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

What AI models run on the RX 7900 XTX?

The RX 7900 XTX has 24 GB of VRAM (960 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
960 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, RTX 3090 Ti, RTX 4090, Mac M2 24GB, Mac M3 24GB, Mac M4 Pro 24GB.

Compare for local LLMs: RX 7900 XTX vs RTX 4090RX 7900 XTX vs RTX 3090RX 7900 XTX vs RTX 5080 all comparisons

Using a specific tool? Ollama models that run on the RX 7900 XTX · 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_K897details
Nous Hermes 2 Mixtral 8x7B DPO47B (13B act.)Q2_K17.6 GBQ2_K201details
Falcon 40B42BQ3_K_M20.2 GBQ2_K48details
Flint40B (1.3B act.)Q3_K_M19.1 GBQ3_K_M1564details
Maxwell V1 Flash37B (2.3B act.)Q3_K_M17.9 GBQ3_K_M865details
Hermes 4.3 36B36BQ3_K_M19.1 GBQ2_K54details
Seed OSS 36B Instruct36BQ3_K_M19.1 GBQ2_K54details
Qwen3.6 35B A3B36B (3.0B act.)Q3_K_M17.0 GBQ3_K_M672details
Ornith 1.5 35B A3B36B (3.0B act.)Q3_K_M16.6 GBQ3_K_M689details
Qwen3.5 35B A3B36B (3.0B act.)Q3_K_M17.0 GBQ3_K_M672details
Sage A1.536B (4.7B act.)Q3_K_M17.0 GBQ3_K_M424details
Qwen AgentWorld 35B A3B35B (3.0B act.)Q3_K_M16.6 GBQ3_K_M665details
Yi 34B 200K34BQ3_K_M17.9 GBQ2_K58details
Yi 34B34BQ3_K_M17.9 GBQ2_K58details
Yi 34B Chat34BQ3_K_M17.9 GBQ2_K58details
Yi 1.5 34B Chat34BQ3_K_M18.1 GBQ2_K57details
Nous Hermes 2 Yi 34B34BQ3_K_M18.4 GBQ2_K56details
DeepSeek R1 Distill Qwen 32B33BQ3_K_M17.5 GBQ2_K60details
Qwen3 32B33BQ3_K_M17.6 GBQ2_K60details
GLM Z1 32B 041433BQ4_K_M19.5 GBQ3_K_M49details
GLM 4 32B 041433BQ4_K_M19.5 GBQ3_K_M49details
Olmo 3 32B Think32BQ4_K_M20.0 GBQ3_K_M49details
Granite 4.0 H Small32B (8.8B act.)Q4_K_M18.9 GBQ4_K_M180details
Gemma 4 31B IT31BQ3_K_M17.8 GBQ2_K60details
Gemma 4 31B31BQ3_K_M17.2 GBQ2_K63details
GLM 4.7 Flash31B (4.6B act.)Q4_K_M18.3 GBQ4_K_M352details
Qwen3 Coder 30B A3B Instruct31B (3.0B act.)Q4_K_M18.8 GBQ3_K_M524details
North Mini Code 1.030B (3.3B act.)Q4_K_M18.8 GBQ4_K_M472details
Hy Memory 30B A3B30B (3.0B act.)Q4_K_M18.5 GBQ3_K_M524details
Muse Glimmer 30B30BQ4_K_M17.9 GBQ4_K_M53details
Granite 4.2 30B29BQ4_K_M19.1 GBQ2_K54details
Raon 1 Mini Preview29B (4.3B act.)Q4_K_M19.4 GBQ4_K_M325details
Qwen3.8 27B28BQ4_K_M17.6 GBQ4_K_M55details
Qwen3.6 27B28BQ4_K_M18.1 GBQ4_K_M53details
Qwen3.5 27B28BQ4_K_M18.1 GBQ4_K_M53details
Gemma 3 27B IT27BQ5_K_M19.9 GBQ4_K_M50details
Arete 27B27BQ4_K_M17.4 GBQ4_K_M56details
MiniMax H3 Image PE27BQ5_K_M19.3 GBQ4_K_M50details
Trinity Mini26B (2.0B act.)Q5_K_M18.4 GBQ5_K_M685details
GLiDE 27B26BQ5_K_M18.7 GBQ5_K_M52details
DiffusionGemma 26B A4B IT26B (4.0B act.)Q5_K_M19.2 GBQ5_K_M324details
Gemma 4 26B A4B IT26B (4.0B act.)Q5_K_M19.4 GBQ5_K_M321details
LFM2.5 24B A2B24B (2.0B act.)Q6_K19.1 GBQ6_K584details
ERNIE 4.5 21B A3B PT22BQ6_K17.9 GBQ6_K53details
