RunsOnMy ModelsGPUs & MacsBy VRAMBest GPUCan I run…?OllamaStart hereAPI pricesVRAM calculatorNew

What AI models run on the RTX 3090?

The RTX 3090 has 24 GB of VRAM (936 GB/s). About 22.8 GB is usable for a model. 209 of 298 open models we track fit at 8K context.

209models fit (8K ctx)
209still fit at 32K ctx
49Blargest that fits
936 GB/smemory bandwidth

This card has 24 GB of VRAM. See every model that fits in 24 GB VRAM. Other 24 GB hardware: RTX 3090 Ti, RTX 4090, RX 7900 XTX, Mac M2 24GB, Mac M3 24GB, Mac M4 Pro 24GB.

Compare for local LLMs: RTX 3090 vs RTX 3090 TiRTX 3090 vs RTX 4070 Ti SuperRTX 3090 vs RTX 5080RTX 3090 vs RTX 4090RTX 3090 vs RTX 5090RTX 3090 vs RX 7900 XTXRTX 3090 vs Mac M4 Max 64GB all comparisons

Using a specific tool? Ollama models that run on the RTX 3090 · LM Studio · llama.cpp offload

Models that fit, largest first

ModelParamsBest quant (8K)Needs32K contextMax tok/s (est.)Details
Kimi Linear 48B A3B Instruct49B (3.0B act.)Q2_K17.3 GBQ2_K874details
Nous Hermes 2 Mixtral 8x7B DPO47B (13B act.)Q2_K17.6 GBQ2_K196details
Falcon 40B42BQ3_K_M20.2 GBQ2_K47details
Flint40B (1.3B act.)Q3_K_M19.1 GBQ3_K_M1525details
Maxwell V1 Flash37B (2.3B act.)Q3_K_M17.9 GBQ3_K_M843details
Hermes 4.3 36B36BQ3_K_M19.1 GBQ2_K53details
Seed OSS 36B Instruct36BQ3_K_M19.1 GBQ2_K53details
Qwen3.6 35B A3B36B (3.0B act.)Q3_K_M17.0 GBQ3_K_M655details
Ornith 1.5 35B A3B36B (3.0B act.)Q3_K_M16.6 GBQ3_K_M672details
Qwen3.5 35B A3B36B (3.0B act.)Q3_K_M17.0 GBQ3_K_M655details
Sage A1.536B (4.7B act.)Q3_K_M17.0 GBQ3_K_M413details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q3_K_M16.6 GBQ3_K_M648details
Yi 34B 200K34BQ3_K_M17.9 GBQ2_K56details
Yi 34B34BQ3_K_M17.9 GBQ2_K56details
Yi 34B Chat34BQ3_K_M17.9 GBQ2_K56details
Yi 1.5 34B Chat34BQ3_K_M18.1 GBQ2_K56details
Nous Hermes 2 Yi 34B34BQ3_K_M18.4 GBQ2_K54details
DeepSeek R1 Distill Qwen 32B33BQ3_K_M17.5 GBQ2_K59details
Qwen3 32B33BQ3_K_M17.6 GBQ2_K58details
GLM Z1 32B 041433BQ4_K_M19.5 GBQ3_K_M48details
GLM 4 32B 041433BQ4_K_M19.5 GBQ3_K_M48details
Olmo 3 32B Think32BQ4_K_M20.0 GBQ3_K_M48details
Granite 4.0 H Small32B (8.8B act.)Q4_K_M18.9 GBQ4_K_M176details
Gemma 4 31B IT31BQ3_K_M17.8 GBQ2_K59details
Gemma 4 31B31BQ3_K_M17.2 GBQ2_K61details
GLM 4.7 Flash31B (4.6B act.)Q4_K_M18.3 GBQ4_K_M343details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)Q4_K_M18.8 GBQ3_K_M511details
North Mini Code 1.030B (3.3B act.)Q4_K_M18.8 GBQ4_K_M461details
Hy Memory 30B A3B30B (3.0B act.)Q4_K_M18.5 GBQ3_K_M510details
Muse Glimmer 30B30BQ4_K_M17.9 GBQ4_K_M51details
Granite 4.2 30B29BQ4_K_M19.1 GBQ2_K53details
Raon 1 Mini Preview29B (4.3B act.)Q4_K_M19.4 GBQ4_K_M317details
Qwen3.8 27B28BQ4_K_M17.6 GBQ4_K_M54details
Qwen3.6 27B28BQ4_K_M18.1 GBQ4_K_M52details
