NADDOD High-Performance Computing (HPC) InfiniBand Network Solution
Based on the high floating-point computing performance of GPU and the high bandwidth and low latency of InfiniBand, NADDOD provides customers with high-performance computing (HPC) solutions based on InfiniBand network and GPU computing nodes, providing powerful performance support for many data-intensive applications such as high-performance computing, artificial intelligence (AI), cloud computing, storage system, etc.
NADDOD Typical HPC Network and Computing Solution
NADDOD high-performance computing solution typically consists of high-speed InfiniBand network and high-performance GPU server. InfiniBand HDR switch and active optical cables (AOC) form a high-speed non-congestion network in LEAF-SPINE architecture. The high-performance GPU servers are equipped with GPU computing platform, which is composed of eight data center A100 scientific data computing accelerator card and InfiniBand HDR network cards.
The solution builds a high-bandwidth and low-latency data transmission network between GPU computing servers and storage systems to meet the performance requirements of data-intensive application systems to the greatest extent, and solve the problems of long AI model iteration cycles and high resource consumption in the deep learning training process in various industries.
NADDOD HPC Solution Advantages
Equipped with InfiniBand HDR 200G network interface, the InfiniBand networking interconnect speed of NADDOD HPC solution is up to 200Gb/s for Server to Server and Server to Storage, accelerating the data transfer of send and receive services.
In this NADDOD HPC solution, the local subnet is configured through the subnet manager and continuous operation is ensured.
Stable and Reliable
Based on InfiniBand SHIELD technology, microsecond-level recovery of network faults is realized.
|Product Name||Product Type||Product Specification|
|MQM8790-HS2F||HDR InfiniBand Switch||HDR InfiniBand switch, 40G QSFP56 ports, 2 power supplies (AC), x86 dual core, standard depth, C2P airflow, rail kit|
|Q56-200G-A10H||NADDOD HPC IB HDR AOC Cable||200G QSFP56 to QSFP56 AOC, 4x50Gb/s PAM4, Mellanox compatible InfiniBand HDR|
|Q2Q56-200G-A5H||NADDOD HPC IB HDR AOC Splitter Cable||200G HDR to 2x100G HDR100 QSFP56 to 2xQSFP56 AOC, 4x50Gb/s PAM4, Mellanox compatible InfiniBand HDR|
|MCX653106A-ECAT-SP||NADDOD HDR IB NIC||Dual QSFP56 InfiniBand and Ethernet (copper and optical), dual ports for HDR100 EDR InfiniBand and 100GbE Ethernet connectivity|
|ND4110W||NADDOD GPU Server||MAX support dual 3rd generation Intel Xeon scalable processors, 10 double-width PCI-E 4.0 x16 card (full height full length)|
|Data Center A100 GPU||DC Scientific Computing Accelerator||TensorFloat 32 156 TFLOPs, FP64 Tensor Core 19.5 TFLOPS, GPU memory 80Gb HBM2e|
Customized Network Design Flexibly Increases Service Performance
Combined with the customized high-speed network built by InfiniBand network card and InfiniBand switch, while fully releasing the performance, NADDOD HPC solution enables the system to dynamically increase the number of nodes according to business needs, and realize the linear growth of storage capacity and storage performance.
Professional and Efficient Technical Support
NADDOD senior HPC solution experts can provide users with 7x24 hour technical support services. From hardware solution design to implementation, and to software configurations and services. We will provide professional full-stack solutions.
Low Latency and Fast Output, Saving Costs
The high-bandwidth and low-latency data transmission network built between GPU computing servers and storage systems through InfiniBand can quickly increase the model iteration speed, shorten the result output cycle as much as possible, and directly reduce the waste of storage space.