512 × AMD Instinct MI355X 288GB HBM3e

Tested system

Processor: AMD EPYC 9575F
Graphics: AMD Instinct MI355X 288GB HBM3e
Accelerator count: 512
Nodes: 64

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 / deepseek-r1Offline · v6.1 · Tokens/s

512 × AMD Instinct MI355X 288GB HBM3e

2,901,950^* Tokens/s

Accuracy target / result: exact_match: 80.69735642661804 TOKENS_PER_SAMPLE: 3898.373746581586
Weight precision: fp4
Division: closed
System type: datacenter
Availability: available
Cores per processor: 64
Processors per node: 2
Nodes: 64
Total accelerators: 512
Software stack: SGLang 0.5.15.post1 (ROCm 7.2.0), PyTorch 2.x
Operating system: Ubuntu 24.04 (node amdgpu driver 6.14.x; container ROCm 7.2.0 userspace)

MLPerf Inference / deepseek-r1Server · v6.1 · Tokens/s

512 × AMD Instinct MI355X 288GB HBM3e

2,405,310^* Tokens/s

Accuracy target / result: exact_match: 80.69735642661804 TOKENS_PER_SAMPLE: 3898.373746581586
Weight precision: fp4
Division: closed
System type: datacenter
Availability: available
Cores per processor: 64
Processors per node: 2
Nodes: 64
Total accelerators: 512
Software stack: SGLang 0.5.15.post1 (ROCm 7.2.0), PyTorch 2.x
Operating system: Ubuntu 24.04 (node amdgpu driver 6.14.x; container ROCm 7.2.0 userspace)

Processing & platform

Processor: AMD EPYC 9575F
Graphics: AMD Instinct MI355X 288GB HBM3e

Connectivity & software

Operating system: Ubuntu 24.04 (node amdgpu driver 6.14.x; container ROCm 7.2.0 userspace)

System configuration

Reported host core count: 64
Processors per node: 2
Accelerator count: 512
Nodes: 64
Software stack: SGLang 0.5.15.post1 (ROCm 7.2.0), PyTorch 2.x
Submission division: closed

* Provisional data