Huawei Atlas · Compute Server
Atlas 650E
The Atlas 650E server is powered by the Ascend 950DT, with 8 NPUs per node delivering up to 12.4 PFLOPS @mxFP4 and up to 4.0 TB/s read/write bandwidth, designed for standard air-cooled data centers. It supports dual-node 16-NPU full-mesh Lingqu interconnect for efficient, reliable AI compute in finance, government and healthcare.
Key Highlights
- Low-precision inference: industry-leading 12.4 PFLOPS @mxFP4 with HiF8/mxFP8/mxFP4. Ultra-high memory speed: industry-leading 4.0 TB/s HBM bandwidth, 768GB HBM per system. Ultra-high interconnect: up to 13.4 TB/s system NPU interconnect (dual-node), dual-node 16-NPU full-mesh over Lingqu.
Technical Specifications
| Form Factor | 14U rack server |
|---|---|
| NPU | 8x Ascend 950DT |
| CPU | 2x Kunpeng 950 |
| AI Compute | 12.48 PFLOPS mxFP4, 6.43 PFLOPS mxFP8/FP8/HiF8, 3.40 PFLOPS FP16/BF16 |
| On-chip Memory | 8x96GB HBM up to 4.0TB/s (768GB total) |
| System Memory | 24x DIMM slots up to 6400MT/s |
| Storage | 8x2.5 NVMe + 2x2.5 SATA (RAID) |
| Lingqu Interconnect | 8x784GB/s bidirectional per node, dual-node 16-NPU full-mesh up to 13.4TB/s |
| Network | UBoE/RoCE 400Gbps per NPU |
| PCIe | up to 5x PCIe 5.0 |
| PSU | 6x3.0kW hot-swap, 5+1 redundant |
| Cooling | air-cooled |
| Max Power | 14.5kW |
| Dimensions | 618.8x447x920mm |
Typical Use Cases
AI agents, AI coding, office assistance, content moderation, financial risk control, intelligent customer serviceAI in finance, government and healthcare.
英文站无对应英文产品页，英文文案为翻译稿；img1: 单机正面视角官方渲染图；datasheet为给定官方彩页落地页（已HTTP 200校验）；双机16 NPU Full-mesh灵衢互联为官网原文。
Related Products
Huawei Atlas
Atlas 200I DK A2 Developer Kit
Easy-to-use AI development kit for fast innovation View details →
Huawei Atlas
Atlas 200I A2 Acceleration Module
Edge AI accelerator module embedded in smart devices, robots and drones View details →
Huawei Atlas
Atlas 300I A2 Inference Card
High-performance AI inference card integrated into servers View details →
Huawei Atlas