4 × NVIDIA GB300

Tested system

Processor: NVIDIA Grace CPU (Arm Neoverse-V2)
Graphics: NVIDIA GB300
Accelerator count: 4
Nodes: 1

Recorded results

Only the same benchmark, version, workload, scenario and unit share a row. Missing results are not zero. System results include the full submitted hardware and software configuration.

MLPerf Inference / gpt-oss-120bOffline · v6.1 · Tokens/s

4 × NVIDIA GB300

65,511.90^* Tokens/s

Accuracy target / result: exact_match: 82.738
Weight precision: mxfp4
Division: closed
System type: datacenter
Availability: available
Cores per processor: 72
Processors per node: 2
Nodes: 1
Total accelerators: 4
Software stack: TensorRT-LLM 1.3.0rc14 (GPT-OSS-120B); vLLM (CentML) 0.20.2rc1.dev2091+g2e1040b7e + NVIDIA Dynamo (Qwen3-VL-235B); CUDA 13.0
Operating system: Ubuntu 24.04.4 LTS (aarch64)

MLPerf Inference / gpt-oss-120bServer · v6.1 · Tokens/s

4 × NVIDIA GB300

58,195.80^* Tokens/s

Accuracy target / result: exact_match: 83.079
Weight precision: mxfp4
Division: closed
System type: datacenter
Availability: available
Cores per processor: 72
Processors per node: 2
Nodes: 1
Total accelerators: 4
Software stack: TensorRT-LLM 1.3.0rc14 (GPT-OSS-120B); vLLM (CentML) 0.20.2rc1.dev2091+g2e1040b7e + NVIDIA Dynamo (Qwen3-VL-235B); CUDA 13.0
Operating system: Ubuntu 24.04.4 LTS (aarch64)

Processing & platform

Processor: NVIDIA Grace CPU (Arm Neoverse-V2)
Graphics: NVIDIA GB300

Connectivity & software

Operating system: Ubuntu 24.04.4 LTS (aarch64)

System configuration

Reported host core count: 72
Processors per node: 2
Accelerator count: 4
Nodes: 1
Software stack: TensorRT-LLM 1.3.0rc14 (GPT-OSS-120B); vLLM (CentML) 0.20.2rc1.dev2091+g2e1040b7e + NVIDIA Dynamo (Qwen3-VL-235B); CUDA 13.0
Submission division: closed

* Provisional data