Back to articles
AI & Cloud

Self-Hosting AI Tools on VPS: Save 80% Compared to ChatGPT and Midjourney

A detailed guide on self-hosting AI tools like ChatGPT and Midjourney on a VPS to save up to 80% compared to commercial cloud services. Includes configuration, optimization, and...

DPTCloud Engineering4 min read
Self-Hosting AI Tools on VPS: Save 80% Compared to ChatGPT and Midjourney

Why Self-Hosting AI on VPS Saves 80% in Costs?

Using commercial AI services like ChatGPT or Midjourney can cost hundreds of dollars monthly, especially with high usage. However, by self-hosting AI tools on a VPS (Virtual Private Server) with ample RAM, you can save up to 80% compared to cloud-based services. This approach is ideal for businesses, developers, or individuals who want full control over their AI environment and avoid reliance on third-party providers.

Requirements and Preparations for Self-Hosting AI on VPS

To successfully deploy, you need a VPS with sufficient hardware. Below are the basic requirements:

  • RAM: Minimum 16GB, preferably 32GB or more to run large language models (LLMs) like Llama 3, Mistral, or Phi.
  • Storage: SSD with at least 100GB to store models and data.
  • Processor: 8-core CPU or higher, preferably with GPU support (e.g., NVIDIA T4, A10G).
  • Bandwidth: Minimum 1Gbps for fast AI query processing.

You can choose a High-RAM Cloud VPS or a Dedicated Server based on your needs. VPS monthly costs range from $10–$50, while commercial services can cost $200–$500/month.

Deploying Self-Hosted AI Tools on VPS

Here are the basic steps to self-host popular AI tools:

1. Self-Hosting LLMs (Large Language Models)

Run models like Llama 3 or Mistral using llama.cpp or Text Generation WebUI:

git clone https://github.com/oobabooga/text-generation-webui.git
cd text-generation-webui
pip install -r requirements.txt
python server.py --model llama-3-8b --n-gpu-layers 40 --n_ctx 4096

After running, access the web interface at http://your-vps-ip:7860.

2. Self-Hosting Midjourney-like Image Generation

Use Stable Diffusion with Automatic1111:

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
cd stable-diffusion-webui
git checkout v1.5
python launch.py --listen 0.0.0.0 --port 7860 --theme dark

Ensure you have a CUDA-supported GPU and enough RAM (5–10GB for SD 1.5).

Optimizing Performance and Reducing Costs

To achieve high performance and cost savings, apply these strategies:

  • Choose the right model: Use smaller models like Phi-2 or Mistral-7B instead of Llama-3-70B if extreme accuracy isn't needed.
  • Run on GPU: Use NVIDIA GPUs with CUDA to accelerate processing.
  • Optimize model loading: Apply quantization (8-bit or 4-bit) to reduce memory usage.
  • API gateway: Use FastAPI or Flask to expose APIs for applications.

Testing and Verifying Performance

Run the following commands to check performance:

# Check GPU status
nvidia-smi

# Monitor RAM and CPU
htop

# Test query speed
curl -X POST http://localhost:7860/api/v1/generate -d '{"prompt": "Hello", "max_tokens": 50}'

Troubleshooting Common Issues

  • Model fails to start: Check RAM, GPU, and CUDA version.
  • Remote access blocked: Verify firewall rules and ports (7860, 8000, etc.).
  • Low performance: Reduce max_tokens, use a smaller model, or enable quantization.

References

DPTCloud Tier III Infrastructure

High-RAM Cloud VPS Plans Optimized for AI, Databases & Cache

Enterprise RAID 10 NVMe SSDs, CMC & FPT District 9 uplinks with Anycast Anti-DDoS and 99.99% SLA.

View 22 VPS Plans
ECO · vn-02Recommended

VPS-4E

297.000 ₫ / mo

10% VAT included

  • CPU: 4 vCPU
  • RAM: 8 GB
  • SSD: 50 GB NVMe
ECO · vn-02

VPS-5E

614.900 ₫ / mo

10% VAT included

  • CPU: 8 vCPU
  • RAM: 16 GB
  • SSD: 100 GB NVMe
ADVANCED · vn-03Recommended

VPS-4T

548.900 ₫ / mo

10% VAT included

  • CPU: 4 vCPU
  • RAM: 8 GB
  • SSD: 50 GB NVMe

Related Articles

View all
ZaloZalo SupportDiscord0384 753 478