Single-Node Performance

SPEC HPC 2021 Tiny Benchmark Results

These benchmarks demonstrate single-node performance across different hardware configurations, executed with various parallelization strategies (MPI, OpenMP) and job submission methods (bare-metal, SLURM).

Dell PowerEdge R6625 (AMD EPYC 9754)

The Dell R6625 systems powered by AMD EPYC 9754 processors represent our primary compute platform.

Scheduling	Parallelization	Ranks	Threads	Base Score	Full Report
Bare-metal	MPI	256	1	13.5	PDF
SLURM	MPI	256	1	12.8	PDF
SLURM	MPI+OpenMP	32	8	14.9	PDF

Gigabyte R183 (AMD EPYC 9754)

The Gigabyte R183 systems also feature AMD EPYC 9754 processors, demonstrating vendor-neutral performance.

Scheduling	Parallelization	Ranks	Threads	Base Score	Full Report
SLURM	MPI	256	1	12.4	PDF
SLURM	MPI+OpenMP	32	8	15.0	PDF

Dell PowerEdge R660 (Intel Xeon 8592+)

The Dell R660 systems powered by Intel Xeon 8592+ processors (dual socket) provide a comparison point for Intel-based architectures.

Scheduling	Parallelization	Ranks	Threads	Base Score	Full Report
SLURM	MPI	128	1	8.92	PDF

Key Observations

• Dell R6625 and Gigabyte R183 platforms deliver consistent performance with the same AMD EPYC 9754 processors

• Dell R660 with Intel Xeon 8592+ shows lower single-node score (8.92) compared to AMD platforms (12.4-15.0), reflecting architectural differences

• AMD EPYC 9754 delivers approximately 70% better per-core performance, while also providing double the core count per node compared to Intel Xeon 8592+

• No significant performance degradation observed across different vendors with the same processor

• SLURM job scheduler introduces minimal overhead compared to bare-metal execution

• Both MPI and OpenMP parallelization strategies are well-supported across all platforms

Detailed hardware specifications, software configurations, compiler flags, and runtime parameters for each system are available in the corresponding PDF reports.