Deployment options
Recommended configurations based on usage scenarios:
- Basic Development Edition - RTX 3090 graphics card (24G memory) + CUDA 11.7 + 32G RAM for video testing up to 1080P
- Business Applications Edition - 2×RTX 4090 (bridged via NVLink) + 64G RAM + PCIe 4.0 SSD, supports 720P batch generation
- Cloud Solutions - AWS p4d.24xlarge instance (8×A100 GPUs) combined with Docker containerized deployment
Performance Tuning Guide
Key optimization parameters:
- Set mixed_precision in config.yaml: fp16
- Adjust the number of MoE specialists to 4 (default 8 reduces speed by 10-151 TP3T)
- Automated Mixing Accuracy with Torch's AMP
- Enable chunking with -chunk_size 32 parameter for video generation
This answer comes from the articleWan: a tool for video and image generation based on the Wan family of modelsThe
May not be reproduced without permission:AI productivity tools " What hardware configuration optimization tips are needed when deploying Wan 2.2 models locally?
