MetaX · Compute Server
N300 Server
The N300 Server, a standard rack-mounted server product developed by MetaX and its partners, is equipped with maximum 16 cards N300 per unit. It features an innovative PCIe Switch full-interconnect topology and powerful multi-precision computing and high-bandwidth interconnection topologies, making it ideal for high-density inference scenarios.
Key Highlights
- Supports multiple topologies including PCIe Switch full-mesh and Common, meeting the demands of high-density and high-performance scenarios
- Compatible with mainstream domestic and international CPUs
- Supports high-capacity 16-GPU and standard 8-GPU rack-mounted server configurations
- With flexible configurations for an optimal price-performance AI inference solution
Technical Specifications
| System Config | 单机最高16张曦思N300 |
|---|---|
| 高配16卡、典配8卡 / up to 16x N300 | |
| 16-GPU high config, 8-GPU standard | |
| Topology | PCIe Switch全互连 / Common等多种拓扑 / PCIe Switch full-mesh & Common topologies |
| Chassis | 标准机架式服务器 / standard rackmount server |
| CPU | 支持国内外主流CPU计算平台 |
| Accelerator | 曦思N300（最大500W/卡 |
| 显存容量未公开） | |
Typical Use Cases
Large Language ModelsHigh-concurrency inferenceData 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。; 整机功耗、内存、网卡/RoCE、N300显存容量官网未公开；无公开整机Datasheet PDF。
