Brand: Huawei | Category: Servers
SKU: HUAW-02313UWN | Part #: 02313UWN | MPN: 02313UWN
Contact for Pricing — Request a Quote
The Huawei Atlas 800T A2 AI Training Server from Huawei is enterprise-grade Servers hardware built for data centers. Huawei's Ascend 910B-based AI training server delivering domestic high-performance compute for large-scale model development and enterprise AI workloads. Available in brand new condition. Available from Omnixon Global — contact our team to request a quote.
| Brand | Huawei |
| Category | Servers |
| SKU | HUAW-02313UWN |
| Part Number | 02313UWN |
| Condition | New |
| Product Line | Atlas 800T A2 |
| Accelerator Type | Ascend 910B |
| Accelerator Count | 8 × Ascend 910B per node |
| Accelerator Memory | 32 GB HBM2E per Ascend 910B (256 GB aggregate) |
| Peak Compute Performance | 640 TFLOPS FP32 per accelerator (5.12 PFLOPS aggregate with 8 units) |
| Tensor Operations | Mixed-precision support (FP32, FP16, BF16, INT8, INT16) |
| Interconnect Fabric | Huawei Ascend Link (proprietary high-speed accelerator interconnect) |
| Host Processor | Third-generation Xeon Scalable (or equivalent multi-socket x86 configuration) |
| Host Memory | 1 TB DDR5 system RAM (upgradeable) |
| Storage Interface | PCIe 5.0 ×16 per accelerator; NVMe M.2 SSD slots |
| Network Connectivity | 4 × 100 GbE RoCE v2 Ethernet ports (collective 400 Gbps) |
| Power Consumption | 6–8 kW typical training load per node |
| Power Supply | Redundant 12 kW+ modular PSUs (N+1 configuration) |
| Form Factor | 4U dual-socket server chassis |
| Cooling | Liquid cooling for accelerators; air-cooled host components |
| Operating System Support | EulerOS, CentOS, Ubuntu LTS with Huawei-optimized drivers |
| Software Stack | MindSpore 2.0+, HiLens, MindX SDK, TensorFlow/PyTorch compatibility layers |
| Management Interface | Intelligent Management Module (IMM) with IPMI 2.0, remote KVM, firmware updates |
| Typical Training Workloads | Batch sizes 64–4096; model sizes 1B–175B+ parameters; distributed training across 2–32 nodes |
| Multi-Node Scaling | Supports seamless distributed training via RoCE collective communications and AllReduce optimization |
The Huawei Atlas 800T A2 AI Training Server is built for enterprise data-center workloads — virtualization (VMware, Proxmox, Nutanix), private cloud, database hosting, and AI/ML training. It fits standard EIA-310 server racks and supports redundant PSUs and hot-swap drives common in production environments.
Key specifications for the Huawei Atlas 800T A2 AI Training Server: new condition; manufacturer Huawei; product line Atlas 800T A2; accelerator type Ascend 910B; accelerator count 8 × Ascend 910B per node; accelerator memory 32 GB HBM2E per Ascend 910B (256 GB aggregate); peak compute performance 640 TFLOPS FP32 per accelerator (5.12 PFLOPS aggregate with 8 units). Manufacturer part number 02313UWN. For the full datasheet with electrical, environmental, and compliance details, contact our pre-sales engineering team.