超算 TOP500 榜单公布：中国依然数量最多，英伟达再发 HGX A100

2021-06-30 06:54IT之家 - 问舟

IT之家 6 月 30 日消息一年一度的高性能计算大会 ISC 如期而至，世界 TOP 500 超级计算机排名昨日正式公布。

对比去年的榜单，今年 TOP 10 的榜单中，来自美国新能源部劳伦斯伯克利国家实验室的 Perlmutter 系统新晋入围。中国超级计算机的数量虽然下降至 186 台（去年 212），但对比位居第二的美国的 123 台，依然遥遥领先。不过美国超算算力更强，美国超算的综合性能为 856.8 Pflop/s，而中国超算为 445.3 Pflop/s。

以下是 Top10 中的系统摘要（引用自 Top500）：

日本的 Fugaku 仍然是第一的超算系统。它拥有 7,630,848 个内核，使其能够达到 442 Pflop/s 的 HPL 基准测试分数。这使它领先第二名 3 倍。
Summit 是美国田纳西州橡树岭国家实验室 (ORNL) 的 IBM 构建的系统，仍然是美国最快的系统，在 HPL 基准测试中的性能为 148.8 Pflop/s，全球排名第二。用于对 TOP500 榜单进行排名。Summit 有 4,356 个节点，每个节点包含两个 Power9 CPU，每个 CPU 有 22 个内核，六个 NVIDIA Tesla V100 GPU，每个都有 80 个流式多处理器 (SM)。这些节点通过 Mellanox 双轨 EDR InfiniBand 网络链接在一起。
美国加利福尼亚州劳伦斯利弗莫尔国家实验室的 Sierra 系统排名第 3。其架构与 #2 的 Summit 系统非常相似。它由 4,320 个节点构建，具有两个 Power9 CPU 和四个 NVIDIA Tesla V100 GPU。Sierra 达到了 94.6 Pflop/s。
由中国国家并行计算机工程与技术研究中心（NRCPC）开发并设立在江苏无锡国家超级计算中心的神威太湖之光以 93 Pflop/s 排名第四.
排名第 5 的 Perlmutter 是 TOP10 中的新晋成员。它基于 HPE Cray“Shasta”平台，以及一个具有基于 AMD EPYC 的节点和 1536 个 NVIDIA A100 加速节点的异构系统。Perlmutter 达到了 64.6 Pflop/s。
Selene 现在排名第 6，是 NVIDIA 内部安装的 NVIDIA DGX A100 SuperPOD。该系统基于 AMD EPYC 处理器，采用 NVIDIA A100 加速，Mellanox HDR InfiniBand 作为网络，实现了 63.4 Pflop/s。
由中国国防科技大学（NUDT）开发并部署在中国广州国家超级计算机中心的系统天河 2A（Milky Way-2A）现在以 61.4 Pflop /s 被列为世界第 7 系统。
名为“JUWELS Booster Module”系统是 No. 8。由 Atos 构建的 BullSequana 系统安装在德国的 Forschungszentrum Juelich (FZJ)。该系统使用带有 NVIDIA A100 的 AMD EPYC 处理器进行加速，并使用 Mellanox HDR InfiniBand 作为类似于 Selene 系统的网络。这是欧洲最强大的系统，算力达 44.1 Pflop/s。
排在第 9 位的 HPC5 是戴尔构建、意大利 Eni SpA 安装的 PowerEdge 系统，由于使用 NVIDIA Tesla V100 作为加速器和 Mellanox HDR InfiniBand 作为网络，实现了 35.5 Pflop/s 的性能。
Frontera 是一款戴尔 C6420 系统，安装在得克萨斯大学的得克萨斯高级计算中心，目前排名第 10。它使用 448,448 个英特尔至强内核实现了 23.5 Pflop/s。

值得注意的是，此次榜单中大多数系统采用了 AMD 和英伟达的方案。另外，名单上更新的两个新系统“超级云”，能够同时满足 AI、高性能计算（HPC）和云端需求，这意味着，高性能计算正在同 AI 加速融合。

