1、White PaperAI NetworkingUpdated July White PaperEmergence of Artificial Intelligence(AI)Artificial Intelligence(AI)has emerged as a revolutionary technology that is transforming many industries and aspects of our daily lives from medicine to financial services and entertainment.The rapid evolution o

2、f real-time gaming,virtual reality,generative AI and metaverse applications are changing the ways in which network,compute,memory,storage and interconnect I/O interact.As AI continues to advance at unprecedented pace networks need to adapt to the colossal growth in traffic transiting hundreds and th

3、ousands of processors with trillions of transactions and gigabits of throughput.As AI quickly moves out of labs and research projects toward mainstream adoption it demands increases in network and computing resources.A common characteristic of these AI workloads is that they are both data and comput

4、e-intensive.A typical AI training workload involves billions of parameters and a large sparse matrix computation distributed across hundreds or thousands of processors CPUs,XPUs or TPUs.These processors compute intensively and then exchange data with their peers.Data from the peers is reduced or mer

5、ged with the local data and then another cycle of processing begins.In this compute-exchange-reduce cycle,approximately 20-50%of the job time is spent communicating across the network so bottlenecks have a substantial impact on job completion time.Networking for AIEthernet has come a long way since

6、its invention by Bob Metcalfe,first introduced as a memo in 1973 and commercialized in 1980.Since then,the technology has evolved multiple times and has been extended all over the world.Its grown in speeds from the initial 10 Mbps to now 800 Gbps per port,with 1.6 Tbps on the horizon,and has evolved