In AI servers and high-performance storage systems, reliable transmission of high-speed signals is fundamental to stable system operation. The LRSV9501-2E is a PCIe 5.0 x16 dual-port MCIO Retimer expansion card launched by Linkreal Co., Ltd. (LR-LINK). Based on the BCM85657 controller, it delivers high-performance signal expansion capabilities for data centers and AI infrastructure. This article provides an in-depth analysis of the product from technical specifications, functional features, application scenarios and other dimensions.
The LRSV9501-2E(Go to purchase) features a low-profile, half-length design (170mm × 68.9mm), compatible with standard servers and compact AI workstations. It comes with both 3U and 2U brackets to fit different chassis.
Core Specifications
Item | Details |
Controller | Broadcom BCM85657 |
Host Interface | PCIe 5.0 x16, 32GT/s |
External Ports | 2×MCIO 8i Connectors |
Protocol Support | PCIe Gen5/Gen4/Gen3/Gen2/Gen1, CXL 2.0 |
Lane Bifurcation | 1×16, 2×8, 4×4 |
Power Consumption | Min 5.313W / Max 12.372W |
Operating Temperature | -10°C ~ 55°C |
The LRSV9501-2E uses the Broadcom BCM85657, a high-performance PCIe 5.0/CXL 2.0 Retimer chip. It is purpose-built to resolve signal integrity issues in high-speed transmission and is widely used in AI servers, high-end storage systems and data center interconnection.
Key Technical Features
· 16-Lane PCIe 5.0 Support: Full x16 PCIe 5.0 link at 32GT/s per lane
· CXL 2.0 Compatibility: Complete support for CXL.io, CXL.cache and CXL.memory
· Signal Regeneration: Restores amplitude and timing via CDR and retiming
· Protocol-Transparent Forwarding: Fully transparent to upper-layer protocols, no driver required
· Low-Latency Design: Optimized signal path to minimize transmission delay
The LRSV9501-2E supports flexible PCIe lane bifurcation to match downstream device requirements:
All 16 lanes operate as a single link, ideal for high-end GPUs or high-performance network adapters. Delivers maximum bandwidth for bandwidth-critical applications.
16 lanes split into two x8 links, supporting two devices simultaneously. Suitable for dual-GPU setups or GPU+NIC combinations.
16 lanes split into four x4 links, supporting four devices at once. Optimized for high-density storage expansion such as multi-port NVMe SSDs.
The LRSV9501-2E is equipped with two MCIO 8i external connectors for high-speed cable connections to external devices. MCIO (Mini Cool Edge IO) is a next-generation high-speed interconnect standard optimized for PCIe 5.0 and CXL.
Advantages of MCIO
· High Signal Integrity: Optimized pin layout and shielding for PCIe 5.0 signals
· Compact Size: Saves space vs. traditional connectors, ideal for high-density deployments
· Cable Extension: Enables external device connection beyond chassis limits
· Hot-Plug Support: Simplifies maintenance and online expansion
Via MCIO cables, the LRSV9501-2E can connect to: NVMe SSD expansion backplanes, GPU compute nodes, CXL memory expansion modules, high-speed network adapters and more.
The LRSV9501-2E fully supports CXL 2.0, a key technology for next-generation data center interconnection. Built on the PCIe 5.0 physical layer, CXL adds memory-semantic access and cache coherence.
Key CXL 2.0 Features
· Memory Pooling: Shared memory resources across multiple devices to improve utilization
· Cache Coherence: Accelerators such as GPUs access CPU cache directly, reducing copy overhead
· Switch Support: Enables complex topologies via CXL 2.0 switching
For memory-intensive AI training and HPC workloads, CXL memory expansion paired with the LRSV9501-2E’s signal enhancement enables high-performance memory pooling.
The LRSV9501-2E supports mainstream server operating systems:
Global OS: Windows, Windows Server, RHEL, CentOS, Ubuntu, Debian, SUSE, Rocky Linux, FusionOS, VMware ESXi
As a protocol-transparent device, the LRSV9501-2E is plug-and-play with no dedicated driver required, simplifying deployment and maintenance.
In AI training servers, the LRSV9501-2E expands external GPU chassis via MCIO cables, overcoming motherboard slot limits to build large-scale GPU clusters. Retimer ensures signal integrity over long distances.
In software-defined storage and all-flash arrays, the card connects external NVMe SSD backplanes. 4×4 bifurcation supports four x4 NVMe devices for high-density storage scaling.
For memory-hungry applications, the card connects CXL memory modules to add terabytes of RAM. CXL 2.0 guarantees cache-coherent memory access.
For 100G/200G/400G network adapters, the LRSV9501-2E provides full PCIe bandwidth and robust signal integrity.
· Motherboard Compatibility: Verify PCIe 5.0 and target bifurcation support
· Thermal Design: Ensure adequate chassis airflow for up to 12W power dissipation
· Cable Selection: Use standard-compliant MCIO high-speed cables
· Lane Planning: Preconfigure bifurcation based on downstream device needs
With the Broadcom BCM85657 controller, flexible lane bifurcation, CXL 2.0 support and MCIO high-speed interfaces, the LRSV9501-2E delivers powerful signal expansion for AI servers and high-performance storage. In the PCIe 5.0 era, Retimer technology is essential for reliable high-speed transmission.
For enterprises building AI infrastructure or upgrading storage systems, the LRSV9501-2E offers a high-performance, highly compatible signal expansion solution worthy of evaluation.