XPerf Inc.

Founded 2025

Validate hardware racks, prevent cluster downtime, and schedule compute workloads for machine learning

Pricing model

Demo

You can use XPerf to test and validate AI computing racks before putting them into active production. It monitors running hardware clusters to detect issues early and trigger automatic fixes before failures interrupt model training. The system coordinates workload scheduling across NVIDIA, AMD, and ASIC accelerators to keep compute utilization high. You can generate tokens-per-second performance baselines, diagnose hardware faults down to their root causes, and run everything on-premise in air-gapped data centers without sending any private telemetry data outside your physical facility.

What XPerf Inc. does

Run pre-deployment validation tests on GPU racksDetect hardware and network faults across compute clustersApply automated self-healing fixes to failed nodesSchedule post-training and inference tasks across acceleratorsTrack tokens-per-second performance benchmarksLog cluster incident histories and root cause diagnosesValidates new compute racks and clusters before deploymentMonitors GPU, memory, and network telemetry for failure signalsExecutes automated self-healing remediation routines when faults occurSchedules inference and post-training jobs across mixed hardware fleetsGenerates tokens-per-second baseline benchmark reportsRuns entirely on-premise within air-gapped network environmentsMaintains an incident knowledge base with historical root causes