Scala Computing

Founded 2018

Simulate AI training and inference performance across hardware topologies and network scales

You can use Scala Computing to simulate AI model training and inference workloads across various network topologies, hardware configurations, and cluster scales. The software uses calibrated workload traces to project performance bottlenecks before spending money on physical hardware. You can compare different cloud capacity providers, tune your live server fleet, and test network fabrics or silicon designs before tape-out. It gives engineering teams verifiable performance evidence to configure training runs, evaluate fabric designs, and validate infrastructure performance claims without running costly physical test clusters.

What Scala Computing does

Simulate large-scale all-reduce operations across multi-GPU setupsAnalyze queue depths and ECN marking across spine switchesCompare provider network fabrics for training throughputEvaluate model training runtimes across varying cluster sizesMeasure mean hop time and packet pause events under loadVerify performance gains of acceleration software on networksSimulates AI training and inference across custom network topologiesUses calibrated workload traces to mirror actual application behaviorModels leaf-spine networks, ECMP routing, and queue depthsEvaluates GPU cluster performance before renting physical capacityTests silicon and switch fabric designs prior to tape-outProjects workload scaling limits across different cloud providers