Brand: Huawei | Category: GPUs
SKU: HUAW-02313TJP | Part #: 02313TJP | MPN: 02313TJP
Contact for Pricing — Request a Quote
The Huawei Ascend 910B represents the pinnacle of indigenous Chinese AI acceleration technology, engineered for datacenter-scale machine learning workloads. Built on advanced process technology, the 910B integrates 64GB of high-bandwidth HBM2e memory supporting up to 1.2TB/s memory bandwidth, enabling efficient training of ultra-large language models and vision transformers. The processor architecture incorporates Huawei's proprietary tensor computation cores optimized for mixed-precision operations (FP32, FP16, INT8), delivering 1,024 TFLOPS peak performance in single-precision compute. Its unified memory hierarchy and on-chip interconnect enable efficient distributed training across multi-card configurations.
The 910B is positioned for enterprise AI infrastructure modernization within heavily regulated computational environments and organizations prioritizing domestic semiconductor supply chains. The accelerator integrates seamlessly with Huawei's MindSpore deep learning framework and CANN (Compute Architecture for Neural Networks) software stack, optimizing end-to-end training and inference pipelines. Its 310W thermal design power with vapor-chamber cooling enables dense datacenter deployment while maintaining thermal stability across sustained workloads.
Launched in Q4 2023, the Ascend 910B addresses critical demand for high-performance AI compute capacity amid evolving global semiconductor trade dynamics. The card supports PCIe 5.0 interconnect for host system integration and features enhanced reliability mechanisms including error-correcting codes, redundancy checking, and comprehensive health monitoring telemetry.
| Manufacturer | Huawei |
| Model | Ascend 910B |
| Product Category | AI Accelerator Card |
| Launch Quarter | Q4 2023 |
| Peak FP32 Performance | 1024 TFLOPS |
| Peak FP16 Performance | 2048 TFLOPS |
| Peak INT8 Performance | 4096 TOPS |
| Memory Type | HBM2e |
| Memory Capacity | 64GB |
| Memory Bandwidth | 1.2TB/s |
| Memory Error Correction | Full ECC support |
| Tensor Cores | Multiple specialized compute clusters with mixed-precision support |
| Interconnect Interface | PCIe 5.0 x16 |
| Multi-Card Connectivity | Huawei HiLink proprietary inter-accelerator fabric with NVSwitch-class bandwidth |
| Thermal Design Power | 310W |
| Cooling Solution | Vapor-chamber with high-performance thermal interface |
| Operating Temperature Range | 0°C to 50°C |
| Form Factor | Full-height, full-length PCIe card |
| Software Framework | MindSpore, CANN (Compute Architecture for Neural Networks) |
| Supported Precision Formats | FP32, FP16, BF16, INT8, INT16 |
| Max Power Consumption | 350W (peak transient) |
| System Requirements | x86 or ARM host with PCIe 5.0 slot and proper airflow |
Available from Omnixon Global. Submit an RFQ and our team will confirm configuration and availability for your order.
| Brand | Huawei |
| Category | GPUs |
| SKU | HUAW-02313TJP |
| Part Number | 02313TJP |
| Condition | New |
| Model | Ascend 910B |
| Product Category | AI Accelerator Card |
| Launch Quarter | Q4 2023 |
| Peak FP32 Performance | 1024 TFLOPS |
| Peak FP16 Performance | 2048 TFLOPS |
| Peak INT8 Performance | 4096 TOPS |
| Memory Type | HBM2e |
| Memory Capacity | 64GB |
| Memory Bandwidth | 1.2TB/s |
| Memory Error Correction | Full ECC support |
| Tensor Cores | Multiple specialized compute clusters with mixed-precision support |
| Interconnect Interface | PCIe 5.0 x16 |
| Multi-Card Connectivity | Huawei HiLink proprietary inter-accelerator fabric with NVSwitch-class bandwidth |
| Thermal Design Power | 310W |
| Cooling Solution | Vapor-chamber with high-performance thermal interface |
| Operating Temperature Range | 0°C to 50°C |
| Form Factor | Full-height, full-length PCIe card |
| Software Framework | MindSpore, CANN (Compute Architecture for Neural Networks) |
| Supported Precision Formats | FP32, FP16, BF16, INT8, INT16 |
| Max Power Consumption | 350W (peak transient) |
| System Requirements | x86 or ARM host with PCIe 5.0 slot and proper airflow |