LightBoat 2300 series FPGA Card

Based on AMD UltraScale+™ structure, the product can be applied in network acceleration in data centres, AI inference and industrial marginal calculations. It excels at having a high throughput, low power dissipation and being flexible to expand.

LightBoat 2300 series FPGA Card is built based on AMD UltraScale+™ structure with integrated VU23P chip, provides 2.252 million arithmetic logic units, 1,320 DSP slices and 99MB UltraRAM while supporting computing power of 4.1 TOPS. Each QSFP28 port supports up to 100Gbps, equals to the power and throughput of two 100Gbps adapters in a single PCIe 4.0 x16 slot. With 34 paths of GTY transmitter-receiver sets (32.75GB/s) and 4 paths of GTM pathways (58GB/s), low latency communications are ensured while being compatible with industrial environments. Efficiency and stable capacity are determined by 75W PCIe power supply plus an outer design of 120W power supply. The product possesses major advantages such as high throughput, being flexibly expandable and low power dissipation, applicable in network acceleration in data centres, AI inference and industrial marginal calculations. 

LightBoat 2300 series FPGA Card is specially designed for intelligent networking and acceleration in data centres, carrying 58 high speed transmitter-receiver sets, supporting 200G SmartNIC, meanwhile, significantly reducing the loading capacity of CPU and optimising TCO. Regarding the characteristics of the product, three main advantages stand out. First, it's based on 16nm UltraScale+™ design and 3D piling technology, achieving super high logic density and balanced efficiency, hence an overall high efficiency structural design. Secondly, it's flexible for expansion, being compatible with marginal calculations and cloud calculations evolving protocols. Thirdly, full coverage rate. It empowers processing work of 5G stations, acceleration of AI inference, high frequency financial transactions and other scenarios. Also, the product supports a heterogeneous collaboration framework between Vitis developing platform and Arm processor, and provides a full flow acceleration solution from hardwares to algorithms, assisting in building intelligent infrastructures.

Several aspects of values are embodied in the said product. 

High efficiency calculations: supports multiple high-speed ports such as dual-path 100G Ethernet, PCIe 4.0 etc., achieving large scale parallel calculations

Low latency design: suitable for applications that require super high instantaneity, e.g. data analysis, data packet monitoring, high frequency transactions etc.

Flexible and programmable: users can customise FPGA logic according to their needs and preferences; able to quickly adapt to different applications and scenarios

High energy efficiency ratio (EER): has a significant rise in EER compared with traditional CPU and GPU

Convenient: able to be deployed and run on both cloud and local, easy to install