Workload and system support

Use this matrix to choose an execution system for your workload. Names match test_template_name in TOML. A check mark means this revision registers an execution strategy for that combination; workload-specific prerequisites still apply.

Compatibility

Workload	Slurm	Standalone	Kubernetes	RunAI	LSF
Aiconfigurator	—	✅	—	—	—
AIDynamo	✅	—	✅	—	—
BashCmd	✅	—	—	—	—
ChakraReplay	✅	—	—	—	—
DDLBTest	✅	—	—	—	—
DeepEP	✅	—	—	—	—
DynamoMocker	—	✅	—	—	—
Fio	✅	✅	—	—	—
JaxToolboxGPT	✅	—	—	—	—
JaxToolboxGrok	✅	—	—	—	—
JaxToolboxNemotron	✅	—	—	—	—
MegatronBridge	✅	—	—	—	—
MegatronRun	✅	—	—	—	—
MoEBenchmark	✅	—	—	—	—
NcclTest	✅	—	✅	✅	—
NeMoLauncher	✅	—	—	—	—
NeMoRun	✅	—	—	—	—
NIXLBench	✅	—	—	—	—
NixlEP	✅	—	—	—	—
NIXLKVBench	✅	—	—	—	—
NixlPerftest	✅	—	—	—	—
OSUBench	✅	—	—	—	—
sglang	✅	—	—	—	—
Sleep	✅	✅	✅	—	✅
SlurmContainer	✅	—	—	—	—
SRTSlurm	✅	—	—	—	—
TritonInference	✅	—	—	—	—
UCCTest	✅	—	—	—	—
vllm	✅	—	—	—	—

Choose a setup path

• Run on Slurm: submit workloads to compute nodes from a login or submit host.

• Run your first experiment: run a local storage experiment with Standalone.

• Run on Kubernetes: run Dynamo or NCCL with the appropriate Kubernetes operators.

RunAI currently has an NCCL integration; use Set up RunAI for its configuration. LSF currently has a Sleep integration for scheduling checks; its fields are under LSF.