# Can I run DeepSeek V3 on Nvidia GeForce RTX 5090 (32GB)?

> **No, not enough memory.** Needs ~383.7 GB even at Q4_K_M, but only ~31 GB is usable.

Last validated: 2026-06-15.

## Memory math
- DeepSeek V3: 671B (MoE, 37B active); Q4_K_M weighs 376.66 GB on disk.
- At Q4_K_M with a 4k context it needs ~383.7 GB (weights + KV cache + ~0.8 GB runtime overhead, ±15% with context length).
- Nvidia GeForce RTX 5090 (32GB): 32 GB vram, ~31 GB usable for model weights.
- Headroom: ~-352.7 GB.

## How to run
Recommended quant: n/a. Best tool on Windows: LM Studio.

```sh
ollama run deepseek-v3:671b
```

## Other models that run on Nvidia GeForce RTX 5090 (32GB)
- [Mixtral 8x7B (46.7B)](https://localmodel.run/can-i-run/mixtral-8x7b/nvidia-rtx-5090-32gb)
- [Command R 35B (35B)](https://localmodel.run/can-i-run/command-r-35b/nvidia-rtx-5090-32gb)
- [Yi 1.5 34B (34B)](https://localmodel.run/can-i-run/yi-1.5-34b/nvidia-rtx-5090-32gb)
- [Qwen2.5 32B (32B)](https://localmodel.run/can-i-run/qwen2.5-32b/nvidia-rtx-5090-32gb)
- [Qwen3 32B (32B)](https://localmodel.run/can-i-run/qwen3-32b/nvidia-rtx-5090-32gb)
- [DeepSeek-R1-Distill-Qwen 32B (32B)](https://localmodel.run/can-i-run/deepseek-r1-distill-qwen-32b/nvidia-rtx-5090-32gb)

Estimate, not a guarantee. Sources: https://ollama.com/library/deepseek-v3, https://ollama.com/library/deepseek-v3/tags, https://huggingface.co/unsloth/DeepSeek-V3-0324-GGUF, https://huggingface.co/deepseek-ai/DeepSeek-V3-0324.
More: https://localmodel.run/can-i-run/deepseek-v3/nvidia-rtx-5090-32gb