NVIDIA MCX653105A-HDAT ConnectX-6 200Gb/s IB/Ethernet Adapter

The NVIDIA ConnectX-6 MCX653105A-HDAT is a single-port QSFP56 smart adapter delivering up to 200Gb/s InfiniBand and Ethernet connectivity. Featuring In-Network Computing acceleration, block-level hardware encryption, and support for both 50G PAM4 and 25G NRZ SerDes, it is engineered for HPC, AI, ML, and hyperscale cloud infrastructures that demand extremely low latency and high message rates.

• Ports: 1x QSFP56
• Max Speed: 200Gb/s
• Host Interface: PCIe Gen 3.0/4.0 x16
• Form Factor: PCIe stand-up

The NVIDIA ConnectX-6 MCX653105A-HDAT is a key element of the NVIDIA Quantum InfiniBand platform. It provides a single port of 200Gb/s InfiniBand and Ethernet connectivity with extremely low latency, high message rate, smart offloads, and NVIDIA In-Network Computing acceleration. The adapter supports both 50G SerDes (PAM4) and 25G SerDes (NRZ) based ports and is compliant with IBTA Specification 1.3.

Complex workloads—high-resolution simulations, extreme-size datasets, and highly parallelized algorithms—demand ultra-fast processing. ConnectX-6 addresses these requirements by offloading computation to the network, saving CPU cycles and increasing network efficiency. Its RDMA technology, end-to-end packet-level flow control, and In-Network Memory capabilities deliver the performance leap required by modern data centers.

• Up to 200Gb/s connectivity on a single QSFP56 port
• Up to 215 million messages/sec with extremely low latency
• Block-level XTS-AES hardware encryption, FIPS compliant
• In-Network Compute acceleration engines and In-Network Memory support
• Best-in-class packet pacing with sub-nanosecond accuracy
• PCIe Gen 3.0 and Gen 4.0 support, x16 host interface
• Supports 50G SerDes (PAM4) and 25G SerDes (NRZ) based ports
• RoHS compliant and ODCC compatible

ConnectX-6 offloads collective and vector collective operations to the network, reducing CPU utilization and accelerating MPI and AI workloads. It supports NVIDIA PeerDirect (GPUDirect) RDMA communication acceleration, enabling direct GPU-to-GPU data paths.

The adapter implements RDMA with send/receive semantics, hardware-based reliable transport, extended reliable connected transport (XRC), dynamically connected transport (DCT), and on-demand paging (ODP). It supports out-of-order RDMA with adaptive routing and MPI tag matching, along with rendezvous protocol offload and burst buffer offload.

NVMe over Fabrics (NVMe-oF) target and initiator offloads optimize remote NVMe storage access, enhancing CPU utilization and scalability. The adapter also supports SRP, iSER, NFS RDMA, and SMB Direct, with T10-DIF signature handover at wire speed.

Block-level encryption (XTS-AES 256/512-bit key) is hardware-offloaded from the CPU, saving latency and CPU utilization. Dedicated encryption keys protect users sharing the same resources. FIPS compliance is available.

• High performance computing (HPC) clusters requiring low-latency, high-message-rate fabrics
• Machine learning and deep learning training platforms
• Big data analytics and hyperscale cloud infrastructures
• NVMe-oF storage environments and disaggregated storage architectures
• Virtualized networks and network function virtualization (NFV)
• Host chaining for economical rack design
• Smart interconnect for x86, Power, Arm, GPU, and FPGA-based platforms

• InfiniBand switches: interoperability up to 200Gb/s (4 lanes of 50Gb/s data rate)
• Ethernet switches: interoperability up to 200Gb/s (4 lanes of 50Gb/s data rate)
• Cables: passive copper cable with ESD protection; powered connectors for optical and active cable support
• PCIe: Gen 4.0, 3.0, 2.0, 1.1 compatible; 16-lane PCIe interface; support for x1, x2, x4, x8, x16 configurations
• Operating systems: RHEL, SLES, Ubuntu, other major Linux distributions, Windows, FreeBSD, VMware, OFED, WinOF-2
• Ethernet standards: IEEE 802.3bj/bm 100GbE, 802.3by 25/50GbE, 802.3ba 40GbE, 802.3ae 10GbE, 802.3az EEE, and more

The MCX653105A-HDAT uses a single QSFP56 cage. It is a PCIe stand-up card with a PCIe Gen 3.0/4.0 x16 host interface.

Specifications are based on NVIDIA product documentation. For parameters not listed here, such as exact power consumption and physical dimensions beyond the publicly specified 167.65mm x 68.90mm (without brackets), please confirm before ordering.

