.png)
4 months ago
7
camd (cheapamd) is a command-line tool that helps you find available AMD hardware across cloud providers. With the massive 192GB memory of MI300X GPUs and powerful EPYC CPUs, AMD offers compelling alternatives to NVIDIA hardware.
- MI300X GPU: 192GB HBM3 memory (2.4x more than H100!)
- High Performance: Excellent compute capabilities
- EPYC CPUs: Best price/performance for CPU workloads
- Availability: Often easier to find than scarce H100s
- 🔍 Multi-Provider Search: Vultr and RunPod support
- 💎 AMD GPU Discovery: Find MI300X (192GB) and MI250X (128GB)
- 💻 AMD CPU Discovery: All EPYC variants (Milan, Rome, Genoa)
- 💰 Price Comparison: Sort by hourly cost
- 🏷️ Spot Pricing: 50% discounts on RunPod
- 📦 Multi-GPU Configs: 1x, 2x, 4x, 8x GPU clusters
- ⚡ Smart Caching: 5-minute cache to reduce API calls
- 🎨 Beautiful CLI: Color-coded output with emojis
- 🔐 Secure: API keys stored locally with 600 permissions
# Download the script
curl -O https://raw.githubusercontent.com/modelturnedgeek/CheaperNvidia/main/camd.py
chmod +x camd.py
# Install system-wide
sudo cp camd.py /usr/local/bin/camd
# Or install for current user
mkdir -p ~/.local/bin
cp camd.py ~/.local/bin/camd
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.bashrc
source ~/.bashrc
- Python 3.6+
- requests library (pip install requests)
You'll be guided to get API keys from:
- RunPod: https://www.runpod.io/console/user/settings
- Vultr: https://my.vultr.com/settings/#settingsapi
# List all AMD hardware (GPUs + CPUs)
camd list
# List only AMD GPUs
camd list gpu
# List only AMD CPUs
camd list cpu
💰 camd v6.0.0 - Checking AMD hardware availability...
━━━ AMD GPU Instances ━━━
MI300X: 192GB HBM3 | 5.3TB/s | 1307.4 TFLOPS
💵 $/hr Provider Model Count VRAM Type Available
─────────────────────────────────────────────────────────────────────────────────
$1.25 RunPod MI300X 1 192GB MI300X-spot ✓
$2.49 RunPod MI300X 1 192GB MI300X-ondemand ✓
$2.50 Vultr MI300X 1 192GB gpu-mi300x-1 ✓
$5.00 Vultr MI300X 2 384GB gpu-mi300x-2 ✓
━━━ AMD CPU Instances ━━━
AMD EPYC processors - Industry leading performance
💵 $/hr Provider Type vCPUs RAM Category
─────────────────────────────────────────────────────────────────────────────────
$0.01 Vultr vhf-1c-1gb-amd 1 1GB High Frequency AMD
$0.01 Vultr vhp-1c-1gb-amd 1 1GB High Performance AMD
$0.02 Vultr vhf-1c-2gb-amd 1 2GB High Frequency AMD
...
| RunPod | ✅ MI300X, MI250X | ❌ | Stable | Best for GPU workloads, spot pricing available |
| Vultr | 🔄 Limited | ✅ EPYC | Stable | Excellent CPU selection, some GPU availability |
- Strengths: GPU-focused, spot instances (50% off), global availability
- GPUs: MI300X ($2.49/hr), MI250X ($1.99/hr estimated)
- Features: Multi-GPU clusters, persistent storage, Jupyter support
- Strengths: Wide CPU selection, hourly billing, 25+ locations
- CPUs: EPYC 7003 (Milan), 7002 (Rome), 9004 (Genoa)
- Types: High Performance (vhp), Optimized Cloud (voc), High Frequency (vhf)
- 70B+ LLMs: Run Llama-70B on a single GPU!
- RAG Systems: Massive context windows
- Multi-modal AI: Image + text models
- Scientific Computing: Large memory requirements
- Web Hosting: Better price/performance than Intel
- Databases: High memory bandwidth
- Containers: Excellent multi-threading
- CI/CD: Cost-effective build servers
# API Keys
export RUNPOD_API_KEY='your-key'
export VULTR_API_KEY='your-key'
# Cache timeout (minutes)
export CAMD_CACHE_MINUTES=5
# Debug mode
export CAMD_DEBUG=1
# Location: ~/.camd/.env
RUNPOD_API_KEY=your_runpod_key
VULTR_API_KEY=your_vultr_key
CAMD_CACHE_MINUTES=5
We welcome contributions! Here's how to add a new provider:
- Create a new provider class inheriting from base
- Implement get_amd_hardware() method
- Add to provider initialization in load_config()
- Submit PR with example output
git clone https://github.com/modelturnedgeek/CheaperNvidia
cd CheaperNvidia
pip install requests # Only dependency
python camd.py setup
"No configuration found"
camd setup # Run setup first
"No AMD hardware found"
- Check API keys are valid
- Ensure you have network connectivity
- Try with debug mode: CAMD_DEBUG=1 camd list
API Rate Limits
- Results are cached for 5 minutes
- Adjust with CAMD_CACHE_MINUTES
MIT License - see LICENSE file
- AMD for making competitive hardware
- Cloud providers offering AMD instances
- The open-source community