此外，英伟达也在本次 ISC 大会上发布了全新的 HGX A100 系统，且已经在英国爱丁堡大学托管的 DiRAC 超级计算机中落地。

据介绍，英伟达 HGX A100 加入了三项关键技术：NVIDIA A100 80GB PCIe GPU、NVIDIA NDR 400G InfiniBand 网络和 NVIDIA Magnum IO GPUDirect Storage 软件。

其中，A100 80GB PCle GPU 采用 NVIDIA Ampere 架构，与 A100 40GB 相比其内存带宽带到 2TB/S，提升 25% 。更高的内存容量和内存带宽，能够将更多的数据和更大的神经网络保存在内存中，从而最大限度地减少节点通信和能耗，研究人员也能获得更高的吞吐率和更快的结果。

目前，英伟达最新 HGX 高性能计算平台已落地英国爱丁堡大学，为其托管的新型 DiRAC 超级计算机 Tursa 提供动力支持。

IT之家提醒，以下为 Top100：

现	前	名称	配置	位置
1	1	Supercomputer Fugaku	Supercomputer Fugaku, A64FX 48C 2.2GHz, Tofu interconnect D	RIKEN Center for Computational Science
2	2	Summit	IBM Power System AC922, IBM POWER9 22C 3.07GHz, NVIDIA Volta GV100, Dual-rail Mellanox EDR Infiniband	DOE/SC/Oak Ridge National Laboratory
3	3	Sierra	IBM Power System AC922, IBM POWER9 22C 3.1GHz, NVIDIA Volta GV100, Dual-rail Mellanox EDR Infiniband	DOE/NNSA/LLNL
4	4	Sunway TaihuLight	Sunway MPP, Sunway SW26010 260C 1.45GHz, Sunway	National Supercomputing Center in Wuxi
5		Perlmutter	HPE Cray EX235n, AMD EPYC 7763 64C 2.45GHz, NVIDIA A100 SXM4 40 GB, Slingshot-10	DOE/SC/LBNL/NERSC
6	5	Selene	NVIDIA DGX A100, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100, Mellanox HDR Infiniband	NVIDIA Corporation
7	6	Tianhe-2A	TH-IVB-FEP Cluster, Intel Xeon E5-2692v2 12C 2.2GHz, TH Express-2, Matrix-2000	National Super Computer Center in Guangzhou
8	7	JUWELS Booster Module	Bull Sequana XH2000 , AMD EPYC 7402 24C 2.8GHz, NVIDIA A100, Mellanox HDR InfiniBand/ParTec ParaStation ClusterSuite	Forschungszentrum Juelich (FZJ)
9	8	HPC5	PowerEdge C4140, Xeon Gold 6252 24C 2.1GHz, NVIDIA Tesla V100, Mellanox HDR Infiniband	Eni S.p.A.
10	9	Frontera	Dell C6420, Xeon Platinum 8280 28C 2.7GHz, Mellanox InfiniBand HDR	Texas Advanced Computing Center/Univ. of Texas
11	10	Dammam-7	Cray CS-Storm, Xeon Gold 6248 20C 2.5GHz, NVIDIA Tesla V100 SXM2, InfiniBand HDR 100	Saudi Aramco
12		ABCI 2.0	PRIMERGY GX2570 M6, Xeon Platinum 8360Y 36C 2.4GHz, NVIDIA A100 SXM4 40 GB, Infiniband HDR	National Institute of Advanced Industrial Science and Technology (AIST)
13		Wisteria/BDEC-01 (Odyssey)	PRIMEHPC FX1000, A64FX 48C 2.2GHz, Tofu interconnect D	Information Technology Center, The University of Tokyo
14	11	Marconi-100	IBM Power System AC922, IBM POWER9 16C 3GHz, Nvidia Volta V100, Dual-rail Mellanox EDR Infiniband	CINECA
15	12	Piz Daint	Cray XC50, Xeon E5-2690v3 12C 2.6GHz, Aries interconnect , NVIDIA Tesla P100	Swiss National Supercomputing Centre (CSCS)
16	13	Trinity	Cray XC40, Xeon E5-2698v3 16C 2.3GHz, Intel Xeon Phi 7250 68C 1.4GHz, Aries interconnect	DOE/NNSA/LANL/SNL
