One Stop Systems (OSS) has introduced a portfolio of next-generation PCI Express (PCIe) 6.0 CopprLink cable adapters and the 4UPro-Max PCIe expansion accelerator. OSS claims these offerings deliver ultra-low latency and high power-handling capabilities tailored for artificial intelligence (AI) and machine learning (ML) workloads in commercial data centers and edge computing applications. OSS is showcasing these new products at SuperComputing Conference 2025 (SC25) in St. Louis.
For PCIe 6.0 connectivity, the OSS HIB6110 cable adapter uses an ultra-low-latency, retimer-based design. It enables servers to connect x16 PCIe slots to industry-standard x16 PCIe CopprLink external cables, supporting up to 256 GB/s throughput to high-speed external devices. OSS says this adapter is compatible with PCIe expansion accelerators as well as current-generation Compute Express Link (CXL) memory expansion and pooling systems, addressing bandwidth, latency, and memory requirements for large AI workloads.
OSS also introduced the HIB6310, a PCIe 6.0 cable adapter with a multi-port, switch-based architecture. The HIB6310 provides up to 1,024 GB/s of PCIe switched-fabric bandwidth, supporting both external (16 lanes) and internal (32 lanes) PCIe 6.0 connectivity from a single x16 slot. This is designed to enhance Remote Direct Memory Access (RDMA) workflows by allowing external sensor data to bypass host system memory, connecting directly to GPU, FPGA, and NVMe storage in OSS Short Depth Servers. The HIB6310 further benefits from OSS proprietary multicast software and advanced server PCIe fabric topologies for complex, high-bandwidth data distribution.
The new OSS 4UPro-Max PCIe expansion system is engineered to scale server platforms by supporting up to eight 675 W, dual-width PCIe accelerators per chassis. OSS states the 4UPro-Max is built for the increasing power demands seen in modern AI accelerators—often exceeding 600 W—by offering high-density connectivity, integrated cooling, and comprehensive management functionality. The system incorporates OSS’s Unified Baseboard Management Controller (U-BMC), combining hardware telemetry, configuration management, and a Redfish API for orchestration. It supports dynamic PCIe switch resource allocation and unified monitoring across multi-rack deployments.
OSS highlights that its products are made in the US and target commercial data center, edge computing, and ruggedized environments. Additional applications include high-performance computing, advanced sensor processing, and AI/ML training and inference.
Source: One Stop Systems
