MetaX · Supernode / Cluster
C550 3D Mesh Supernode
The C550 3D Mesh Supernode, developed by MetaX and its partners, delivers powerful multi-precision computing capabilities. It supports 3D Mesh topology with low-cost, low-latency electrical interconnection for a supernode of up to 8 servers and 64 cards. It perfectly supports ultra-large-scale computing centers, general-purpose computing, and data processing applications.
Key Highlights
- Standard OAM server featuring self-developed GPGPU and mainstream CPU
- Single supernode supports 64-card in-memory semantic electrical interconnect with ultra-low communication latency
- 3D mesh architecture with open bus technology enables hyperscale data center deployment
- Full-stack solution from chip to cabinet for rapid deployment
Technical Specifications
| Supernode Scale | 最高8机64卡，3D Mesh电互连拓扑 / up to 8 servers 64 GPUs |
|---|---|
| 3D Mesh electrical topology | |
| Interconnect | 单超节点64卡内存语义电互连，通信延迟极低 / 64-card in-memory semantic electrical interconnect, ultra-low latency |
| Base Form | 标准OAM服务器，内置自研GPGPU与主流CPU / standard OAM server with self-developed GPGPU + mainstream CPU |
| Delivery | 从芯片到机柜全栈解决方案，支持快速部署 / chip-to-cabinet full-stack solution, rapid deployment |
| Accelerator | 曦云C550 OAM模组（64GB |
| 896GB/s聚合 | |
| 最大450W/模组） | |
Typical Use Cases
Large Language ModelsLarge-scale trainingHigh-concurrency inferenceIntelligent Computing CenterData Processing
img1: main; alt=官网产品页主视觉(banner); 官方图右下角标注"部分图片由大模型生成"，为含绿色科技背景的产品渲染主图。; 单卡精度算力、整机功耗、交换机与网卡型号官网未公开；无公开Datasheet PDF。
