NVIDIA · consumer

Titan V

Titan V has 32 GB of VRAM at 868 GB/s — about 29.76 GiB usable after driver and compositor overhead. 2022 of 2118 indexed models fit at 4K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
HBM2
Bandwidth
868 GB/s
4096-bit bus
Tensor FP16
119 TF
dense
TDP
250 W
$5999 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1737audio tts 21vision language 181video 16image 2embedding 26audio asr 39

What fits at 4K context

largest quantization that fits, per model · 2022 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Kimi-Dev-72BUD-IQ2_M72.7B27.56 GiB1.25 GiB29.74 GiB0.02 GiB22±12.9%
Qwen3-TTS-12Hz-0.6B-BaseF32915M28.88 GiB0.00 GiB29.72 GiB0.04 GiB22±12.9%
Rombo-LLM-V3.0-Qwen-72bI1-Q2_K_S72.7B27.54 GiB1.25 GiB29.72 GiB0.04 GiB22±12.9%
Qwen2.5-72B-Instruct-abliteratedI1-Q2_K_S72.7B27.54 GiB1.25 GiB29.72 GiB0.04 GiB22±12.9%
Qwen2.5-72B-Instruct-abliterated-v2I1-Q2_K_S72.7B27.54 GiB1.25 GiB29.72 GiB0.04 GiB22±12.9%
MiroThinker-v1.0-72BI1-Q2_K_S72.7B27.54 GiB1.25 GiB29.72 GiB0.04 GiB22±12.9%
Malaysian-Qwen2.5-72B-InstructI1-Q2_K_S72.7B27.54 GiB1.25 GiB29.72 GiB0.04 GiB22±12.9%
Qwen2.5-72BI1-Q2_K_S72.7B27.54 GiB1.25 GiB29.72 GiB0.04 GiB22±12.9%
Tower-Plus-72B-ultra-uncensored-hereticI1-Q2_K_S72.7B27.54 GiB1.25 GiB29.72 GiB0.04 GiB22±12.9%
Qwen3.5-35B-A3BMoEQ6_K36.0B28.82 GiB0.08 GiB29.70 GiB0.06 GiB124±37%
Qwen3.6-35B-A3BMoEQ6_K36.0B28.82 GiB0.08 GiB29.70 GiB0.06 GiB124±37%
deepseek-llm-67b-chatI1-Q3_K_S67.4B27.30 GiB1.48 GiB29.68 GiB0.08 GiB22±12.9%
deepseek-llm-67b-baseI1-Q3_K_S67.4B27.30 GiB1.48 GiB29.68 GiB0.08 GiB22±12.9%
openbuddy-deepseek-67b-v15.3-4kI1-Q3_K_S67.4B27.30 GiB1.48 GiB29.68 GiB0.08 GiB22±12.9%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedIQ2_XXS109B28.09 GiB0.75 GiB29.67 GiB0.09 GiB84±37%
Qwen3-Coder-Next-Opus-4.6-Reasoning-DistilledMoEIQ3_XXS—28.78 GiB0.09 GiB29.67 GiB0.09 GiB139±37%
Hypernova-60B-2605MoEI1-IQ3_S58.7B28.73 GiB0.15 GiB29.66 GiB0.10 GiB104±37%
Qwen3-53B-A3B-2507-THINKING-TOTAL-RECALL-v2-MASTER-CODERMoEI1-Q4_K_S53.0B28.13 GiB0.66 GiB29.58 GiB0.18 GiB78±37%
Huihui-Qwen3-Coder-Next-abliteratedMoEI1-IQ3_XXS79.7B28.68 GiB0.09 GiB29.57 GiB0.19 GiB139±37%
Bernini-RQ8_014.3B28.71 GiB0.00 GiB29.56 GiB0.20 GiB22±12.9%
Salience-1.5-ProMoEQ6_K_L36.0B28.66 GiB0.08 GiB29.54 GiB0.22 GiB125±37%
Qwable-v1MoEQ6_K_L36.0B28.66 GiB0.08 GiB29.54 GiB0.22 GiB125±37%
T-SearchMoEQ6_K_L36.0B28.66 GiB0.08 GiB29.54 GiB0.22 GiB125±37%
Snowpiercer-15B-v4BF1615.0B27.90 GiB0.78 GiB29.53 GiB0.23 GiB22±12.9%
Seed-OSS-36B-Instruct-biprojected-norm-preserving-abliteratedI1-Q6_K36.2B27.63 GiB1.00 GiB29.53 GiB0.23 GiB22±12.9%
Seed-OSS-36B-InstructQ6_K36.2B27.63 GiB1.00 GiB29.53 GiB0.23 GiB22±12.9%
Hermes-4.3-36B-hereticI1-Q6_K36.2B27.63 GiB1.00 GiB29.53 GiB0.23 GiB22±12.9%
Hermes-4.3-36BQ6_K36.2B27.63 GiB1.00 GiB29.53 GiB0.23 GiB22±12.9%
Seed-OSS-36B-BaseQ6_K36.2B27.63 GiB1.00 GiB29.53 GiB0.23 GiB22±12.9%
OLMo-2-1124-13B-InstructF1613.7B25.55 GiB3.13 GiB29.52 GiB0.24 GiB22±12.9%
Melody1437-27BQ3_K_M27.8B28.40 GiB0.25 GiB29.52 GiB0.24 GiB22±12.9%
HuatuoGPT-o1-72BIQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
EVA-Qwen2.5-72B-v0.2IQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
Qwen2.5-Math-72B-InstructIQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
Qwen2.5-72B-InstructIQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
magnum-v4-72bI1-IQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
KAT-Dev-72B-ExpIQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
Homer-v1.0-Qwen2.5-72BIQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
Qwen2.5-VL-72B-InstructIQ2_M73.4B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
Chronos-Platinum-72BIQ2_M72.7B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
UI-TARS-72B-DPOIQ2_M73.4B27.32 GiB1.25 GiB29.50 GiB0.26 GiB22±12.9%
Maenad-70BI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-ReasonerI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Rombos-LLM-70b-Llama-3.3I1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
L3.3-Electra-R1-70bI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Latxa-Llama-3.1-70B-Instruct-v2I1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Llama-3.3_70_b_uncensored_continuedI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Llama-3.3-70B-Instruct-abliteratedI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
grok-oss-Revenant-70BI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Llama-3.1-Nemotron-70B-Instruct-HFI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
L3.3-70B-Euryale-v2.3I1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Hermes-3-Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Hermes-4-70B-hereticI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Llama-3.3-70B-InstructIQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Llama-3.1-70BIQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Anubis-70B-v1.2IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Hermes-4-70BIQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
Golem-70B-v1bI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
DeepSeek-R1-Distill-Llama-70B-abliteratedI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
DeepSeek-R1-Distill-Llama-70B-hereticI1-IQ3_XS70.6B27.29 GiB1.25 GiB29.47 GiB0.29 GiB22±12.9%
●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 Titan V run?
2022 of 2118 indexed open-weight models fit a Titan V at 4,096 context with f16 KV cache, the largest being Kimi-Dev-72B at UD-IQ2_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Titan V 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 Titan V fast for local AI?
Its memory bandwidth is 868 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.