The RTX 5090 redefines what is possible with consumer GPUs. With 32GB GDDR7 and an unprecedented 145 tokens per second, it can load 70B parameter models with Q4 quantization. The ultimate card for running the largest local AI models.
| GPU | VRAM | Speed | Bandwidth | Price |
|---|---|---|---|---|
| RTX 3090 | 24 GB | 87 tok/s | 936 GB/s | $900 |
| RTX 5080 | 16 GB | 94 tok/s | 960 GB/s | $999 |
| RTX 5090 | 32 GB | 145 tok/s | 1792 GB/s | $2,499 |
| RTX 4090 | 24 GB | 104 tok/s | 1008 GB/s | $2,574 |
Qwen / 35B / Q4_K_M / ~20 GB
Best for: Reasoning, Coding, Agent scenarios·Pop: 90/100
Perf: ~41.4 tok/s · first token ~1.0s
Fits in 32 GB VRAM with room to spare. Best for reasoning, coding, agent scenarios on RTX 5090.
ollama run qwen3.5:35b-a3b-instruct-q4_K_M
Qwen / 27B / Q4_K_M / ~16 GB
Best for: Chat, Coding, Complex reasoning·Pop: 82/100
Perf: ~51.6 tok/s · first token ~0.4s
Fits in 32 GB VRAM with room to spare. Best for chat, coding, complex reasoning on RTX 5090.
ollama run qwen3.5:27b-instruct-q4_K_M
LFM2 / 24B / Q4_K_M / ~14 GB
Best for: Local AI agents, privacy-first tool calling, MCP workflows·Pop: 80/100
Perf: ~57.0 tok/s · first token ~0.4s
Fits in 32 GB VRAM with room to spare. Best for local ai agents, privacy-first tool calling, mcp workflows on RTX 5090.
ollama run liquidai/lfm2:24b-a2b-instruct-q4_K_M
Qwen / 14B / Q4_K_M / ~11 GB
Best for: Coding, Quality·Pop: 84/100
Perf: ~90.1 tok/s · first token ~0.4s
Fits in 32 GB VRAM with room to spare. Best for coding, quality on RTX 5090.
ollama run qwen3:14b-q4_K_M
Qwen / 14B / Q4_K_M / ~11 GB
Best for: Coding, Chat·Pop: 80/100
Perf: ~90.1 tok/s · first token ~0.4s
Fits in 32 GB VRAM with room to spare. Best for coding, chat on RTX 5090.
ollama run qwen2.5:14b-instruct-q4_K_M
Qwen / 14B / Q4_K_M / ~11 GB
Best for: Coding·Pop: 79/100
Perf: ~90.1 tok/s · first token ~0.4s
Fits in 32 GB VRAM with room to spare. Best for coding on RTX 5090.
ollama run qwen2.5-coder:14b-q4_K_M
Mistral / 12B / Q4_K_M / ~9.5 GB
Best for: Chat, Translation·Pop: 78/100
Perf: ~102.7 tok/s · first token ~0.3s
Fits in 32 GB VRAM with room to spare. Best for chat, translation on RTX 5090.
ollama run mistral-nemo:12b-q4_K_M
Gemma / 12B / Q4_K_M / ~9.5 GB
Best for: Chat, Quality·Pop: 76/100
Perf: ~102.7 tok/s · first token ~0.3s
Fits in 32 GB VRAM with room to spare. Best for chat, quality on RTX 5090.
ollama run gemma3:12b-instruct-q4_K_M
DeepSeek / 14B / Q4_K_M / ~11 GB
Best for: Reasoning, Quality·Pop: 74/100
Perf: ~90.1 tok/s · first token ~0.4s
Fits in 32 GB VRAM with room to spare. Best for reasoning, quality on RTX 5090.
ollama run deepseek-r1-distill:qwen-14b-q4_K_M
Phi / 14B / Q4_K_M / ~11 GB
Best for: Coding, Quality·Pop: 69/100
Perf: ~90.1 tok/s · first token ~0.4s
Fits in 32 GB VRAM with room to spare. Best for coding, quality on RTX 5090.
ollama run phi3:medium-q4_K_M
With 32GB VRAM, the RTX 5090 can run up to 70b parameter models. Top recommendations include Qwen3.5 35B-A3B Instruct, Qwen3.5 27B Instruct, LFM2 24B-A2B Instruct.
The RTX 5090 achieves 145 tokens per second with Qwen3 8B at Q4 quantization. Smaller models run faster, larger models slower.
32GB VRAM is excellent for local AI. You can comfortably run up to 70b parameter models with room for KV cache. 10 of our top 10 recommended models run at full speed.
Install Ollama from ollama.com, then run models directly. For example: ollama run qwen3.5:35b-a3b-instruct-q4_K_M. Ollama automatically detects your NVIDIA GPU and uses CUDA acceleration.
Use our interactive wizard to compare models across Apple Silicon and NVIDIA GPUs.
Open ModelFit Wizard →