The MCX653105A-HDAT is the single-port 200Gb/s stand-up PCIe adapter with a Gen 3.0/4.0 x16 host interface. For dual-port configurations, refer to the MCX653106A-HDAT. For OCP 3.0 form factors, refer to the MCX653435A-HDAT. For cold plate liquid-cooling variants, contact your sales representative.

• In-Network Computing: Offloads collective and vector operations to the network, freeing CPU cycles and boosting application efficiency.
• High message rate and low latency: Up to 215 million messages/sec with sub-nanosecond packet pacing supports demanding HPC and AI workloads.
• Security built in: Hardware block-level encryption and FIPS compliance protect data in transit without sacrificing performance.
• Storage optimization: NVMe-oF target and initiator offloads improve CPU utilization and scalability for disaggregated storage.
• Flexible deployment: Supports InfiniBand and Ethernet, 50G PAM4 and 25G NRZ SerDes, and a broad range of operating systems.
• Virtualization-ready: SR-IOV, ASAP2, and OVS offloads deliver efficient virtualized network performance.

Hong Kong Starsurge Group Co., Limited provides global delivery, multilingual support, and tailored network infrastructure solutions. Our experienced sales and technical team supports industries including government, healthcare, manufacturing, education, finance, and enterprise. We offer:

• Pre-sales technical consultation and compatibility verification for ConnectX-6 deployments
• Custom software development and network management systems
• IoT solutions and system integration
• Responsive after-sales support and RMA assistance

Contact us for volume pricing, lead times, and certification documents for this product.

• ESD handling: Observe standard ESD precautions during installation and removal. Use grounded wrist straps and ESD-safe work surfaces.
• Thermal: Ensure adequate airflow in the server chassis. The adapter requires proper cooling to maintain performance and reliability.
• Power: Verify that the server provides sufficient power and PCIe slot bandwidth for the adapter's x16 host interface.
• Firmware and drivers: Install the latest NVIDIA OFED or WinOF-2 drivers and firmware for optimal performance and compatibility.
• Compatibility: Confirm that the server's PCIe slot, bracket type (tall vs. short), and cooling configuration match the adapter variant before installation.

• Confirm required port count and speed (single-port 200Gb/s for MCX653105A-HDAT).
• Verify PCIe slot availability and generation (Gen 3.0 or Gen 4.0 x16).
• Check bracket type (tall vs. short) and server chassis compatibility.
• Confirm cable type and reach (passive copper vs. optical/active).
• Ensure adequate cooling and airflow in the server chassis.
• Request firmware/driver package (OFED or WinOF-2) and certification documents if required.
• Confirm quantity, lead time, and warranty terms with supplier.

• MCX653106A-HDAT — dual-port 200Gb/s, PCIe Gen 3.0/4.0 x16
• MCX653105A-ECAT — single-port 100Gb/s, PCIe Gen 3.0/4.0 x16
• MCX653106A-ECAT — dual-port 100Gb/s, PCIe Gen 3.0/4.0 x16
• MCX653435A-HDAT — single-port 100Gb/s, OCP 3.0 small form factor
• MCX654105A-HCAT — single-port 200Gb/s, Socket Direct

• NVIDIA ConnectX-6 InfiniBand Adapters Product Brief
• NVIDIA OFED and WinOF-2 Driver Installation Guide
• NVIDIA In-Network Computing and SHARP Technology Overview
• InfiniBand Trade Association (IBTA) Specification 1.3
• Data Center ESD Handling Best Practices

• Model: MCX653105A-HDAT
• Brand: NVIDIA
• Ports: 1x QSFP56
• Max Speed: 200Gb/s
• Host Interface: PCIe Gen 3.0/4.0 x16
• Form Factor: PCIe stand-up
• Protocols: InfiniBand & Ethernet
• Encryption: XTS-AES, FIPS

Hong Kong Starsurge Group Co., Limited is a technology-driven provider of network hardware, IT services, and system integration solutions. Founded in 2008, the company serves customers worldwide with products including network switches, NICs, wireless access points, controllers, cables, and related networking equipment. Backed by an experienced sales and technical team, Starsurge supports industries such as government, healthcare, manufacturing, education, finance, and enterprise.

The company also offers IoT solutions, network management systems, custom software development, multilingual support, and global delivery. With a customer-first approach, Starsurge focuses on reliable quality, responsive service, and tailored solutions that help clients build efficient, scalable, and dependable network infrastructure.