What Is Functional Performance Test?_
This test validates that each component—GPU compute, networking, and cooling—operates correctly and achieves the expected performance metrics, such as GPU utilization, memory bandwidth, and inter-node latency. It is typically run after rack assembly and cabling are complete, using standardized benchmarks or workload scripts. The test identifies hardware faults, configuration errors, or thermal throttling before the system enters production.
Technical Details
The test suite often includes GPU stress tests (e.g., matrix multiplication or AI inference workloads) and network throughput checks (e.g., MPI all-reduce or NCCL tests). For NVLink-based systems, the test verifies GPU-to-GPU copper spine bandwidth without relying on external fiber links typically used for scale-out networking. Results are compared against OEM-provided baselines, and any deviation beyond acceptable tolerances triggers diagnostic steps.
How Leviathan Systems Works with Functional Performance Test
In our field work, Functional Performance Tests are the final validation step before handover, executed after liquid cooling loop commissioning and structured cabling verification. We run these tests on each rack in the NVL72 cluster to confirm that all 72 GPUs and their NVLink domains perform within spec.
Related service
Network Testing & Commissioning →