How to install DeepSeek on a server – step-by-step instructions for a VPS

DeepSeek AI – an advanced open-source artificial intelligence model that copes perfectly with its tasks even without using a GPU. Combined with Ollama, the technology makes it possible to run AI locally, allowing the user to fully control performance and the level of privacy.

Hosting DeepSeek on a personal server guarantees a high level of security, as the risk of data being intercepted via the API is eliminated. With the official DeepSeek servers frequently overloaded, a local installation lets you use only your own computing resources without sharing them with outsiders.

---

What specifications will we need?

Let's look at a comparison of models with different system requirements. All models are adapted to run on both CPU and GPU. However, given that the installation is done on a server, this article will focus mainly on setting up and running the model on the computing power of the processor.

Model version	Size	RAM	Suitable plan
deepseek-r1:1.5b	1.1 GB	4 GB	DE-7950X3D-4	Simple tasks, great for testing and trial runs.
deepseek-r1:7b	4.7 GB	10 GB	DE-7950X3D-5	Writing and translating texts. Developing simple code.
deepseek-r1:14b	9 GB	20 GB	DE-7950X3D-6	Advanced capabilities in development and copywriting. An excellent balance of speed and functionality.
deepseek-r1:32b	19 GB	40 GB	DE-7950X3D-6	Power comparable to ChatGPT o1 mini. In-depth data analysis.
deepseek-r1:70b	42 GB	85 GB	Minimum dedicated server	High-level computing for business tasks. Deep data analysis and complex development.
deepseek-r1:671b	720 GB	768 GB	Minimum dedicated server	The main and most advanced DeepSeek R1 model. Provides computing capabilities on a par with the latest ChatGPT models. We recommend hosting it on a powerful dedicated server with NVMe drives.

---

Step-by-step installation of the DeepSeek 14B model on a server

As an example, we will install the 14B model, which provides the optimal balance between performance and resource consumption. These instructions apply to any model from the available list, allowing you to install a different version if necessary.

WARNING

The installation uses the Hi-CPU Pulsar plan with the Ubuntu 22.04 operating system, which is ideal for DeepSeek.

Update the system repositories and packages:

sudo apt update && sudo apt upgrade -y

Install Ollama – a manager for deploying DeepSeek:

curl -fsSL https://ollama.com/install.sh | sh && sudo systemctl start ollama

Download the DeepSeek model you have chosen:

ollama pull deepseek-r1:14b

INFO

Note: You can replace deepseek-r1:14b with another version, for example, deepseek-r1:32b. Installing the 14B model takes approximately 2 minutes on a Hi-CPU Pulsar server thanks to the high-speed network connection.

Run the DeepSeek model:

ollama run deepseek-r1:14b

After you run this command, an interface for entering messages will appear, allowing you to start a dialogue with the artificial intelligence.

---

Additional commands

You can also send a single query:

ollama run deepseek-r1:14b "What is heavier: 1 kg of iron or 1 kg of feathers?"

To create public API access:

ollama serve

To check the installed models:

ollama list

---

Recommendations for choosing a server

For stable operation of DeepSeek R1 models, it is recommended to use servers with a sufficient amount of RAM (preferably with some headroom) and fast NVMe drives.

---

Our resources

Telegram channel: https://t.me/
F3 Cloud: https://f3cloud.com