Qwen3.5 27B28BQ4_K_M18.1 GBQ4_K_M52details
Gemma 3 27B IT27BQ5_K_M19.9 GBQ4_K_M49details
Arete 27B27BQ4_K_M17.4 GBQ4_K_M54details
MiniMax H3 Image PE27BQ5_K_M19.3 GBQ4_K_M49details
Trinity Mini26B (2.0B act.)Q5_K_M18.4 GBQ5_K_M668details
GLiDE 27B26BQ5_K_M18.7 GBQ5_K_M50details
DiffusionGemma 26B A4B IT26B (4.0B act.)Q5_K_M19.2 GBQ5_K_M316details
Gemma 4 26B A4B IT26B (4.0B act.)Q5_K_M19.4 GBQ5_K_M313details
LFM2.5 24B A2B24B (2.0B act.)Q6_K19.1 GBQ6_K570details
ERNIE 4.5 21B A3B PT22BQ6_K17.9 GBQ6_K52details
ERNIE 4.5 21B A3B Thinking22BQ6_K17.9 GBQ6_K52details
gpt-oss Safeguard 20B22B (4.8B act.)F1613.8 GBF16305details
gpt-oss 20B21B (4.2B act.)F1613.8 GBF16338details
Kimi VL A3B Thinking 250616B (3.0B act.)Q8_016.8 GBQ8_0302details
Kimi VL A3B Thinking16B (3.0B act.)Q8_017.2 GBQ8_0294details
Kimi VL A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_0302details
DeepSeek MoE 16B Base16B (2.8B act.)Q8_018.6 GBQ5_K_M311details
LLaDA2.1 Mini16B (1.4B act.)Q8_017.1 GBQ8_0615details
Moonlight 16B A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_0293details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q8_016.4 GBQ8_0331details
DeepSeek V2 Lite16B (2.7B act.)Q8_016.4 GBQ8_0331details
DeepSeek V2 Lite Chat16B (2.7B act.)Q8_016.4 GBQ8_0331details
Qwen2.5 Coder 14B Instruct15BQ8_016.8 GBQ6_K60details
DeepSeek R1 Distill Qwen 14B15BQ8_016.8 GBQ6_K60details
Qwen3 14B15BQ8_016.6 GBQ8_060details
Phi 415BQ8_016.7 GBQ6_K60details
Gemma 3 12B IT12BQ8_013.4 GBQ8_075details
Llama Guard 4 12B12BQ8_014.2 GBQ8_073details
KORMo Web 11B11BQ8_013.0 GBQ8_079details
Llama 3.2 11B Vision Instruct11BF1620.1 GBQ8_048details
Llama 3.2 11B Vision11BQ8_012.4 GBQ8_083details
GLM 4.6V Flash10BF1618.5 GBF1650details
GLM 4.1V 9B Thinking10BF1618.5 GBF1650details
Step3 VL 10B10BF1617.1 GBF1657details
SenseNova Loop 9B Preview9.7BF1617.8 GBF1652details
Qwen3.5 9B9.7BF1618.3 GBF1651details
Ornith 1.5 9B9.7BF1617.8 GBF1652details
SenseNova Loop 9B9.4BF1618.7 GBF1650details
MiMo V2.6 Distill Qwen 9B9.4BF1617.8 GBF1652details
GLM 4 9B9.4BF1618.5 GBF1650details
Apertus v1.5 8B8.9BF1618.2 GBQ8_053details
InternLM3 8B Instruct8.8BF1617.4 GBF1653details
Granite 4.2 8B8.8BF1618.2 GBQ8_053details
Granite 4.1 8B8.8BF1618.2 GBQ8_053details
Granite 4.1 8B Llmon Preview8.8BF1618.2 GBQ8_053details
Qwen3 VL 8B Instruct8.8BF1617.1 GBF1657details
Molmo2 8B8.7BF1617.1 GBF1657details
LFM2.5 8B A1B8.5B (1.0B act.)F1616.5 GBF16468details
Qwen2.5 VL 7B Instruct8.3BF1615.3 GBF1661details
UI TARS 1.5 7B8.3BF1615.3 GBF1661details
olmOCR 2 7B 10258.3BF1615.3 GBF1661details
olmOCR 7B 0225 Preview8.3BF1616.6 GBF1656details
UI TARS 7B DPO8.3BF1615.3 GBF1661details
Qwen3 8B8.2BF1617.1 GBF1657details
DeepSeek R1 0528 Qwen3 8B8.2BF1617.1 GBF1657details
