Can I run this LLM? — fit by model × hardware

Computed with the open FitLLM engine — accurate per-layer KV-cache modeling, not a naive estimate. Updated 2026-06.

Each page is computed by the open FitLLM engine from official model configs.

Guides

• Why naive VRAM calculators are wrong

Reference

• LLM memory & architecture reference

Best GPU / Mac by model

• Best GPU or Mac for Qwen 3.6 35B-A3B
• Best GPU or Mac for Qwen 3.6 27B
• Best GPU or Mac for Gemma 4 31b
• Best GPU or Mac for Gemma 4 26b A4B
• Best GPU or Mac for Gemma 4 12b

GPU vs GPU

• RTX 4090 vs RTX 3090
• RTX 5090 vs RTX 4090
• RTX 5080 vs RTX 4080 SUPER
• RTX 5090 vs RTX 5080
• RTX 4090 vs RTX 5080
• RTX 3090 vs RTX 4080 SUPER

By model × hardware

• qwen 3 6 35b a3b on rtx 3090
• qwen 3 6 27b on rtx 3090
• gemma 4 31b on rtx 3090
• gemma 4 26b a4b on rtx 3090
• gemma 4 12b on rtx 3090
• qwen 3 6 35b a3b on rtx 4090
• qwen 3 6 27b on rtx 4090
• gemma 4 31b on rtx 4090
• gemma 4 26b a4b on rtx 4090
• gemma 4 12b on rtx 4090
• qwen 3 6 35b a3b on rtx 4080 super
• qwen 3 6 27b on rtx 4080 super
• gemma 4 31b on rtx 4080 super
• gemma 4 26b a4b on rtx 4080 super
• gemma 4 12b on rtx 4080 super
• qwen 3 6 35b a3b on rtx 5090
• qwen 3 6 27b on rtx 5090
• gemma 4 31b on rtx 5090
• gemma 4 26b a4b on rtx 5090
• gemma 4 12b on rtx 5090
• qwen 3 6 35b a3b on rtx 5080
• qwen 3 6 27b on rtx 5080
• gemma 4 31b on rtx 5080
• gemma 4 26b a4b on rtx 5080
• gemma 4 12b on rtx 5080
• qwen 3 6 35b a3b on rtx 3060 12gb
• qwen 3 6 27b on rtx 3060 12gb
• gemma 4 31b on rtx 3060 12gb
• gemma 4 26b a4b on rtx 3060 12gb
• gemma 4 12b on rtx 3060 12gb
• qwen 3 6 35b a3b on m5 max 128gb
• qwen 3 6 27b on m5 max 128gb
• gemma 4 31b on m5 max 128gb
• gemma 4 26b a4b on m5 max 128gb
• gemma 4 12b on m5 max 128gb
• qwen 3 6 35b a3b on m5 max 64gb
• qwen 3 6 27b on m5 max 64gb
• gemma 4 31b on m5 max 64gb
• gemma 4 26b a4b on m5 max 64gb
• gemma 4 12b on m5 max 64gb
• qwen 3 6 35b a3b on m5 pro 48gb
• qwen 3 6 27b on m5 pro 48gb
• gemma 4 31b on m5 pro 48gb
• gemma 4 26b a4b on m5 pro 48gb
• gemma 4 12b on m5 pro 48gb

All numbers are computed by the open-source fitllm-engine (MIT) from official model config.json values — reproduce or audit them yourself. Estimates; real usage varies with runtime (llama.cpp / MLX / Ollama), driver and display. Found a mismatch? Report it. · FitLLM home