Specially built for ultra-low latency electronic exchanges, AMD launches FPGA acceleration card Alveo UL3524

Thanks to CTOnews.com netizen OC_Formula for the clue delivery! CTOnews.com, Oct. 2, users who have been exposed to stock speculation may understand that a delay of less than half a second can determine the profit or loss of a trade, and that using faster or more sophisticated algorithms is important for electronic exchanges.

AMD last week launched an Alveo UL3524 accelerator card, a new accelerator card designed for ultra-low latency electronic exchanges, which claims to reduce operation latency to 1/7 and run more complex machine learning algorithms on custom chips.

CTOnews.com query found that the card is based on Xilinx's 16nm Virtex UltraScale+ FPGA design, with 64 transceivers, 780000 lookup tables and 1680 DSP, software support is available through the Vivado suite, and customers are free to deploy algorithms.

According to reports, Alveo UL3524 has been delivered to the industry's leading exchanges for deployment and supports a variety of partner products that can trade at a speed of less than 3 nanoseconds.

Hamid Salehi, director of product marketing at AMD, said that in ultra-low latency transactions, one nanosecond can determine the profit or loss of the transaction. The Alveo UL3524 accelerator card uses 64 ultra-low latency transceivers, 780000 FPGA lookup tables (LUT) and 1680 digital signal processors (DSP) to accelerate customized transaction algorithms in hardware.