Graph500

The Graph500 is a rating of supercomputer systems, focused on data-intensive loads. The project was announced on International Supercomputing Conference in June 2010. The first list was published at the ACM/IEEE Supercomputing Conference in November 2010. New versions of the list are published twice a year. The main performance metric used to rank the supercomputers is GTEPS (giga- traversed edges per second).

Richard Murphy from Sandia National Laboratories, says that "The Graph500's goal is to promote awareness of complex data problems", instead of focusing on computer benchmarks like HPL (High Performance Linpack), which TOP500 is based on.[1]

Despite its name, there were several hundreds of systems in the rating, growing up to 174 in June 2014.[2]

The algorithm and implementation that won the championship is published in the paper titled "Extreme scale breadth-first search on supercomputers".[3]

There is also list Green Graph 500, which uses same performance metric, but sorts list according to performance per Watt, like Green 500 works with TOP500 (HPL).

Benchmark

[edit]

The benchmark used in Graph500 stresses the communication subsystem of the system, instead of counting double precision floating-point.[1] It is based on a breadth-first search in a large undirected graph (a model of Kronecker graph with average degree of 16). There are three computation kernels in the benchmark: the first kernel is to generate the graph and compress it into sparse structures CSR or CSC (Compressed Sparse Row/Column); the second kernel does a parallel BFS search of some random vertices (64 search iterations per run); the third kernel runs a single-source shortest paths (SSSP) computation. Six possible sizes (Scales) of graph are defined: toy (226 vertices; 17 GB of RAM), mini (229; 137 GB), small (232; 1.1 TB), medium (236; 17.6 TB), large (239; 140 TB), and huge (242; 1.1 PB of RAM).[4]

The reference implementation of the benchmark contains several versions:[5]

  • serial high-level in GNU Octave
  • serial low-level in C
  • parallel C version with usage of OpenMP
  • two versions for Cray-XMT
  • basic MPI version (with MPI-1 functions)
  • optimized MPI version (with MPI-2 one-sided communications)

The implementation strategy that have won the championship on the Japanese K computer is described in.[6]

Top 10 ranking

[edit]

According to June 2024 release of the list, for the BFS results section, Fugaku ranks highest, but in the SSSP results section Wuhan Supercomputer ranks highest, then Pengcheng Cloudbrain-II, then Fugaku; table shows for BFS results:[7]

Rank Country Site Machine (architecture) Number of nodes Number of cores Problem scale GTEPS
1  Japan RIKEN Advanced Institute for Computational Science Supercomputer Fugaku (Fujitsu A64FX) 152064 7299072 42 166029
2  China Wuhan Kunpeng 920+Tesla A100 252 6999552 41 115357.6
3  USA Frontier HPE Cray EX235a 9248 8730112 40 29654.6
4  China Pengcheng Lab Pengcheng Cloudbrain-II (Kunpeng 920+Ascend 910) 488 93696 40 25242.9
5  USA DOE/SC/Argonne National Laboratory HPE Cray EX - Intel Exascale Compute Blade 4096 25591808 40 24250.2
6  China National Supercomputing Center in Wuxi Sunway TaihuLight (Sunway MPP) 40768 10599680 40 23755.7

Spain (Barcelona), has a new supercomputer MareNostrum 5 ACC, ranked 8th.

2022

[edit]

According to November 2022 release of the list:[8]

Rank Country Site Machine (architecture) Number of nodes Number of cores Problem scale GTEPS
1  Japan RIKEN Advanced Institute for Computational Science Supercomputer Fugaku (Fujitsu A64FX) 158976 7630848 41 102955
2  China Pengcheng Lab Pengcheng Cloudbrain-II (Kunpeng 920+Ascend 910) 488 93696 40 25242.9
3  China National Supercomputing Center in Wuxi Sunway TaihuLight (Sunway MPP) 40768 10599680 40 23755.7
4  Japan Information Technology Center, University of Tokyo Wisteria/BDEC-01 (PRIMEHPC FX1000) 7680 368640 37 16118
5  Japan Japan Aerospace Exploration Agency TOKI-SORA (PRIMEHPC FX1000) 5760 276480 36 10813
6  EU EuroHPC/CSC LUMI-C (HPE Cray EX) 1492 190976 38 8467.71
7  US Oak Ridge National Laboratory OLCF Summit (IBM POWER9) 2048 86016 40 7665.7
8  Germany Leibniz Rechenzentrum SuperMUC-NG (ThinkSystem SD530 Xeon Platinum 8174 24C 3.1GHz Intel Omni-Path) 4096 196608 39 6279.47
9  Germany Zuse Institute Berlin Lise (Intel Omni-Path) 1270 121920 38 5423.94
10  China National Engineering Research Center for Big Data Technology and System DepGraph Supernode (DepGraph (+GPU Tesla A100)) 1 128 33 4623.379

2020

[edit]

Arm-based Fugaku took the top spot of the list.[9]

2016

[edit]

According to June 2016 release of the list:[10]

