NVIDIA · datacenter

Tesla P100 16GB

Tesla P100 16GB has 16 GB of VRAM at 732 GB/s — about 14.88 GiB usable after driver and compositor overhead. 1855 of 2118 indexed models fit at 16K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
16 GB
HBM2
Bandwidth
732 GB/s
4096-bit bus
Tensor FP16
—
dense
TDP
250 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1590vision language 162video 15embedding 26audio asr 39audio tts 21image 2

What fits at 16K context

largest quantization that fits, per model · 1855 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Hy-MT2-30B-A3BMoEQ3_K_M30.1B13.45 GiB0.42 GiB14.87 GiB0.01 GiB111±37%
granite-20b-code-instruct-8kQ5_K_M20.1B13.79 GiB0.00 GiB14.87 GiB0.01 GiB30±22%
granite-20b-code-base-8kI1-Q5_K_M20.1B13.79 GiB0.00 GiB14.87 GiB0.01 GiB30±22%
WizardCoder-Python-34B-V1.0I1-IQ3_XS33.7B12.93 GiB0.84 GiB14.87 GiB0.01 GiB30±22%
Phind-CodeLlama-34B-Python-v1I1-IQ3_XS33.7B12.93 GiB0.84 GiB14.87 GiB0.01 GiB30±22%
Phind-CodeLlama-34B-v2I1-IQ3_XS33.7B12.93 GiB0.84 GiB14.87 GiB0.01 GiB30±22%
granite-34b-code-base-8kI1-IQ3_S33.7B13.79 GiB0.00 GiB14.87 GiB0.01 GiB30±22%
Skyfall-31B-v4.2-hereticI1-Q3_K_S31.4B12.80 GiB0.95 GiB14.87 GiB0.01 GiB30±22%
Skyfall-31B-v4.2I1-Q3_K_S31.4B12.80 GiB0.95 GiB14.87 GiB0.01 GiB30±22%
Qwen3.6-28BMoEI1-Q3_K_L28.2B13.77 GiB0.09 GiB14.87 GiB0.01 GiB154±37%
Qwen3.5-28BMoEI1-Q3_K_L28.7B13.77 GiB0.09 GiB14.87 GiB0.01 GiB154±37%
GLM-4.7-Flash-REAP-23B-A3BMoEQ4_123.0B13.62 GiB0.23 GiB14.86 GiB0.02 GiB107±37%
solar-pro-preview-instructKV unresolvedQ4_K_M22.1B12.40 GiB1.41 GiB14.86 GiB0.02 GiB30±22%
Skywork-R1V3-38BIQ3_M38.4B13.79 GiB0.00 GiB14.86 GiB0.02 GiB30±22%
GLM-4.7-FlashMoEQ3_K_M31.2B13.61 GiB0.23 GiB14.85 GiB0.03 GiB121±37%
Gemma-4-31B-Isometry-RPI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Gemma-4-Dark-Gemistry-31BI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Prosopon-31BI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Gemma-4-Novelist-Eclipse-31BI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Giftige-Blume-31B-v1-StyleSwapI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
G4-MeroMero-31B-StyleSwapI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Gemma-4-31B-StyleTune-heretic-araI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Pantheon-Reasoning-31B-1.1I1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Gemma-4-31B-StyleTuneI1-IQ3_XS32.7B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
Barcenas-StyleTune-31B-FableI1-IQ3_XS32.1B12.74 GiB1.03 GiB14.85 GiB0.03 GiB30±22%
GLM-4.7-Flash-hereticMoEIQ3_M29.9B13.60 GiB0.23 GiB14.85 GiB0.03 GiB121±37%
Pantheon-Reasoning-26B-A4B-1.1MoEIQ4_XS26.5B13.59 GiB0.26 GiB14.84 GiB0.04 GiB30±22%
diffusiongemma-26B-A4B-it-HERETIC-UncensoredMoEIQ4_NL25.8B13.59 GiB0.26 GiB14.84 GiB0.04 GiB30±22%
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoEI1-IQ3_M30.0B13.00 GiB0.83 GiB14.84 GiB0.04 GiB73±37%
Qwen3-53B-A3B-2507-THINKING-TOTAL-RECALL-v2-MASTER-CODERMoEI1-IQ2_XXS53.0B13.10 GiB0.74 GiB14.83 GiB0.05 GiB92±37%
