Intel Omni-Path Architecture Once Again Leads on the Top500

Intel® Omni-Path Architecture (Intel® OPA) still leads the rankings among 100 Gbps fabrics in the fastest supercomputers in the world1. Continued high installation numbers of Intel OPA illustrate that application performance at customer sites and the price/performance of Intel’s fabric compel decision-makers at the leading research institutions around the world to choose Intel OPA over InfiniBand* EDR.

Since the last Top500 listing in June of 2017, Intel OPA has increased another 20 percent in total FLOPS and listings in the top 100 systems2, including:

  • #7 Oakforest-PACS at JCAHPC in Japan, now the fastest supercomputer in Japan
  • #12 Stampede 2 at the Texas Advanced Computing Center (TACC) at University of Texas in the U.S.
  • #13 MareNostrum at Barcelona Supercomputing Center (BSC) in Spain
  • #14 Marconi at CINECA in Italy

Additionally, Penguin Computing has installed nine Intel OPA-based systems under the CTS-1 program, which appear on the Top500. Overall, Intel OPA claims 40 percent more systems in the top 100 entries on the Top500 list than InfiniBand EDR (100 Gbps)3, and 80 percent of those in the top 20 100 Gbps systems4.

Intel® Omni-Path Architecture (Intel® OPA) still leads the rankings among 100 Gbps fabrics in the fastest supercomputers in the world

Over the last two years, and highlighted in the last six months, decision makers are seeing greater price/performance, value, and raw performance delivered by Intel® Omni-Path Architecture, leading them to choose Intel-based solutions over InfiniBand* architecture.

Deeper analysis of the November 2017 Top500 list reveals further positive trends:

  • In terms of delivered performance, 10 percent of total FLOPS from all Top500 supercomputers comes from Intel OPA-connected systems5.
  • Looking at clusters with just 100 Gbps fabrics on the Top500, Intel OPA-based systems deliver a total of 83.4 petaFLOPS, compared to 69.5 petaFLOPS from InfiniBand—that’s 1.2X more petaFLOPS from Intel OPA-connected supercomputers6.

Considering that 63.6 percent of servers on the Top500 with 100 Gbps fabrics use Intel OPA, compared to 36.4 percent connected with InfiniBand, those numbers are not surprising7. For a new technology that’s only 18 months out of the launch gate, Intel OPA is making a significant impact on supercomputing.63.6 percent of servers on the Top500 with 100 Gbps fabrics use Intel OPA, compared to 36.4 percent connected with InfiniBand

Intel OPA’s leading position highlights that, while the benchmark claims from system integrators and manufacturers are important, in the end, the application performance at the customer’s site and the price/performance are critical to decision makers. It appears that over the last 18 months and highlighted in the last six months, those decision makers are seeing greater price/performance, value, and raw performance delivered by Intel OPA, leading them to choose Intel technology-based solutions over InfiniBand.

We are excited to continue Intel OPA’s momentum in 2018 and to help enable optimal HPC performance for more customers. For more information on how Intel Omni-Path Architecture can help your institution or company deliver its next scientific breakthrough, please visit intel.com/omnipath.

Performance estimates were obtained prior to implementation of recent software patches and firmware updates intended to address exploits referred to as "Spectre" and "Meltdown."  Implementation of these updates may make these results inapplicable to your device or system.

Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. For more information go to www.intel.com/benchmarks.

1 https://www.top500.org/lists/2017/11/
2 Based on FLOPS for Intel OPA systems ranked in the top 100 of the November 2017 Top500 List and FLOPS for Intel OPA systems ranked in top 100 of the June 2017 Top500 List and on the number of Intel OPA systems in the top 100 of the November 2017 Top500 List and the number of Intel OPA systems in the top 100 of the June 2017 Top500 List. https://www.top500.org/lists/2017/11/, https://www.top500.org/lists/2017/06/
3 Based on the number of Intel OPA systems in the top 100 of the November 2017 Top500 List and the number of Infiniband EDR systems in the top 100 of the November 2017 Top500 List. https://www.top500.org/lists/2017/11/
4 Based on the number of Intel OPA systems in the top 20 100 Gbps systems on the November 2017 Top500 List and the number of Ininiband EDR systems in the top 20 100 Gbps systems on the November 2017 Top500 List. https://www.top500.org/lists/2017/11/
5 Based on total FLOPS for November 2017 Top500 List Intel OPA systems compared to total FLOPS for all November 2017 Top500 List systems. https://www.top500.org/lists/2017/11/
6 Based on FLOPS for Intel OPA systems with 100 Gbps fabrics in the November 2017 Top500 list compared to FLOPS for InfiniBand EDR systems with 100 Gbps fabrics in the November 2017 Top500 list. https://www.top500.org/lists/2017/11/
7 Based on the number of servers for November 2017 Top 500 List Intel OPA systems compared to the number of servers for November 2017 Top 500 List InfiniBand EDR systems. https://www.top500.org/lists/2017/11/