Rank Site Machine (architecture) Number of nodes Number of cores Problem scale GTEPS
1 Riken Advanced Institute for Computational Science K computer (Fujitsu custom) 82944 663552 40 38621.4
2 National Supercomputing Center in Wuxi Sunway TaihuLight (NRCPC - Sunway MPP) 40768 10599680 40 23755.7
3 Lawrence Livermore National Laboratory IBM Sequoia (Blue Gene/Q) 98304 1572864 41 23751
4 Argonne National Laboratory IBM Mira (Blue Gene/Q) 49152 786432 40 14982
5 Forschungszentrum Jülich JUQUEEN (Blue Gene/Q) 16384 262144 38 5848
6 CINECA Fermi (Blue Gene/Q) 8192 131072 37 2567
7 Changsha, China Tianhe-2 (NUDT custom) 8192 196608 36 2061.48
8 CNRS/IDRIS-GENCI Turing (Blue Gene/Q) 4096 65536 36 1427
8 Science and Technology Facilities Council – Daresbury Laboratory Blue Joule (Blue Gene/Q) 4096 65536 36 1427
8 University of Edinburgh DIRAC (Blue Gene/Q) 4096 65536 36 1427
8 EDF R&D Zumbrota (Blue Gene/Q) 4096 65536 36 1427
8 Victorian Life Sciences Computation Initiative Avoca (Blue Gene/Q) 4096 65536 36 1427

2014

[edit]

According to June 2014 release of the list:[2]

Rank Site Machine (architecture) Number of nodes Number of cores Problem scale GTEPS
1 RIKEN Advanced Institute for Computational Science K computer (Fujitsu custom) 65536 524288 40 17977.1
2 Lawrence Livermore National Laboratory IBM Sequoia (Blue Gene/Q) 65536 1048576 40 16599
3 Argonne National Laboratory IBM Mira (Blue Gene/Q) 49152 786432 40 14328
4 Forschungszentrum Jülich JUQUEEN (Blue Gene/Q) 16384 262144 38 5848
5 CINECA Fermi (Blue Gene/Q) 8192 131072 37 2567
6 Changsha, China Tianhe-2 (NUDT custom) 8192 196608 36 2061.48
7 CNRS/IDRIS-GENCI Turing (Blue Gene/Q) 4096 65536 36 1427
7 Science and Technology Facilities Council - Daresbury Laboratory Blue Joule (Blue Gene/Q) 4096 65536 36 1427
7 University of Edinburgh DIRAC (Blue Gene/Q) 4096 65536 36 1427
7 EDF R&D Zumbrota (Blue Gene/Q) 4096 65536 36 1427
7 Victorian Life Sciences Computation Initiative Avoca (Blue Gene/Q) 4096 65536 36 1427

2013

[edit]

According to June 2013 release of the list:[11]

Rank Site Machine (architecture) Number of nodes Number of cores Problem scale GTEPS
1 Lawrence Livermore National Laboratory IBM Sequoia (Blue Gene/Q) 65536 1048576 40 15363
2 Argonne National Laboratory IBM Mira (Blue Gene/Q) 49152 786432 40 14328
3 Forschungszentrum Jülich JUQUEEN (Blue Gene/Q) 16384 262144 38 5848
4 RIKEN Advanced Institute for Computational Science K computer (Fujitsu custom) 65536 524288 40 5524.12
5 CINECA Fermi (Blue Gene/Q) 8192 131072 37 2567
6 Changsha, China Tianhe-2 (NUDT custom) 8192 196608 36 2061.48
7 CNRS/IDRIS-GENCI Turing (Blue Gene/Q) 4096 65536 36 1427
7 Science and Technology Facilities Council - Daresbury Laboratory Blue Joule (Blue Gene/Q) 4096 65536 36 1427
7 University of Edinburgh DIRAC (Blue Gene/Q) 4096 65536 36 1427
7 EDF R&D Zumbrota (Blue Gene/Q) 4096 65536 36 1427
7 Victorian Life Sciences Computation Initiative Avoca (Blue Gene/Q) 4096 65536 36 1427

See also

[edit]

References

[edit]
  1. ^ a b The Exascale Report (March 15, 2012). "The Case for the Graph 500 – Really Fast or Really Productive? Pick One". Inside HPC.
  2. ^ a b "June 2014 | Graph 500". Archived from the original on June 28, 2014. Retrieved June 26, 2014.
  3. ^ Ueno, Koji; Suzumura, Toyotaro; Maruyama, Naoya; Fujisawa, Katsuki; Matsuoka, Satoshi (2016). "Extreme scale breadth-first search on supercomputers". 2016 IEEE International Conference on Big Data (Big Data). pp. 1040–1047. doi:10.1109/BigData.2016.7840705. ISBN 978-1-4673-9005-7. S2CID 8680200.
  4. ^ Performance Evaluation of Graph500 on Large-Scale Distributed Environment // IEEE IISWC 2011, Austin, TX; presentation
  5. ^ "Graph500: адекватный рейтинг" (in Russian). Open Systems #1 2011.
  6. ^ Ueno, K.; Suzumura, T.; Maruyama, N.; Fujisawa, K.; Matsuoka, S. (December 1, 2016). "Extreme scale breadth-first search on supercomputers". 2016 IEEE International Conference on Big Data (Big Data). pp. 1040–1047. doi:10.1109/BigData.2016.7840705. ISBN 978-1-4673-9005-7. S2CID 8680200.
  7. ^ "Complete Results - Graph 500". 2024. Retrieved July 20, 2024.
  8. ^ "November 2022; Graph 500". June 14, 2017. Retrieved November 18, 2022.
  9. ^ "Fujitsu and RIKEN Take First Place in Graph500 Ranking with Supercomputer Fugaku". HPCwire. June 23, 2020. Retrieved August 8, 2020.
  10. ^ "June 2016 | Graph 500". Archived from the original on June 24, 2016. Retrieved July 6, 2016.
  11. ^ "June 2013 | Graph 500". Archived from the original on June 21, 2013. Retrieved June 19, 2013.
[edit]