StorageReview Breaks Pi Calculation Record with 314 Trillion Digits on Dell PowerEdge R7725
Key Takeaways
- ▸StorageReview achieved a new world pi calculation record of 314 trillion digits on a single 2U Dell PowerEdge R7725 server, breaking the previous 300 trillion-digit record
- ▸The system leveraged dual AMD EPYC 9965 CPUs (384 cores total) and 2.4PB of Micron 6550 Ion NVMe SSDs, achieving storage performance of up to 280GB/s
- ▸Design improvements in 17th-generation Dell PowerEdge servers—including direct-connect PCIe backplanes—enabled significant performance gains (72-383%) over previous-generation infrastructure
Summary
StorageReview has set a new world record by calculating pi to 314 trillion digits using a single 2U Dell PowerEdge R7725 server, surpassing the previous record of 300 trillion digits held by Linus Media Group and KIOXIA. The achievement was accomplished using dual AMD EPYC 192-core CPUs and forty 61.44TB Micron 6550 Ion SSDs, with 34 drives allocated to the y-cruncher computation and 6 configured in RAID10 for data recording. The compute run began on July 31, 2025, and completed during the second day of SC25, the international HPC conference.
The record demonstrates how on-premises, single-system infrastructure can outperform hyperscale cloud solutions for storage-intensive computational workloads. StorageReview's progression from 105 trillion digits (early 2024) to 202 trillion (mid-2024) to the current 314 trillion digits showcases the efficiency gains possible through careful hardware selection and system tuning. The key to success was leveraging the Dell PowerEdge R7725's Gen5 E3.S backplane with direct-connect architecture, which delivered up to 280GB/s read and write performance across all 40 bays simultaneously—a 72-383% performance improvement compared to previous-generation systems.
- The achievement validates that optimized single-system on-premises infrastructure can outperform large hyperscale cloud clusters for specific demanding computational workloads
Editorial Opinion
This record underscores an important inflection point in HPC: not every computational challenge benefits from massive distributed cloud infrastructure. By focusing on hardware-software co-optimization—pairing modern high-density NVMe SSDs with carefully engineered server architectures—StorageReview demonstrates that carefully-tuned single systems can achieve results that rival or exceed expensive, power-hungry multi-node clusters. This has meaningful implications for organizations evaluating cloud versus on-premises solutions for storage-intensive workloads.