Granite 3.3 8B Instruct8.2BF1617.0 GBQ8_057details
Granite 3.1 8B Instruct8.2BF1617.0 GBQ8_057details
Granite 3.2 8B Instruct8.2BF1617.0 GBQ8_057details
Apertus 8B Instruct 25098.1BF1616.7 GBF1658details
Llama 3.1 8B Instruct8.0BF1616.6 GBF1658details
Llama 3.1 8B8.0BF1616.6 GBF1658details
Hermes 3 Llama 3.1 8B8.0BF1616.6 GBF1658details
Llama Guard 3 8B8.0BF1616.6 GBF1658details
DeepSeek R1 Distill Llama 8B8.0BF1616.6 GBF1658details
Meta Llama 3 8B Instruct8.0BF1616.6 GBF1658details
Meta Llama Guard 2 8B8.0BF1616.6 GBF1658details
Meta Llama 3.1 8B Instruct8.0BF1616.6 GBF1658details
Aya Expanse 8B8.0BF1616.9 GBF1658details
Ling 3.0 Tiny7.9B (1.4B act.)F1615.6 GBF16338details
MiMo 7B RL7.8BF1616.4 GBF1660details
InternLM2.5 7B Chat7.7BF1616.0 GBF1660details
InternLM2 Chat 7B7.7BF1616.0 GBF1660details
Qwen2.5 7B Instruct7.6BF1615.3 GBF1661details
DeepSeek R1 Distill Qwen 7B7.6BF1615.3 GBF1661details
Qwen2.5 Coder 7B Instruct7.6BF1615.3 GBF1661details
Olmo 3 7B Instruct7.3BF1616.7 GBF1664details
Olmo 3 1025 7B7.3BF1616.7 GBF1664details
Olmo 3 7B Think7.3BF1616.7 GBF1664details
Mistral 7B v0.17.2BF1614.9 GBF1665details
Mistral 7B Instruct v0.27.2BF1614.9 GBF1665details
Mistral 7B Instruct v0.17.2BF1614.9 GBF1665details
Falcon 7B7.2BF1614.0 GBF1665details
Falcon 7B Instruct7.2BF1614.0 GBF1665details
Granite 4.0 H Tiny6.9B (1.5B act.)F1613.6 GBF16319details
OLMoE 1B 7B 09246.9B (1.3B act.)F1614.4 GBF16365details
DeepSeek Coder 7B Instruct v1.56.9BF1617.2 GBQ5_K_M68details
DeepSeek Coder 6.7B Instruct6.7BF1617.0 GBQ4_K_M69details
DeepSeek Coder 6.7B Base6.7BF1617.0 GBQ4_K_M69details
Llama 2 7B HF6.7BF1617.0 GBQ4_K_M69details
Yi 6B 200K6.1BF1612.3 GBF1677details
Yi 6B6.1BF1612.3 GBF1677details
Molmo2 4B4.9BF1610.0 GBF16106details
WeVisOne VL 4B4.8BF1610.0 GBF16106details
Mage VL Cowork4.7BF1610.0 GBF16106details
Qianfan OCR4.7BF1610.1 GBF16106details
Qwen3.5 4B4.7BF169.2 GBF16108details
SingVL S1 4B4.5BF169.0 GBF16111details
LightOnOCR 3 4B4.5BF169.0 GBF16111details
Intern Decision 4B4.5BF169.0 GBF16111details
Qwen3 VL 4B Instruct4.4BF169.3 GBF16116details
Gemma 3 4B IT4.3BF168.4 GBF16121details
MedGemma 1.5 4B IT4.3BF168.4 GBF16120details
MedGemma 4B IT4.3BF168.4 GBF16120details
TranslateGemma 4B IT4.3BF169.2 GBF16109details
Gemma 3 4B PT4.3BF168.4 GBF16121details
Phi 3.5 Vision Instruct4.1BF1611.2 GBF16113details
Qwen3 4B4.0BF169.3 GBF16116details
Qwen3 4B Instruct 25074.0BF169.3 GBF16116details
Granite Vision 4.1 4B4.0BF167.6 GBF16137details
Phi 4 Mini Instruct3.8BF168.9 GBF16122details
Phi 3 Mini 4K Instruct3.8BF1610.6 GBF16122details
Phi 3.5 Mini Instruct3.8BF1610.6 GBF16122details
Phi 3 Mini 128K Instruct3.8BF1610.6 GBF16122details
Qwen2.5 VL 3B Instruct3.8BF168.0 GBF16125details
Granite 4.2 3B3.7BF168.0 GBF16128details
