K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
16 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
vision language 162text 1600video 15audio asr 39image 2embedding 26audio tts 21

What fits at 4K context

largest quantization that fits, per model · 1865 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Darwin-35B-A3B-OpusMoEIQ3_XXS36.0B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
Aurora-Code-1MoEIQ3_XXS34.7B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
grug-35b-v2MoEIQ3_XXS35.1B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
grug-35bMoEIQ3_XXS35.1B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
WorldSim-Opus-3.6-35B-A3BMoEIQ3_XXS35.1B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
Qwen3.6-35B-A3B-AnkoMoEIQ3_XXS35.1B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
KAT-Coder-V2.5-DevMoEIQ3_XXS34.7B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
Ornith-1.0-35BMoEIQ3_XXS34.7B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
Nex-N2-miniMoEIQ3_XXS35.1B13.85 GiB0.02 GiB14.88 GiB0.00 GiB206±37%
Qwen3-14B-GPT-5.2-High-Reasoning-DistillQ3_K_M14.8B13.64 GiB0.18 GiB14.87 GiB0.01 GiB37±22%
granite-20b-code-instruct-8kQ5_K_M20.1B13.79 GiB0.00 GiB14.87 GiB0.01 GiB37±22%
granite-20b-code-base-8kI1-Q5_K_M20.1B13.79 GiB0.00 GiB14.87 GiB0.01 GiB37±22%
ALIA-40b-fc-2606I1-IQ2_M40.4B13.54 GiB0.21 GiB14.87 GiB0.01 GiB37±22%
ALIA-40b-instruct-2606I1-IQ2_M40.4B13.54 GiB0.21 GiB14.87 GiB0.01 GiB37±22%
Salience-1.5-FlashMoEIQ3_M31.1B13.77 GiB0.11 GiB14.87 GiB0.01 GiB154±37%
granite-34b-code-base-8kI1-IQ3_S33.7B13.79 GiB0.00 GiB14.87 GiB0.01 GiB37±22%
Magistry-24B-v1.1Q4_K_M23.6B13.57 GiB0.18 GiB14.87 GiB0.01 GiB37±22%
Olmo-3.1-32B-InstructIQ3_M32.2B13.48 GiB0.28 GiB14.86 GiB0.02 GiB37±22%
Olmo-3.1-32B-ThinkIQ3_M32.2B13.48 GiB0.28 GiB14.86 GiB0.02 GiB37±22%
Olmo-3-32B-ThinkIQ3_M32.2B13.48 GiB0.28 GiB14.86 GiB0.02 GiB37±22%
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoEI1-Q3_K_M30.0B13.64 GiB0.21 GiB14.86 GiB0.02 GiB106±37%
Skywork-R1V3-38BIQ3_M38.4B13.79 GiB0.00 GiB14.86 GiB0.02 GiB37±22%
gpt-oss-20b-hereticMoEQ4_K_S20.9B13.83 GiB0.03 GiB14.85 GiB0.03 GiB111±37%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEQ4_123.6B13.82 GiB0.02 GiB14.85 GiB0.03 GiB176±37%
deepseek-coder-33b-instructIQ3_S33.3B13.49 GiB0.27 GiB14.84 GiB0.04 GiB37±22%
North-Mini-Code-1.0MoEQ3_K_L30.5B13.74 GiB0.11 GiB14.83 GiB0.05 GiB154±37%
Mistral-MOE-4X7B-Dark-MultiVerse-Uncensored-Enhanced32-24BMoEQ4_K_M24.2B13.65 GiB0.14 GiB14.83 GiB0.05 GiB21±37%
QwQ-32B-Preview-abliterated-linear25I1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
openhands-lm-32b-v0.1I1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
Qwen2.5-Coder-32B-abliteratedI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
m1-32bI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
XMainframe-v2-Instruct-32bI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
Qwen2.5-Coder-32B-Python-SpecialistI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
Qwen2.5-32b-RP-InkI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
Qwen2.5-Coder-32BIQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
DeepSeek-R1-Distill-Qwen-32B-hereticI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
InnoSpark-HPC-RM-32BI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
Qwen2.5-Coder-32B-Instruct-UncensoredI1-IQ3_S32.8B13.45 GiB0.28 GiB14.83 GiB0.05 GiB37±22%
cogito-v1-preview-qwen-32BI1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
QwQ-32B-Snowdrop-v0I1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
DeepSeek-R1-Distill-Qwen-32B-UncensoredI1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
RoguePlanet-DeepSeek-R1-Qwen-32B-RPI1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
Qwen3-VL-32B-Instruct-ultra-uncensored-hereticI1-IQ3_S33.4B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
Huihui-Qwen3-VL-32B-Instruct-abliteratedI1-IQ3_S33.4B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
ColorGUI-32BI1-IQ3_S33.4B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
Qwen3-32B-UncensoredI1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
Qwen3-32B-abliteratedI1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
AReaL-boba-2-32BI1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
Assistant_Pepe_32BI1-IQ3_S32.8B13.44 GiB0.28 GiB14.82 GiB0.06 GiB37±22%
Qwen3.5-27B-Engineer-Deckard-GeminiI1-IQ4_XS27.7B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredI1-IQ4_XS27.4B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-ThinkingI1-IQ4_XS27.4B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
Huihui-Qwen3.5-27B-abliteratedI1-IQ4_XS27.8B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
Qwen3.5-27B-Unredacted-MAXI1-IQ4_XS27.4B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
Qwen3.5-27B-hereticI1-IQ4_XS27.4B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
Qwen3.5-27B-DerestrictedI1-IQ4_XS27.8B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-DistilledI1-IQ4_XS27.8B13.68 GiB0.07 GiB14.81 GiB0.07 GiB37±22%
codellama-13b-oasst-sft-v10Q8_013.0B12.88 GiB0.88 GiB14.80 GiB0.08 GiB37±22%
chronos-hermes-13b-v2Q8_013.0B12.88 GiB0.88 GiB14.80 GiB0.08 GiB37±22%
WhiteRabbitNeo-13B-v1Q8_013.0B12.88 GiB0.88 GiB14.80 GiB0.08 GiB37±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 16GB run?
1865 of 2118 indexed open-weight models fit a Tesla V100 16GB at 4,096 context with q4_0 KV cache, the largest being Darwin-35B-A3B-Opus at IQ3_XXS. That covers text, vision-language, image, video and speech models.
How much usable memory does a Tesla V100 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 Tesla V100 16GB 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.