INFRASTRUCTURE
“ Distributed training hits a wall when the network between nodes can’ t keep up,” Colm explains.“ Elastic Fabric Adapter( EFA) is built specifically for this, delivering lower, more consistent latency and higher throughput than traditional TCP. EC2 UltraClusters take it further by packing thousands of accelerated instances into a single Availability Zone, all connected through a petabit-scale, non-blocking EFA network. These UltraServers are deployed in EC2 UltraClusters that scale to hundreds of thousands of chips, letting customers scale training from days to hours without hitting the network bottleneck that typically limits multi-node jobs.”
Spatial AI tackles this pipeline congestion from the opposite direction: reducing and filtering high-density point-cloud data directly at the sensor before it ever hits the network.
AMAZON EC2
Amazon Elastic Compute Cloud( Amazon EC2) is a cloud computing service that provides resizable, secure compute capacity as virtual servers. It offers over 1,000 instance types across Intel, AMD, and Arm processors and serves developers, enterprises, and teams running web apps, HPC, machine learning, and Windows workloads.
Cormac explains the necessity of this upstream data reduction:“ The key is to avoid moving or processing more data than you actually need. High-density point clouds contain enormous amounts of information, but not every point is relevant to every task. Across Hexagon, we’ re building a more connected ecosystem between sensors and software that allows spatial data to move more efficiently from the field into downstream workflows.
CREDIT: AWS aimagazine. com 103