Granite 4.1 3B Llmon Preview3.7BF168.0 GBF16128details
Raon 1 Nano3.5BF167.0 GBF16149details
Granite 4.1 3B3.4BF167.6 GBF16137details
Granite 4.0 Micro3.4BF167.6 GBF16137details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16452details
DeepSeek VL2 Tiny3.4B (1.2B act.)F167.4 GBF16398details
Tiny Aya Global3.3BF167.5 GBF16139details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16463details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16466details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16466details
Granite 3.1 3B A800M Instruct3.3B (883M act.)F167.1 GBF16530details
Llama 3.2 3B Instruct3.2BF167.5 GBF16145details
Llama 3.2 3B3.2BF167.5 GBF16146details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF16145details
d1 3B3.1BF165.8 GBF16173details
LFM2.5 VL 3B3.1BF165.8 GBF16173details
Qwen2.5 3B Instruct3.1BF166.7 GBF16152details
SmolLM3 3B3.1BF166.9 GBF16152details
SmolLM3 3B Base3.1BF166.9 GBF16152details
Phi 22.8BF168.2 GBF16168details
LFM2.5 2.6B2.7BF1610.8 GBF1687details
Granite 3.3 2B Instruct2.5BF165.8 GBF16185details
MiniCPM5 2B2.5BF165.7 GBF16186details
WeVisOne VL 2B2.4BF165.4 GBF16230details
Cosmos Reason2 2B2.4BF165.4 GBF16230details
Qwen3.5 2B2.3BF164.6 GBF16240details
SmolVLM Instruct2.2BF165.4 GBF16258details
Qwen3 1.7B2.0BF165.4 GBF16230details
d1 2B1.9BF164.0 GBF16270details
LFM3 VL 2B1.9BF164.0 GBF16270details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16263details
LFM3 1.7B1.7BF167.2 GBF16135details
LFM3 1.7B Base1.7BF164.0 GBF16270details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16273details
SmolLM2 1.7B1.7BF165.2 GBF16273details
SmolLM 1.7B1.7BF165.2 GBF16273details
Granite 4.0 1B1.6BF164.3 GBF16286details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16303details
OLMo 2 0425 1B1.5BF164.4 GBF16315details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBF16348details
Llama 3.2 1B1.2BF163.2 GBF16377details
Llama 3.2 1B Instruct1.2BF163.2 GBF16377details
LFM2.5 1.2B Instruct1.2BF165.0 GBF16200details
LFM2 1.2B1.2BF163.0 GBF16399details
HunyuanOCR1.1BF163.5 GBF16418details
PaddleOCR VL 1.6959MF162.5 GBF16488details
Qwen3.5 0.8B873MF162.4 GBF16601details
LightOnOCR 3 0.8B853MF162.4 GBF16617details
FastVLM 0.5B759MF162.2 GBF16617details
Qwen3 0.6B752MF163.0 GBF16620details
GOT OCR2.0716MF162.8 GBF16654details
Nemotron 3.5 Parse631MF161.9 GBF16716details
GOT OCR 2.0 HF561MF162.5 GBF16835details
SmolVLM2 500M Video Instruct507MF161.6 GBF161141details
SmolVLM 500M Instruct507MF161.6 GBF161141details
Qwen2.5 0.5B Instruct494MF161.7 GBF16947details
LFM2.5 VL 450M449MF161.3 GBF161316details
SmolLM2 360M Instruct362MF161.5 GBF161290details
SmolLM2 360M362MF161.5 GBF161290details
LFM2.5 350M354MF161.9 GBF16658details