17	15	SuperMUC-NG	ThinkSystem SD650, Xeon Platinum 8174 24C 3.1GHz, Intel Omni-Path	Leibniz Rechenzentrum
18	16	Hawk	Apollo 9000, AMD EPYC 7742 64C 2.25GHz, Mellanox HDR Infiniband	HLRS - Höchstleistungsrechenzentrum Stuttgart
19		Ghawar-1	HPE Cray EX, AMD EPYC 7702 64C 2GHz, Slingshot-10	Saudi Aramco
20	17	Lassen	IBM Power System AC922, IBM POWER9 22C 3.1GHz, Dual-rail Mellanox EDR Infiniband, NVIDIA Tesla V100	DOE/NNSA/LLNL
21	18	PANGEA III	IBM Power System AC922, IBM POWER9 18C 3.45GHz, Dual-rail Mellanox EDR Infiniband, NVIDIA Volta GV100	Total Exploration Production
22		HiPerGator AI	NVIDIA DGX A100, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100, Infiniband HDR	University of Florida
23		Maru	ThinkSystem SD650 V2, Xeon Platinum 8368Q 38C 2.6GHz, Infiniband HDR	Korean Meteorological Administration
24		Guru	ThinkSystem SD650 V2, Xeon Platinum 8368Q 38C 2.6GHz, Infiniband HDR	Korean Meteorological Administration
25	19	TOKI-SORA	PRIMEHPC FX1000, A64FX 48C 2.2GHz, Tofu interconnect D	Japan Aerospace eXploration Agency
26		Pioneer-EUS	NDv4 cluster, AMD EPYC 7V12 48C 2.45GHz, NVIDIA A100, Infiniband HDR	Azure East US
27		Pioneer-SCUS	NDv4 cluster, AMD EPYC 7V12 48C 2.45GHz, NVIDIA A100, Infiniband HDR	Azure South Central US
28		Pioneer-WEU	NDv4 cluster, AMD EPYC 7V12 48C 2.45GHz, NVIDIA A100, Infiniband HDR	Azure West Europe
29		Pioneer-WUS2	NDv4 cluster, AMD EPYC 7V12 48C 2.45GHz, NVIDIA A100, Infiniband HDR	Azure West US 2
30	20	Cori	Cray XC40, Intel Xeon Phi 7250 68C 1.4GHz, Aries interconnect	DOE/SC/LBNL/NERSC
31	21	Nurion	Cray CS500, Intel Xeon Phi 7250 68C 1.4GHz, Intel Omni-Path	Korea Institute of Science and Technology Information
32	22	Oakforest-PACS	PRIMERGY CX1640 M1, Intel Xeon Phi 7250 68C 1.4GHz, Intel Omni-Path	Joint Center for Advanced High Performance Computing
33	23	HPC4	Proliant DL380 Gen10, Xeon Platinum 8160 24C 2.1GHz, Mellanox InfiniBand EDR, NVIDIA Tesla P100	Eni S.p.A.
34	24	Tera-1000-2	Bull Sequana X1000, Intel Xeon Phi 7250 68C 1.4GHz, Bull BXI 1.2	Commissariat a l'Energie Atomique (CEA)
35	25	Stampede2	PowerEdge C6320P/C6420, Intel Xeon Phi 7250 68C 1.4GHz/Platinum 8160, Intel Omni-Path	Texas Advanced Computing Center/Univ. of Texas
36		MeluXina - Accelerator Module	BullSequana XH2000, AMD EPYC 7452 32C 2.35GHz, NVIDIA A100 40GB, Mellanox HDR InfiniBand/ParTec ParaStation ClusterSuite	LuxProvide
37		Cactus	HPE Cray EX, AMD EPYC 7742 64C 2.25GHz, Slingshot-10	GDIT/NOAA/WCOSS
38		Dogwood	HPE Cray EX, AMD EPYC 7742 64C 2.25GHz, Slingshot-10	GDIT/NOAA/WCOSS