ERNIE 4.5 21B A3B Thinking22BQ6_K17.9 GBQ6_K53details
gpt-oss Safeguard 20B22B (4.8B act.)F1613.8 GBF16313details
gpt-oss 20B21B (4.2B act.)F1613.8 GBF16347details
Kimi VL A3B Thinking 250616B (3.0B act.)Q8_016.8 GBQ8_0309details
Kimi VL A3B Thinking16B (3.0B act.)Q8_017.2 GBQ8_0301details
Kimi VL A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_0309details
DeepSeek MoE 16B Base16B (2.8B act.)Q8_018.6 GBQ5_K_M319details
LLaDA2.1 Mini16B (1.4B act.)Q8_017.1 GBQ8_0631details
Moonlight 16B A3B Instruct16B (3.0B act.)Q8_016.8 GBQ8_0301details
DeepSeek Coder V2 Lite Instruct16B (2.7B act.)Q8_016.4 GBQ8_0339details
DeepSeek V2 Lite16B (2.7B act.)Q8_016.4 GBQ8_0340details
DeepSeek V2 Lite Chat16B (2.7B act.)Q8_016.4 GBQ8_0339details
Qwen2.5 Coder 14B Instruct15BQ8_016.8 GBQ6_K61details
DeepSeek R1 Distill Qwen 14B15BQ8_016.8 GBQ6_K61details
Qwen3 14B15BQ8_016.6 GBQ8_061details
Phi 415BQ8_016.7 GBQ6_K62details
Gemma 3 12B IT12BQ8_013.4 GBQ8_077details
Llama Guard 4 12B12BQ8_014.2 GBQ8_075details
KORMo Web 11B11BQ8_013.0 GBQ8_081details
Llama 3.2 11B Vision Instruct11BF1620.1 GBQ8_049details
Llama 3.2 11B Vision11BQ8_012.4 GBQ8_085details
GLM 4.6V Flash10BF1618.5 GBF1651details
GLM 4.1V 9B Thinking10BF1618.5 GBF1651details
Step3 VL 10B10BF1617.1 GBF1659details
SenseNova Loop 9B Preview9.7BF1617.8 GBF1654details
Qwen3.5 9B9.7BF1618.3 GBF1652details
Ornith 1.5 9B9.7BF1617.8 GBF1654details
SenseNova Loop 9B9.4BF1618.7 GBF1651details
MiMo V2.6 Distill Qwen 9B9.4BF1617.8 GBF1654details
GLM 4 9B9.4BF1618.5 GBF1651details
Apertus v1.5 8B8.9BF1618.2 GBQ8_054details
InternLM3 8B Instruct8.8BF1617.4 GBF1655details
Granite 4.2 8B8.8BF1618.2 GBQ8_055details
Granite 4.1 8B8.8BF1618.2 GBQ8_055details
Granite 4.1 8B Llmon Preview8.8BF1618.2 GBQ8_055details
Qwen3 VL 8B Instruct8.8BF1617.1 GBF1659details
Molmo2 8B8.7BF1617.1 GBF1659details
LFM2.5 8B A1B8.5B (1.0B act.)F1616.5 GBF16480details
Qwen2.5 VL 7B Instruct8.3BF1615.3 GBF1663details
UI TARS 1.5 7B8.3BF1615.3 GBF1663details
olmOCR 2 7B 10258.3BF1615.3 GBF1663details
olmOCR 7B 0225 Preview8.3BF1616.6 GBF1658details
UI TARS 7B DPO8.3BF1615.3 GBF1663details
Qwen3 8B8.2BF1617.1 GBF1659details
DeepSeek R1 0528 Qwen3 8B8.2BF1617.1 GBF1659details
Granite 3.3 8B Instruct8.2BF1617.0 GBQ8_059details
Granite 3.1 8B Instruct8.2BF1617.0 GBQ8_059details
Granite 3.2 8B Instruct8.2BF1617.0 GBQ8_059details
Apertus 8B Instruct 25098.1BF1616.7 GBF1660details
Llama 3.1 8B Instruct8.0BF1616.6 GBF1660details
Llama 3.1 8B8.0BF1616.6 GBF1660details
Hermes 3 Llama 3.1 8B8.0BF1616.6 GBF1660details
Llama Guard 3 8B8.0BF1616.6 GBF1660details
DeepSeek R1 Distill Llama 8B8.0BF1616.6 GBF1660details
Meta Llama 3 8B Instruct8.0BF1616.6 GBF1660details
Meta Llama Guard 2 8B8.0BF1616.6 GBF1660details
