AI Servers

FusionServer AI servers are designed for data centers and are applicable to AI, HPC, database, and video analysis scenarios

Superior Heterogeneous Computing Capability

A 4 U chassis supports a maximum of eight full-height full-length dual-slot heterogeneous accelerator cards with a maximum power consumption of 350 W or 32 half-height half-length heterogeneous accelerator cards with a maximum power consumption of 75 W.

Supports GPUDirect RDMA and Peer-to-Peer.

Flexible Heterogeneous Topology Configuration

Supports one full-width or two half-width heterogeneous computing nodes, one-click topology switching, and multiple topologies with CPU/GPU configuration ratios of 1:2, 1:4, and 1:8.

Fully Modular Design

Adopts decoupled CPU modules and heterogeneous modules to support the long-term evolution of CPUs and heterogeneous components.

The PSUs, drives, and fan modules are modular and support hot swap and redundancy.