39		Earth Simulator -SX-Aurora TSUBASA	SX-Aurora TSUBASA A401-8, Vector Engine Type20B 8C 1.6GHz, Infiniband HDR200	Japan Agency for Marine-Earth Science and Technology
40		Amazon EC2 Instance Cluster us-east-1a	Amazon EC2 r5.24xlarge, Xeon Platinum 8260 24C 2.4GHz, 25G Ethernet	Descartes Labs
41		NVIDIA Cambridge-1 DGX SuperPOD	NVIDIA DGX A100, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100 80GB, Infiniband HDR	NVIDIA Corporation
42	26	DGX SuperPOD	NVIDIA DGX-2H, Xeon Platinum 8174 24C 3.1GHz, NVIDIA Tesla V100, Mellanox InfiniBand EDR	NVIDIA Corporation
43		JURECA Data Centric Module	BullSequana XH2000, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100 40GB, Mellanox HDR InfiniBand/ParTec ParaStation ClusterSuite	Forschungszentrum Juelich (FZJ)
44	27	Gadi	PRIMERGY CX2570 M5, ThinkSystem SD650, Xeon Platinum 8274/8268, NVIDIA Tesla V100 SXM2, Mellanox HDR Infiniband	National Computational Infrastructure (NCI Australia)
45	28	Taiwania 2	QCT QuantaGrid D52G-4U/LC, Xeon Gold 6154 18C 3GHz, Mellanox InfiniBand EDR, NVIDIA Tesla V100 SXM2	National Center for High Performance Computing
46		Dragão	Supermicro SYS-4029GP-TVRT, Xeon Gold 6230R 26C 2.1GHz, NVIDIA Tesla V100, Infiniband EDR	Petróleo Brasileiro S.A
47		Raven-GPU	ThinkSystem SD650-N V2, Xeon Platinum 8360Y 36C 2.4GHz, NVIDIA A100, Mellanox HDR Infiniband	Max-Planck-Gesellschaft MPI/IPP
48	29	AiMOS	IBM Power System AC922, IBM POWER9 20C 3.45GHz, NVIDIA Volta GV100, Dual-rail Mellanox EDR Infiniband	Rensselaer Polytechnic Institute Center for Computational Innovations (CCI)
49	30	Taranis	Bull Sequana XH2000 , AMD EPYC 7742 64C 2.25GHz, Mellanox InfiniBand HDR100	Meteo France
50	31	TSUBAME3.0	SGI ICE XA, IP139-SXM2, Xeon E5-2680v4 14C 2.4GHz, Intel Omni-Path, NVIDIA Tesla P100 SXM2	GSIC Center, Tokyo Institute of Technology
51	32	Roxy	Apollo 2000, Xeon Gold 6248 20C 2.5GHz, Infiniband EDR	Government
52		HoreKa-Green	ThinkSystem SD650 V2, Xeon Platinum 8368 38C 2.4GHz, NVIDIA A100 40GB, Mellanox HDR Infiniband	Karlsruher Institut für Technologie (KIT)
53		Narwhal	HPE Cray EX, AMD EPYC 7H12 64C 2.6GHz, Slingshot-10	Navy DSRC
54	33	Plasma Simulator	SX-Aurora TSUBASA A412-8, Vector Engine Type10AE 8C 1.58GHz, Infiniband HDR 200	National Institute for Fusion Science (NIFS)
55	34	Belenos	Bull Sequana XH2000 , AMD EPYC 7742 64C 2.25GHz, Mellanox HDR100	Meteo France
56	35	PupMaya	Apollo 2000, Xeon Gold 6148 20C 2.4GHz, Infiniband EDR	Government
57	36	Artemis	NVIDIA DGX-2, Xeon Platinum 8168 24C 2.7GHz, Infiniband EDR, NVIDIA Tesla V100	Group 42