Meta Llama 3.1 8B Instruct8.0BF1616.6 GBF1660details
Aya Expanse 8B8.0BF1616.9 GBF1660details
Ling 3.0 Tiny7.9B (1.4B act.)F1615.6 GBF16347details
MiMo 7B RL7.8BF1616.4 GBF1661details
InternLM2.5 7B Chat7.7BF1616.0 GBF1662details
InternLM2 Chat 7B7.7BF1616.0 GBF1662details
Qwen2.5 7B Instruct7.6BF1615.3 GBF1663details
DeepSeek R1 Distill Qwen 7B7.6BF1615.3 GBF1663details
Qwen2.5 Coder 7B Instruct7.6BF1615.3 GBF1663details
Olmo 3 7B Instruct7.3BF1616.7 GBF1666details
Olmo 3 1025 7B7.3BF1616.7 GBF1666details
Olmo 3 7B Think7.3BF1616.7 GBF1666details
Mistral 7B v0.17.2BF1614.9 GBF1666details
Mistral 7B Instruct v0.27.2BF1614.9 GBF1666details
Mistral 7B Instruct v0.17.2BF1614.9 GBF1666details
Falcon 7B7.2BF1614.0 GBF1667details
Falcon 7B Instruct7.2BF1614.0 GBF1667details
Granite 4.0 H Tiny6.9B (1.5B act.)F1613.6 GBF16327details
OLMoE 1B 7B 09246.9B (1.3B act.)F1614.4 GBF16374details
DeepSeek Coder 7B Instruct v1.56.9BF1617.2 GBQ5_K_M69details
DeepSeek Coder 6.7B Instruct6.7BF1617.0 GBQ4_K_M71details
DeepSeek Coder 6.7B Base6.7BF1617.0 GBQ4_K_M71details
Llama 2 7B HF6.7BF1617.0 GBQ4_K_M71details
Yi 6B 200K6.1BF1612.3 GBF1679details
Yi 6B6.1BF1612.3 GBF1679details
Molmo2 4B4.9BF1610.0 GBF16109details
WeVisOne VL 4B4.8BF1610.0 GBF16109details
Mage VL Cowork4.7BF1610.0 GBF16109details
Qianfan OCR4.7BF1610.1 GBF16109details
Qwen3.5 4B4.7BF169.2 GBF16111details
SingVL S1 4B4.5BF169.0 GBF16114details
LightOnOCR 3 4B4.5BF169.0 GBF16114details
Intern Decision 4B4.5BF169.0 GBF16114details
Qwen3 VL 4B Instruct4.4BF169.3 GBF16119details
Gemma 3 4B IT4.3BF168.4 GBF16124details
MedGemma 1.5 4B IT4.3BF168.4 GBF16124details
MedGemma 4B IT4.3BF168.4 GBF16124details
TranslateGemma 4B IT4.3BF169.2 GBF16112details
Gemma 3 4B PT4.3BF168.4 GBF16124details
Phi 3.5 Vision Instruct4.1BF1611.2 GBF16116details
Qwen3 4B4.0BF169.3 GBF16119details
Qwen3 4B Instruct 25074.0BF169.3 GBF16119details
Granite Vision 4.1 4B4.0BF167.6 GBF16141details
Phi 4 Mini Instruct3.8BF168.9 GBF16125details
Phi 3 Mini 4K Instruct3.8BF1610.6 GBF16126details
Phi 3.5 Mini Instruct3.8BF1610.6 GBF16126details
Phi 3 Mini 128K Instruct3.8BF1610.6 GBF16126details
Qwen2.5 VL 3B Instruct3.8BF168.0 GBF16128details
Granite 4.2 3B3.7BF168.0 GBF16131details
Granite 4.1 3B Llmon Preview3.7BF168.0 GBF16131details
Raon 1 Nano3.5BF167.0 GBF16153details
Granite 4.1 3B3.4BF167.6 GBF16141details
Granite 4.0 Micro3.4BF167.6 GBF16141details
DeepSeek OCR 23.4B (1.2B act.)F166.6 GBF16464details
DeepSeek VL2 Tiny3.4B (1.2B act.)F167.4 GBF16408details
Tiny Aya Global3.3BF167.5 GBF16143details
Unlimited OCR 23.3B (1.1B act.)F166.6 GBF16475details
Unlimited OCR3.3B (1.1B act.)F166.6 GBF16478details
DeepSeek OCR3.3B (1.1B act.)F166.6 GBF16478details
Granite 3.1 3B A800M Instruct3.3B (883M act.)F167.1 GBF16543details
