NVIDIA · 12 GB GDDR · 672 GB/s

Yes. GeForce RTX 5070 fits Mistral Nemo 12B Instruct at Q4_K_M, the largest catalog model that clears the fit line at 8K context.

NVIDIA compare table: 12 GB GDDR7, 192-bit, 672 GB/s. Source.

Recommended quant, 8,192 tokens.
ModelQuantWeightsTotalVerdictSpeedOllama
Mistral Nemo 12B InstructQ4_K_M6.96 GiB9.15 GiBFits42–72 tok/sollama run mistral-nemo:12b
Gemma 4 12BQ4_K_M6.63 GiB7.95 GiBFits49–75 tok/sollama run gemma4:12b
Gemma 3 12B InstructQ4_K_M6.80 GiB8.52 GiBFits45–74 tok/sollama run gemma3:12b
Qwen3.5 9BQ4_K_M5.29 GiB6.39 GiBFits62–95 tok/sollama run qwen3.5:9b
Qwen3 8BQ4_K_M4.68 GiB6.62 GiBFits59–107 tok/sollama run qwen3:8b
Llama 3.1 8B InstructQ4_K_M4.58 GiB6.39 GiBFits62–109 tok/sollama run llama3.1:8b
Qwen2.5 7B InstructQ4_K_M4.36 GiB5.56 GiBFits72–115 tok/sollama run qwen2.5:7b
Gemma 4 E4BQ4_K_M4.64 GiB5.66 GiBFits70–108 tok/sollama run gemma4:e4b
Gemma 4 E2BQ4_K_M2.89 GiB3.69 GiBFits114–173 tok/sollama run gemma4:e2b
Gemma 3 4B InstructQ4_K_M2.32 GiB3.24 GiBFits133–216 tok/sollama run gemma3:4b
Phi-4-mini InstructQ4_K_M2.32 GiB4.01 GiBFits104–216 tok/sollama run phi4-mini
Llama 3.2 3B InstructQ4_K_M1.88 GiB3.42 GiBFits125–266 tok/sollama run llama3.2:3b
SmolLM3 3BQ4_K_M1.78 GiB2.99 GiBFits147–281 tok/s—
Llama 3.2 1B InstructQ4_K_M0.75 GiB1.58 GiBFits343–666 tok/sollama run llama3.2:1b
Qwen2.5 14B InstructQ4_K_M8.37 GiB10.9 GiBTight35–60 tok/sollama run qwen2.5:14b
DeepSeek-R1 Distill Qwen 14BQ4_K_M8.37 GiB10.9 GiBTight35–60 tok/sollama run deepseek-r1:14b
Qwen3 14BQ4_K_M8.38 GiB10.6 GiBTight36–60 tok/sollama run qwen3:14b
Phi-4Q4_K_M8.43 GiB11.0 GiBTight34–59 tok/sollama run phi4:14b
Qwen3.6 35B-A3BUD-Q4_K_M20.6 GiB22.4 GiBOffload25–36 tok/s—
Qwen2.5 32B InstructQ4_K_M18.5 GiB22.0 GiBOffload2–4 tok/sollama run qwen2.5:32b
DeepSeek-R1 Distill Qwen 32BQ4_K_M18.5 GiB22.0 GiBOffload2–4 tok/sollama run deepseek-r1:32b
Qwen3 32BQ4_K_M18.4 GiB22.0 GiBOffload2–4 tok/sollama run qwen3:32b
Gemma 4 31BQ4_K_M17.1 GiB19.9 GiBOffload3–5 tok/sollama run gemma4:31b
Qwen3 30B-A3BQ4_K_M17.3 GiB19.5 GiBOffload29–42 tok/sollama run qwen3:30b-a3b
Qwen3.8 27BUD-Q4_K_M15.3 GiB17.3 GiBOffload5–7 tok/s—
Gemma 3 27B InstructQ4_K_M15.4 GiB17.8 GiBOffload4–6 tok/sollama run gemma3:27b
Gemma 4 26B-A4BUD-Q4_K_M15.8 GiB17.5 GiBOffload33–47 tok/sollama run gemma4:26b-a4b
Mistral Small 3.2 24B InstructQ4_K_M13.3 GiB15.9 GiBOffload6–9 tok/s—
Devstral Small 2 24B InstructQ4_K_M13.3 GiB15.9 GiBOffload6–9 tok/s—
gpt-oss-20bMXFP411.3 GiB12.6 GiBOffload140–204 tok/sollama run gpt-oss:20b
gpt-oss-120bMXFP459.0 GiB62.8 GiBNo—ollama run gpt-oss:120b
Qwen2.5 72B InstructQ4_K_M44.2 GiB49.5 GiBNo—ollama run qwen2.5:72b
Llama 3.3 70B InstructQ4_K_M39.6 GiB44.7 GiBNo—ollama run llama3.3:70b

Exact answers

gpt-oss-20boffload · MXFP4Phi-4tight · Q4_K_MQwen3 14Btight · Q4_K_MMistral Small 3.2 24B Instructoffload · Q4_K_MGemma 4 12Bfits · Q4_K_MQwen3.8 27Boffload · UD-Q4_K_M

FAQ

What does the verdict for GeForce RTX 5070 mean?

Memory is the GGUF file, plus KV cache at 8,192 tokens, plus a 0.5 GiB engine allowance, then 5% headroom. Fits means that total is at most 85% of fast memory. Tight means it still fits. Offload means it needs system RAM. No means it does not fit in VRAM plus system RAM after an 8 GB Windows reserve. The formula is on the methodology page.

What memory did you count as usable?

Fast memory is the full 12 GB. System RAM is assumed to be 32 GB DDR5-5600, with 8 GB kept for Windows, and you can change that on the homepage calculator.

Why is speed a range?

Decode speed is memory bandwidth divided by the bytes read per token. The high end is 80% of that roofline on the weights alone. The low end is 55% once the KV cache at the working context is included. It is not a benchmark.

How should I run Qwen2.5 14B Instruct?

The table uses Q4_K_M. Ollama's matching command is `ollama run qwen2.5:14b`. That blob is the library default and can differ from the cited GGUF by a few hundred MB.