GPU benchmark
& health test.

Choose a GCAT configuration that matches your NVIDIA GPU, then run the generated command on a supported Linux machine. The runner checks prerequisites before you approve the test. It runs hardware diagnostics, sustained-load tests and performance measurements, then cleans up its workloads.

A GPU benchmark measures performance, such as compute throughput and memory bandwidth. A GPU stress test applies sustained load to check behavior under demanding workloads. Hardware diagnostics check for faults. GCAT combines these checks in a reviewed plan for your hardware configuration.

GCAT checks supported GPU hardware and measures compute performance, memory bandwidth and PCIe bandwidth. Multi-GPU and cluster profiles also check supported collective and network performance. Each configuration has its own reviewed tests and thresholds.

GCAT has reviewed profiles for NVIDIA A10, H100, H200, B200 and B300 configurations. Available tests depend on the current catalogue. The exact GPU model, form factor, number of GPUs and cluster size must match an available profile. Check the configuration selector before starting.

Consumer GeForce RTX GPUs are not currently supported. GCAT runs on supported NVIDIA server hardware, rather than measuring browser graphics performance. Your browser displays progress; the Linux runner performs the GPU tests on hardware you control.

Select a cluster configuration and paste one SSH alias, hostname, IP address or user@host per line. Run the generated command from a Linux controller with passwordless SSH access and verified host keys for every node. The runner coordinates testing across the machines; you do not need to run a command on each node.

No. Run the test yourself, locally or over SSH from your controller to your own GPU servers. Your SSH keys stay on your machine, and Liquid Compute does not need inbound SSH access to your servers.

No. You can choose your hardware configuration, generate the setup command and follow testing progress without logging in or filling in a contact form.

Compare measurements from matching GPU models and configurations. Compute throughput, memory bandwidth, PCIe bandwidth and cluster communication measure different parts of the system. Your GCAT report shows each check, its measurements and the reviewed profile threshold where applicable; it does not reduce every result to a single score.

Duration depends on the profile. Full diagnostics may take 90 minutes or longer. The selected configuration shows its maximum duration, and the test shows actual progress. Keep the controller running until execution and cleanup finish.

The GPUs must be available exclusively for testing. GCAT produces load and needs the required administrator privileges. Preflight checks compatibility before testing; the runner does not automatically upgrade your GPU driver.

The runner performs cleanup of its own workloads and temporary files. It does not enroll machines, install an SSH certificate authority or create a persistent connection to Liquid Compute. If cleanup cannot be verified, it gives you instructions to investigate.

The test continues on your controller. Return in the same browser to follow progress. Closing the terminal or stopping the controller can interrupt testing; the runner enforces local deadlines and attempts cleanup.