Llama 3.2 3B Instruct3.2BF167.5 GBF16149details
Llama 3.2 3B3.2BF167.5 GBF16149details
Hermes 3 Llama 3.2 3B3.2BF167.5 GBF16149details
d1 3B3.1BF165.8 GBF16178details
LFM2.5 VL 3B3.1BF165.8 GBF16178details
Qwen2.5 3B Instruct3.1BF166.7 GBF16156details
SmolLM3 3B3.1BF166.9 GBF16156details
SmolLM3 3B Base3.1BF166.9 GBF16156details
Phi 22.8BF168.2 GBF16173details
LFM2.5 2.6B2.7BF1610.8 GBF1689details
Granite 3.3 2B Instruct2.5BF165.8 GBF16189details
MiniCPM5 2B2.5BF165.7 GBF16191details
WeVisOne VL 2B2.4BF165.4 GBF16236details
Cosmos Reason2 2B2.4BF165.4 GBF16236details
Qwen3.5 2B2.3BF164.6 GBF16246details
SmolVLM Instruct2.2BF165.4 GBF16265details
Qwen3 1.7B2.0BF165.4 GBF16236details
d1 2B1.9BF164.0 GBF16277details
LFM3 VL 2B1.9BF164.0 GBF16277details
DeepSeek R1 Distill Qwen 1.5B1.8BF164.2 GBF16270details
LFM3 1.7B1.7BF167.2 GBF16138details
LFM3 1.7B Base1.7BF164.0 GBF16277details
SmolLM2 1.7B Instruct1.7BF165.2 GBF16280details
SmolLM2 1.7B1.7BF165.2 GBF16280details
SmolLM 1.7B1.7BF165.2 GBF16280details
Granite 4.0 1B1.6BF164.3 GBF16294details
Qwen2.5 1.5B Instruct1.5BF163.8 GBF16311details
OLMo 2 0425 1B1.5BF164.4 GBF16323details
DeepSeek Coder 1.3B Instruct1.3BF164.5 GBF16356details
Llama 3.2 1B1.2BF163.2 GBF16387details
Llama 3.2 1B Instruct1.2BF163.2 GBF16387details
LFM2.5 1.2B Instruct1.2BF165.0 GBF16205details
LFM2 1.2B1.2BF163.0 GBF16410details
HunyuanOCR1.1BF163.5 GBF16429details
PaddleOCR VL 1.6959MF162.5 GBF16501details
Qwen3.5 0.8B873MF162.4 GBF16616details
LightOnOCR 3 0.8B853MF162.4 GBF16633details
FastVLM 0.5B759MF162.2 GBF16633details
Qwen3 0.6B752MF163.0 GBF16636details
GOT OCR2.0716MF162.8 GBF16670details
Nemotron 3.5 Parse631MF161.9 GBF16735details
GOT OCR 2.0 HF561MF162.5 GBF16856details
SmolVLM2 500M Video Instruct507MF161.6 GBF161170details
SmolVLM 500M Instruct507MF161.6 GBF161170details
Qwen2.5 0.5B Instruct494MF161.7 GBF16972details
LFM2.5 VL 450M449MF161.3 GBF161349details
SmolLM2 360M Instruct362MF161.5 GBF161323details
SmolLM2 360M362MF161.5 GBF161323details
LFM2.5 350M354MF161.9 GBF16675details

Bigger models with CPU offload

These popular models don't fit in the RX 7900 XTX'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 · ~8.1–16 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

Comparing cards? The RX 7900 XTX (launch MSRP $999) is in best GPUs for local LLMs under $1,000. See also Mac vs PC.

Just too big at 4-bit

These need up to 35% more memory than the RX 7900 XTX 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 RX 7900 XTX?

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 RX 7900 XTX'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: AMD specifications. Model data: Hugging Face Hub API. Speed is a bandwidth ceiling (bandwidth ÷ bytes read per token), not a benchmark. Methodology · Report an error.