-
ReturnGPU ServerGeneral Purpose ServerData Center InfrastructureNetwork
Specifically designed for AIGC, integrates 128 QSFP112 ports, accelerate AI network performance by 1.6X over traditional Ethernet, features an independent management plane, supports the third-party network operating system, and can be deployed as an AIGC network switch and data center switch.
• 102.4 Tbps switching capacity, End-to-End 400G lossless data transmission.
• High-density ports with 128 QSFP112, elastic scalability supporting AI compute clusters with hundreds of thousands of GPUs.
• Supports adaptive routing, fine-grained load balancing, achieving 95% efficient bandwidth, significantly enhancing LLM training efficiency.
• Supports advanced congestion control algorithm such as RTTCC, preventing network congestion and reducing Job Completion Time (JCT) by 30%.
• Enables multi-tenant, multi-job scenarios, achieving performance isolation for different AI workloads, providing consistent and predictable performance.
• Open hardware platform, adhering to the S3IP-UNP standard, supporting UXOS, compatible with SONiC and other third-party network operating systems.
• Based on the OpenBMC open-source system, an independent management plane is constructed.
