Customization: | Available |
---|---|
Type: | Wired |
Bandwidth: | 200g |
Suppliers with verified business licenses
Audited by an independent third-party inspection agency
Basic information | Model: MCX75510AAS-HEAT | NVIDIA ConnectX-7 series |
Brand: NVIDIA/Mellanox | Support dual protocols (InfiniBand/ Ethernet) | |
hardware configuration | Interface Type: Single-Port OSFP | Backwards compatibility QSFP56/QSFP28 |
Bus Specification: PCIe 5.0 x16 | Compatible with PCIe 4.0/3.0. | |
Form Factor: Half-Height Half-Length (HHHL) | Size: 68.90mm x 167.65mm | |
Operating Temp: 0°C to 55°C | Industrial heat dissipation design | |
performance parameter | Data Rate: 200Gb/s (NDR200) | Upgradeable to 400Gb/s |
Latency: Ultra-Low Latency | Optimize HPC/AI workload | |
Protocol: InfiniBand/Ethernet | Automatic switching protocol stack | |
Features: RDMA, NVMe over Fabric Offload | Accelerate storage and compute unloading | |
Virtualization: SR-IOV | Up to 512 virtual functions | |
Network and expansion | upgrade.Interface Compatibility: QSFP112 | Support future 800G infrastructure upgrade. |
PCIe Speed: Gen 4.0/5.0 @ 16GT/s/32GT/s | Adapting to full bandwidth requirements | |
Safety and compatibility | OS Support: Linux/Windows/VMware ESXi | Driver supports RoCEv2/IBTA. |
Security: Secure Boot/Encryption Offload | Hardware-level data protection | |
Encryption: Not Supported | Rely on external security protocols (such as IPsec) | |
Certification: FCC/CE | Meet the mainstream certification standards | |
Application scenario | Application: HPC/AI Clusters, Large-Scale AI Training | Support communication optimization of 100 billion-level parameter model |
Other characteristics | Networking Compute Engine: NVIDIA Networking Compute Engine | Hardware Accelerated Network Computing Task |
Power Consumption: ≤25W (Full Load) | Need active radiator cooperation |