NVIDIA · 32GB VRAM
A GeForce RTX 5090 has 32GB of VRAM. Checked against 225 models from the Ollama library, 201 have good/excellent reference fits on the GPU at 4K context and 9 more run with part of the model in system RAM.
201
run well
9
with trade-offs
225
builds evaluated
The 32 GB desktop model has 8 GB more capacity than the RTX 4090. Compare the exact quantization before assuming a larger model will fit; model weights and context still compete for the same pool.
ollama ps. Intel Macs use the Ollama CPU path; Apple Silicon uses its integrated Metal GPU.Hardware specifications · Ollama support table · reviewed 2026-09-15.
Compare GeForce RTX 4090 (24 GB). Use the same model, quantization and context to see what the capacity difference changes.
Editorial examples chosen for their documented tasks, all Q4_K_M at 4,096 tokens. Check the verdict before downloading. These are estimates, not benchmark rankings.
A small instruction-tuned model for trying local chat with more memory headroom. Review summaries against the source text.
Task and build source · Catalog checked 2026-10-13
Est. memory 2.9 GB · medium confidence
Model fully fits in usable GPU VRAM.
ollama run llama3.2:3bAn 8B instruction-tuned alternative to compare on your own prompts. It needs more memory than the 3B example and can still give incorrect answers.
Task and build source · Catalog checked 2026-10-13
Est. memory 5.8 GB · medium confidence
Model fully fits in usable GPU VRAM.
ollama run llama3.1:8bA code-focused instruct build. Review generated changes and run your tests; a memory fit is not evidence that its code is correct.
Task and build source · Catalog checked 2026-10-13
Est. memory 5.3 GB · medium confidence
Model fully fits in usable GPU VRAM.
ollama run qwen2.5-coder:7bA 14B instruct build for comparing answers when memory allows. Larger size does not establish better results for every task.
Task and build source · Catalog checked 2026-10-13
Est. memory 10.2 GB · medium confidence
Model fully fits in usable GPU VRAM.
ollama run qwen2.5:14bAfter opening Ollama's prompt, enter /set parameter num_ctx 4096 to match this calculation, within the model's supported maximum. Install a current Ollama release that supports the build. Setup and troubleshooting.
Largest first — these have good/excellent reference fits with 32GB of VRAM at 4K context. A good fit can include a small estimated CPU offload. These are technical defaults, not a quality ranking.
| Model | Build | Memory needed | Verdict |
|---|---|---|---|
| dolphin-mixtralOllama Library | dolphin-mixtral:8x7b-v2.6-q4_K_MQ4_K_M | ~29.5 GB | Good fit |
| nous-hermes2-mixtralOllama Library | nous-hermes2-mixtral:8x7b-dpo-q4_K_MQ4_K_M | ~29.5 GB | Good fit |
| mixtralMistral AI | mixtral:8x7b-text-v0.1-q4_K_MQ4_K_M | ~29.5 GB | Good fit |
| nemotron-3.5-lightningNVIDIA | nemotron-3.5-lightning:30bQ4_K_M | ~29.3 GB | Good fit |
| nemotron-cascade-2NVIDIA | nemotron-cascade-2:30bQ4_K_M | ~28.1 GB | Good fit |
| notuxOllama Library | notux:8x7b-v1-q4_K_MQ4_K_M | ~27.5 GB | Good fit |
| alfredOllama Library | alfred:40bQ4_0Memory scenario exceeds documented maximum of 2,048 tokens. | ~24.8 GB | Good fit |
| qwen3.6Alibaba Cloud | qwen3.6:27b-q4_K_MQ4_K_M | ~23.9 GB | Good fit |
| codeboogaOllama Library | codebooga:34b-v0.1-q4_K_MQ4_K_M | ~21.5 GB | Excellent fit |
| phind-codellamaMicrosoft | phind-codellama:34b-python-q4_K_MQ4_K_M | ~21.5 GB | Excellent fit |
| qwqAlibaba Cloud | qwq:32b-preview-q4_K_MQ4_K_M | ~21.4 GB | Excellent fit |
| olmo-3.1Allen Institute for AI | olmo-3.1:32bQ4_K_M | ~21.0 GB | Excellent fit |
| command-rCohere | command-r:35b-08-2024-q4_K_MQ4_K_M | ~20.9 GB | Excellent fit |
| qwen3-coderAlibaba Cloud | qwen3-coder:30bQ4_K_M | ~19.4 GB | Excellent fit |
| clefOllama Library | clef:27bQ4_K_M | ~18.5 GB | Excellent fit |
| qwen3.8Alibaba Cloud | qwen3.8:27bQ4_K_M | ~18.3 GB | Excellent fit |
| muse-glimmerOllama Library | muse-glimmer:30bQ4_K_M | ~17.4 GB | Excellent fit |
| mistral-small3.1Mistral AI | mistral-small3.1:24bQ4_K_M | ~16.6 GB | Excellent fit |
| devstral-small-2Mistral AI | devstral-small-2:24bQ4_K_M | ~16.3 GB | Excellent fit |
| mistral-small3.2Mistral AI | mistral-small3.2:24bQ4_K_M | ~16.3 GB | Excellent fit |
| lfm2Ollama Library | lfm2:24bQ4_K_M | ~16.1 GB | Excellent fit |
| devstralMistral AI | devstral:24bQ4_K_M | ~15.4 GB | Excellent fit |
| magistralMistral AI | magistral:24bQ4_K_M | ~15.4 GB | Excellent fit |
| solar-proUpstage | solar-pro:22bQ4_K_M | ~15.0 GB | Excellent fit |
| mistral-smallMistral AI | mistral-small:22b-instruct-2409-q4_K_MQ4_K_M | ~14.7 GB | Excellent fit |
Showing the 25 largest of 201 models that run well.
Each model is evaluated at 4K context against 32GB of VRAM, counting model weights, KV cache, compute buffers and runtime overhead, minus a reserve for the display and operating system. Download sizes come from the official Ollama registry.
Reference scenario: Windows 11, 32GB of system RAM, an illustrative 8-core/16-thread CPU and 500GB free disk. Available RAM, driver version and free VRAM are unknown; the engine applies reserves. Supported GPU drivers/backends are assumed installed. Your own machine may differ — RAM, free disk space and whether an acceleration backend is actually installed all change the answer, which is what the PC scan measures directly.