filmov
tv
ACCL: an open-source FPGA accelerated communication library for scale-out applications

Показать описание
The XACC Tech Talks are a series of virtual talks covering a broad range of topics related to Xilinx Adaptive Compute.
ACCL: an open-source FPGA accelerated communication library for scale-out applications
Dr. Lucian Petrica, AMD AECG Research Labs
This talk will present ACCL, an open-source FPGA-accelerated communication collectives library designed to enable scale-out of applications running primarily on AMD FPGAs. Compared to previous collective communication solutions for FPGA, ACCL is flexible and extensible, easily portable, and fast. We evaluate ACCL on 8 Alveo nodes at XACC ETHZ and demonstrate that ACCL outperforms OpenMPI over 100 Gbps TCP-IP for large messages.
ACCL: an open-source FPGA accelerated communication library for scale-out applications
Dr. Lucian Petrica, AMD AECG Research Labs
This talk will present ACCL, an open-source FPGA-accelerated communication collectives library designed to enable scale-out of applications running primarily on AMD FPGAs. Compared to previous collective communication solutions for FPGA, ACCL is flexible and extensible, easily portable, and fast. We evaluate ACCL on 8 Alveo nodes at XACC ETHZ and demonstrate that ACCL outperforms OpenMPI over 100 Gbps TCP-IP for large messages.