With the rapid development of data centers and artificial intelligence, the need for high-speed, low-latency, and low-power data transmission has become critical. FS offers a wide range of Infiniband cables and optical transceiver products tailored for various high-performance computing scenarios. The purpose of this article is to provide an overview of the capabilities and application environments of Feminium (FS) 200G Infiniband HDR optical cables, high-speed cables, and optical modules, and to provide insight into the latest advancements.
The Infiniband 200G QSFP56 fiber optic cable seamlessly integrates fiber optics and electronic chip technologies to provide high-speed transmission, long-distance transmission capabilities, low power consumption, strong anti-interference capabilities, and flexible installation. For a wide range of applications in datacom, data centers, and high-performance computing, F&S offers two types of Infiniband HDR fiber optic cables: direct and breakout. These fiber optic cables ensure high transmission reliability through their low power consumption, low electromagnetic interference, and effective heat dissipation.
The NVIDIA Infiniband QSFP56 200G optical cable is an active optical cable based on QSFP56 VCSEL that supports longer transmission distances than passive copper cables in data centers. It is available in lengths of 1m, 2m, 3m, 5m, 10m, 15m, 20m and 30m, 50m, 100m and is designed for 200GB S Infiniband HDR systems. These cables are hot-swappable for easy installation and replacement, providing higher port bandwidth, density, and configurability at a lower cost. However, due to the presence of modules at both ends, they are not suitable for inter-cabinet cabling and are usually used in scenarios with distances of less than 30m.
The NVIDIA Infiniband 200G QSFP56 to 2x100G QSFP56 breakout cable is an affordable 200GB s HDR to 2x100GB s HDR100 active breakout cable based on the QSFP56 VCSEL. Available in lengths from 1m to 30m, it is suitable for interconnects between one 200G port and two 100G ports, providing an efficient connection for short-distance interconnects. This setup enables higher port bandwidth, density, and configurability at a lower cost, reducing the power requirements of the data center.
NVIDIA Infiniband 2x200G QSFP56 to 2X200G QSFP56 breakout cables are based on QSFP56 VCSELs and range in lengths from 3m to 30m. This breakout cable complies with SFF-8665, SFF-8636, and Infiniband HDR standards. It provides connectivity to two QSFP56 200G ports at one end and can connect to two QSFP56 200G ports at the other end for Tor leaf switches and spine switches with fat-tree ridge redundancy.
The diagram illustrates a high-bandwidth, low-latency, and highly reliable Infiniband connectivity scheme. Breakout cables connect leaf switches to GPU servers equipped with HCA cards and storage devices, while direct-attach cables connect to the backbone switches at the same rate.
The Infiniband 200G QSFP56 high-speed cable from FS provides a cost-effective solution for creating high-speed, low-latency 200G HDR connections in Infiniband switching networks. Feminine (FS) is available in different length options to meet the distance requirements of NVIDIA GPU-accelerated AI end-to-end systems.
The Passive NVIDIA Infiniband QSFP56 200G High-Speed Cable is a direct-attach copper cable that complies with the 200GB S Infiniband HDR standard. The copper cable consists of eight pairs of high-speed copper wires, each capable of transmitting data at rates up to 50GB s and ranging in length from 0Ranging from 5m to 2m.
The NVIDIA Infiniband QSFP56 200G to QSFP56 2X100G high-speed cable is a breakout cable designed based on the 200GB S Infiniband HDR standard. Unlike the 200G HDR QSFP56 to QSFP56 high-speed cables, this high-speed cable connects to a 200G QSFP56 port on one end and two 100G QSFP56 ports on the other. This configuration facilitates high-speed data transmission, bridging the gap from 200G to 100G. It offers 05m、1m、1.The cable length of 5m and 2m can be flexibly adapted to various deployment scenarios.
Sharing Functionality: Each QSFP56 port includes an EEPROM that provides host-readable product information.
Efficient short-haul interconnects for higher port bandwidth, density, and configurability.
Low latency, high bandwidth, low power consumption, and superior reliability.
Applications in data centers, artificial intelligence, and other fields, especially where high-speed data transmission is required.
The diagram shows the use of Turbo (FS) HDR Infiniband high-speed cables in a short-distance connection between a Tor switch and a switch server.
The NVIDIA Infiniband MMA1T00-HS Compatible QSFP56 200G SR4 Optical Transceiver features a QSFP56 form factor and MPO MTP-12 UPC connector designed for short-range applications in 200GB S Infiniband environments. It uses a 4-channel 53GB s PAM4 850nm VCSEL transmitter and PIN receiver with a built-in CDR chip for transmit and receive channels, and the power consumption is less than 5W. The transceiver can transmit distances of up to 100 m over OM4 multimode fiber optic cables with FEC enabled.
The NVIDIA Infiniband MMS1W50-HM Compatible QSFP56 200G FR4 Optical Transceiver is designed for mid- to long-range applications in 200GB S HDR Infiniband in a duplex LC connector and QSFP56 package. It uses a non-co-dipolarized EML CWDM transmitter to multiplex into a single optical data stream using wavelength division multiplexing technology, and uses an optical demultiplexer to demultiplex it into four optical data streams at the receiving end. The transceiver can transmit up to 2km on 1310nm OS2 single-mode fiber, effectively reducing the cost of optical fiber cabling for customers.
Considering the different features and functions of each module, the specific application of these transceivers depends on the requirements of the current environment. The 200G HDR SR4 optical module is suitable for interconnection between Infiniband HDR devices, such as switch-to-NIC or switch-to-switch connections, with transmission distances ranging from 30m to 100m.
The 200GBASE HDR FR4 optical module is designed for network environments with transmission distances ranging from 500m to 2km. It is ideal for Infiniband medium to long-range systems that extend Infiniband to remote infrastructure and the edge. This allows for local and encrypted infiniband connections between remote data centers or between data centers and remote storage infrastructure, facilitating high availability.
In summary, FemtoSpeed (FS) HDR Infiniband optical cables, high-speed cables, and optical modules have low latency, high bandwidth, low power consumption, and high reliability. They have been widely used in data centers, artificial intelligence, and various fields, and have won unanimous praise from customers.
Q1: Can Infiniband cables be used with non-Mellanox switches?
FS offers Infiniband cables that are compatible with devices other than Mellanox switches, as long as they are reasonably compatible with the switch, e.g. H3C and HW are optional.
Q2: Can I use an Infiniband cable on an Ethernet switch?
The Infiniband cable provided by FS is compatible with Infiniband switches and Ethernet switches, while the Ethernet cable is only compatible with Ethernet switches.
Q3: What is the latency and bandwidth of an HDR Infiniband cable?
The Infiniband cable provided by Flying Speed (FS) has a latency of 127 s with a bandwidth of 200 Gbps.
Q4: What lasers are used in 200G optical modules?
There are two types of lasers: direct modulation (DML) and external modulation (EML). Direct modulation lasers include VCSELs, FPs, DFBs, and are primarily used in close-range and low-power scenarios. EML lasers consist of a CW laser and an EAM modulator and are used for high-speed or long-range scenarios.
Q5: How many modulation methods are there for 200G optical transceivers?
There are two modulation methods: PAM4 modulation and NRZ modulation. The 200G QSFP56 optical module uses 4x50G PAM4 modulation, and the 200G QSFP-DD optical module uses 8x25G NRZ modulation.