gemma-4-26B-A4B-it-Claude-Opus-DistillMoEIQ4_NL26.5B13.58 GiB0.26 GiB14.83 GiB0.05 GiB30±22%
gemma-4-26B-A4B-it-Claude-Opus-Distill-v2MoEIQ4_NL26.5B13.58 GiB0.26 GiB14.83 GiB0.05 GiB30±22%
gemma-4-26B-A4BMoEIQ4_NL26.5B13.58 GiB0.26 GiB14.83 GiB0.05 GiB30±22%
Delphi-25B-SimpleRL-MathI1-IQ3_XXS25.0B9.03 GiB4.71 GiB14.81 GiB0.07 GiB30±22%
Qwen3-Coder-REAP-25B-A3BMoEQ4_024.9B13.39 GiB0.42 GiB14.80 GiB0.08 GiB103±37%
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-IQ2_M39.5B13.31 GiB0.42 GiB14.79 GiB0.09 GiB30±22%
Aurora-Code-1MoEI1-Q3_K_M34.7B13.70 GiB0.09 GiB14.79 GiB0.09 GiB165±37%
Qwen3.5-21B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-Q5_K_S21.3B13.50 GiB0.21 GiB14.77 GiB0.11 GiB30±22%
Qwen3.6-21B-IQ-Ultra-Heretic-Uncensored-ThinkingI1-Q5_K_S21.3B13.50 GiB0.21 GiB14.77 GiB0.11 GiB30±22%
TildeOpen-30B-Instruct-LVI1-IQ3_S30.7B12.62 GiB1.05 GiB14.76 GiB0.12 GiB30±22%
dolphin-2.6-mixtral-8x7bMoEI1-IQ2_S46.7B13.16 GiB0.56 GiB14.76 GiB0.12 GiB53±37%
xLAM-8x7b-rMoEIQ2_S46.7B13.16 GiB0.56 GiB14.76 GiB0.12 GiB53±37%
reka-flash-3.1Q4_K_L20.9B13.10 GiB0.58 GiB14.76 GiB0.12 GiB30±22%
reka-flash-3Q4_K_L20.9B13.10 GiB0.58 GiB14.76 GiB0.12 GiB30±22%
GLM-4-32B-0414-Korean-CultureI1-IQ3_S32.6B13.40 GiB0.27 GiB14.75 GiB0.13 GiB30±22%
Rocinante-XL-16B-v1Q6_K16.1B12.76 GiB0.95 GiB14.75 GiB0.13 GiB30±22%
OpenAI-gpt-oss-20B-Claude-4.5-Opus-Heretic-UncensoredMoEI1-Q4_K_S20.9B13.65 GiB0.11 GiB14.75 GiB0.13 GiB90±37%
gpt-oss-20b-uncensoredMoEI1-Q4_K_S20.9B13.65 GiB0.11 GiB14.75 GiB0.13 GiB90±37%
gpt-oss-safeguard-20bMoEI1-Q4_K_S21.5B13.65 GiB0.11 GiB14.75 GiB0.13 GiB90±37%
Huihui-gpt-oss-20b-BF16-abliterated-v2MoEI1-Q4_K_S20.9B13.65 GiB0.11 GiB14.75 GiB0.13 GiB90±37%
metatune-gpt20b-R1.09MoEI1-Q4_K_S21.5B13.65 GiB0.11 GiB14.75 GiB0.13 GiB90±37%
gpt-oss-20b-DerestrictedMoEQ4_K_S20.9B13.65 GiB0.11 GiB14.75 GiB0.13 GiB90±37%
OmniAtlas-Qwen3-30B-A3BI1-Q3_K_M31.7B13.70 GiB0.00 GiB14.75 GiB0.13 GiB30±22%
Qwen3-Omni-30B-A3B-CaptionerI1-Q3_K_M31.7B13.70 GiB0.00 GiB14.75 GiB0.13 GiB30±22%
GLM-Z1-32B-0414Q3_K_S32.6B13.38 GiB0.27 GiB14.74 GiB0.14 GiB30±22%
GLM-4-32B-0414Q3_K_S32.6B13.38 GiB0.27 GiB14.74 GiB0.14 GiB30±22%
Qwen3.6-14B-A3B-FableVibesMoEQ8_013.8B13.65 GiB0.09 GiB14.74 GiB0.14 GiB110±37%
Qwen3.6-14B-A3B-VibeForged-v2MoEQ8_013.8B13.65 GiB0.09 GiB14.74 GiB0.14 GiB110±37%
Goetia-26B-A4B-v1.3-Absolute-Heretic-ARAMoEI1-Q4_025.8B13.49 GiB0.26 GiB14.74 GiB0.14 GiB30±22%
Frank-26B-A4BMoEI1-Q4_026.5B13.49 GiB0.26 GiB14.74 GiB0.14 GiB30±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 P100 16GB run?
1855 of 2118 indexed open-weight models fit a Tesla P100 16GB at 16,384 context with q4_0 KV cache, the largest being Hy-MT2-30B-A3B at Q3_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Tesla P100 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 P100 16GB fast for local AI?
Its memory bandwidth is 732 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.