Knowledge Base · Local AI
How much VRAM to run DeepSeek locally?
By Tiny · how I pick
As an Amazon Associate I earn from qualifying purchases. The graphics card links below are affiliate links.
The DeepSeek most people run at home is one of the R1 distilled models, and they fit on regular graphics cards. At 4-bit:
- 7B: about 5 GB
- 14B: about 10 GB
- 32B: about 20 GB
The full DeepSeek-R1 (671B) is a data center model that needs about 380 GB spread across several graphics cards. For a home build, ignore it and pick a distill that fits your card.
Words you'll see in this guide
- VRAM: the memory on your graphics card. AI models load into it, so more VRAM means you can run bigger models.
- Distilled model: a smaller model trained to copy how a bigger one answers. Much easier to run, a bit less capable.
- Parameters (7B, 14B, 671B): a rough measure of a model's size. "8B" means 8 billion. Bigger models are usually smarter and need more memory.
- Quantization (4-bit): shrinking a model by storing its numbers less precisely, so it fits in less memory with a small drop in quality. Think of saving a photo as a smaller JPEG. "4-bit" (Q4) is the common setting.
- Context: how much text the model can keep in mind at once, measured in tokens. Longer context uses more memory.
VRAM by DeepSeek model
| Model | VRAM (4-bit) | Runs on |
|---|---|---|
| R1 distill 1.5B | About 1.5 GB | Almost anything, even a Raspberry Pi |
| R1 distill 7B | About 5 GB | An 8 GB card like the RTX 4060 |
| R1 distill 8B | About 6 GB | An 8 GB card |
| R1 distill 14B (best balance) | About 10 GB | A 16 GB card, with long conversations |
| R1 distill 32B | About 20 GB | A 24 GB card like the RTX 3090 or RTX 4090 |
| R1 distill 70B | About 40 to 48 GB | Two 24 GB cards, or a 48 GB workstation card |
| Full R1 671B | About 380 GB | A multi-card server, not a home setup |
- 16 GB card: the 14B distill. It's the best balance of smarts and speed.
- 8 GB card: the 7B or 8B, which handle everyday use well.
- 24 GB card: the 32B distill, which gets close to the full model's quality.
How to run it
The easiest way is Ollama with Open WebUI. Check what your machine can handle first with the Server Calculator. Then:
- Install Ollama (the guide above walks through it).
- Download and start the 14B distill:
ollama run deepseek-r1:14bYou should see: a
>>>prompt once the download finishes. - Type a question. For a ChatGPT-style window in your browser, add Open WebUI.
Where to go next
- Buying a card for this? See the best GPU for local AI under $500 and what a 16 GB vs 24 GB card runs.
- Comparing models? See VRAM for Qwen 3 and VRAM for Llama 3 70B.
Quick answers
How much VRAM do I need to run DeepSeek locally?
Most people run the DeepSeek-R1 distilled models, which fit on regular graphics cards at 4-bit: about 5GB for the 7B, 10GB for the 14B, and 20GB for the 32B. The full DeepSeek-R1 (671B) needs roughly 380GB across many data center cards.
Can I run the full DeepSeek-R1 671B at home?
Realistically, no. It needs around 380GB of memory even compressed. At home, you run the distilled versions from 1.5B to 70B.
Which DeepSeek distill should I run?
The 14B on a 16GB card is the sweet spot for most people. On 8GB, run the 7B or 8B. On 24GB, the 32B gets close to the full model's quality. The 70B needs about 40 to 48GB, usually two 24GB cards.
Are the DeepSeek distills as good as the full model?
Not quite, but close for their size. They're smaller models trained to copy the full R1's reasoning, and the bigger the distill, the closer it gets. The 14B and 32B are useful for everyday reasoning and coding help.
Free email updates
New DeepSeek releases come fast. Subscribe and I'll tell you which ones fit the card you already have.
No spam. Unsubscribe anytime. Privacy policy.