Hsinchu, Taiwan – September 22, 2025 – Edgecore Networks, a leading provider of open networking solutions, is pleased to announce the launch of the AGS8600, a next-generation GPU-based AI server engineered for large-scale AI/ML training and high-performance computing (HPC) workloads. The AGS8600 is designed with eight AMD Instinct™ MI325X GPUs and dual AMD EPYC™ 9005 processors, delivering outstanding performance, scalability, and efficiency for modern data center environments. 

As enterprises, cloud providers, and research institutions accelerate adoption of generative AI, large language model (LLM) training, and HPC applications, the AGS8600 meets the demand for higher compute density, efficient interconnects, and optimized software frameworks. Its advanced Infinity Fabric™ interconnects, high-speed I/O, and enterprise-grade storage provide a scalable platform that empowers customers to tackle the most demanding AI workloads. 

Key Features of the AGS8600: 

• Unmatched GPU Power: Eight AMD Instinct™ MI325X GPUs with CDNA™ 3 architecture and 256 GB HBM3 memory per GPU deliver up to 2.6 PFLOPS FP8 performance for AI model training. 

• Scalable CPU Performance: Dual AMD EPYC™ 9005/Turin processors, up to 64 cores per CPU, provide robust compute capabilities for AI and HPC workloads. 

• High-Bandwidth Interconnect: Seven Infinity Fabric™ links per GPU enable full GPU-to-GPU connectivity, complemented by PCIe Gen5 interfaces for scale-out expansion. 

• Flexible I/O & Networking: Supports up to 8 x 400G or 200G Ethernet NICs, ensuring seamless integration into high-speed data center networks. 

• Enterprise-Grade Storage: High-capacity NVMe SSDs with front-end and backend expansion options for fast, high-throughput data access. 

• Reliable System Design: A 8U chassis with redundant high-efficiency PSUs and hot-swappable fans ensures stable operation under heavy workloads. 

• Open Software Ecosystem: Pre-validated with AMD ROCm™, RCCL, PyTorch, TensorFlow, JAX, and ONNX Runtime for ready-to-deploy AI and HPC applications. 

The AGS8600 empowers enterprises, research institutions, and service providers to accelerate innovation in generative AI, automation, and HPC applications with superior performance, reliability, and scalability.  

The AGS8600 is now available for deployment. To learn more, please visit https://www.edge-core.com/product/ags8600/ или свяжитесь [email protected].