Azure HBv2 joins Graph500 Top 20 list

This post has been republished via RSS; it originally appeared at: Azure Compute articles.


The Azure HPC team is proud to join the Top 20 list in the Graph500 ranking announced at the Supercomputing 2020 Conference. We used Azure HBv2 VM instances for our Graph500 run, achieving 1152 GTEPS (Giga Traversed Edges Per Seconds) for the Graph500 Breadth First Search kernel. This resulted in Azure ranking 17th all-time on the Graph500. To the best of our knowledge, this is the first public cloud cluster entry in Graph500 ranking list.



Graph500 Benchmark:

Graphs are one of the most ubiquitous models in analytical workloads. They are powerful representations of many types of relations and process dynamics. They are used various scientific and engineering fields like medical informatics, social networks, symbolic networks, financial applications, cyber security, etc. Basic graph algorithms such as breadth-first search, depth-first search, shortest path, minimum spanning tree, etc. are key components in many modern real-life applications.

The Graph500 benchmark is designed to represent data intensive and irregular applications that use graph algorithm-based processing methods. Further, it is used to evaluate the scalability of supercomputing clusters in the context of data-intensive applications with irregular communication characteristics. The Graph500 benchmark stresses hardware and runtime systems by forcing massive amounts of communication and synchronization thereby modeling more realistic application workloads. Graph500 Consists of two sets of benchmarks – Search (Breadth First Search) and Shortest Path. We used the BFS benchmark for this submission.



Submission Details:

We used 640 Azure HBv2 VM instances for this Graph500 experiment. HBv2 VM instances offer 200 gigabits per second HDR InfiniBand networking that enables Graph500 benchmark to efficiently scale across hundreds of compute nodes. Each HBv2 VM features 120 AMD EPYC™ 7002-series CPU cores with clock frequencies up to 3.3 GHz, 480 GB of RAM, 350 GB/s of memory bandwidth, and a large L3 cache of 480 MB. The System and Benchmark Configuration that we used are depicted below:


System Configuration:

  • 640 x Azure HBv2 VMs (AMD “Rome” 7V12, 2.45-3.3 GHz, 480 GB of RAM)
  • Operating System: CentOS-HPC 7.7
  • Network: InfiniBand HDR 200 Gbps
  • MPI Library: HPC-X

Benchmark Configuration:

REMEMBER: these articles are REPUBLISHED. Your best bet to get a reply is to follow the link at the top of the post to the ORIGINAL post! BUT you're more than welcome to start discussions here:

This site uses Akismet to reduce spam. Learn how your comment data is processed.