Bigger models with CPU offload

These popular models don't fit in the RTX 3090's VRAM at 4-bit, but run with llama.cpp offload, keeping part of the model in system RAM. Speed range is DDR4-3200 to DDR5-6000 (estimate). Method.

ModelParamsOffload plan
GLM 5.3 Flash321B MoEOffload Q2_K, 128 GB RAM · ~8.0–16 tok/s
gpt-oss 120B117B MoEOffload Q4_K_M, 64 GB RAM · ~17–34 tok/s
DeepSeek V4 Flash 0731304B MoEOffload Q3_K_M, 128 GB RAM · ~6.8–13 tok/s
DeepSeek V4 Flash291B MoEOffload Q3_K_M, 128 GB RAM · ~7.6–15 tok/s
GLM 4.5 Air110B MoEOffload Q4_K_M, 64 GB RAM · ~6.5–13 tok/s
MiniMax M2.5229B MoEOffload Q4_K_M, 128 GB RAM · ~6.1–12 tok/s
Hy3299B MoEOffload Q3_K_M, 128 GB RAM · ~4.5–9.1 tok/s
Nex N2.6321B MoEOffload Q2_K, 128 GB RAM · ~7.5–15 tok/s

Comparing cards? The RTX 3090 (launch MSRP $1499) is in best GPUs for local LLMs under $1,500. See also Mac vs PC.

Just too big at 4-bit

These need up to 35% more memory than the RTX 3090 has at Q4_K_M and 8K context. A smaller quant or shorter context may work.

ModelQ4_K_M needsShort by
Hermes 4.3 36B23.0 GB0.2 GB
Seed OSS 36B Instruct23.0 GB0.2 GB
Flint24.5 GB1.7 GB
Falcon 40B25.2 GB2.4 GB
Nous Hermes 2 Mixtral 8x7B DPO28.0 GB5.2 GB
Kimi Linear 48B A3B Instruct28.9 GB6.1 GB

Questions

What is the biggest model that runs on the RTX 3090?

Of the models we track, the largest that fits at 8K context is Kimi Linear 48B A3B Instruct (49B) at Q2_K, needing about 17.3 GB.

How much of the RTX 3090's memory can a model use?

About 22.8 GB: we assume 95% of the 24 GB is usable after driver and display overhead.

Spec source: NVIDIA specifications. Model data: Hugging Face Hub API. Speed is a bandwidth ceiling (bandwidth ÷ bytes read per token), not a benchmark. Methodology · Report an error.