58	37		Cray XC40, Xeon E5-2695v4 18C 2.1GHz, Aries interconnect	United Kingdom Meteorological Office
59	38	JOLIOT-CURIE ROME	Bull Sequana XH2000 , AMD Rome 7H12 64C 2.6GHz, Mellanox HDR100	CEA/TGCC-GENCI
60	39	Theta	Cray XC40, Intel Xeon Phi 7230 64C 1.3GHz, Aries interconnect	DOE/SC/Argonne National Laboratory
61	40	Christofari	NVIDIA DGX-2, Xeon Platinum 8168 24C 2.7GHz, Mellanox InfiniBand EDR, NVIDIA Tesla V100	SberCloud
62	41	Flow	PRIMEHPC FX1000, A64FX 48C 2.2GHz, Tofu interconnect D	Information Technology Center, Nagoya University
63	42	MareNostrum	Lenovo SD530, Xeon Platinum 8160 24C 2.1GHz, Intel Omni-Path	Barcelona Supercomputing Center
64	43	ONYX	Cray XC40, Xeon E5-2699v4 22C 2.2GHz, Aries interconnect	ERDC DSRC
65	44	JUWELS Module 1	Bull Sequana X1000, Xeon Platinum 8168 24C 2.7GHz, Mellanox EDR InfiniBand/ParTec ParaStation ClusterSuite	Forschungszentrum Juelich (FZJ)
66	45		HPE SGI 8600, Xeon Gold 6248R 24C 3GHz, NVIDIA Tesla V100 SXM2, Infiniband EDR	Japan Atomic Energy Agency(JAEA)/ National Institutes for Quantum and Radiological Science and Technology(QST)
67		SQUID - CPU Nodes	NEC LX 103Bj-8, Xeon Platinum 8368 38C 2.4GHz, Infiniband HDR200	Osaka University
68	55	Lise	Bull intel Cluster, Intel Xeon Platinum 9242 48C 2.3GHz, Intel Omni-Path	HLRN at ZIB/Konrad Zuse-Zentrum Berlin
69		Karolina, GPU partition	Apollo 6500, AMD EPYC 7452 32C 2.35GHz, NVIDIA A100 SXM4 40 GB, Infiniband HDR200	IT4Innovations National Supercomputing Center, VSB-Technical University of Ostrava
70	46	Pleiades	SGI ICE X, Intel Xeon E5-2670/E5-2680v2/E5-2680v3/E5-2680v4 2.6/2.8/2.5/2.4 GHz, Infiniband FDR	NASA/Ames Research Center/NAS
71	47	Emmy+	BullSequana X, Intel Xeon Platinum 9242/Gold 6148, Intel Omni-Path	HLRN+ at GWDG/University of Göttingen
72		Aitken	HPE Apollo 9000/HPE SGI 8600, AMD EPYC 7742 64C 2.25GHz/Intel Xeon Gold 6248 20C 2.5GHz, Mellanox EDR/HDR Infiniband	NASA/Ames Research Center/NAS
73	48	Sawtooth	HPE SGI 8600, Xeon Platinum 8268 24C 2.9GHz, Infiniband EDR/HDR	Idaho National Laboratory
74	49		Cray XC50, Xeon Platinum 8160 24C 2.1GHz, Aries interconnect	Japan Meteorological Agency
75	50		Cray XC50, Xeon Platinum 8160 24C 2.1GHz, Aries interconnect	Japan Meteorological Agency
76	51	COBRA	Intel Compute Module HNS2600BP, Xeon Gold 6148 20C 2.4GHz, Intel Omni-Path	Max-Planck-Gesellschaft MPI/IPP
77	52	Shaheen II	Cray XC40, Xeon E5-2698v3 16C 2.3GHz, Aries interconnect	King Abdullah University of Science and Technology
78	53	Electra	HPE SGI 8600/SGI ICE-X, E5-2680V4/ Xeon Gold 6148 20C 2.4GHz, Infiniband EDR/FDR-56 IB	NASA/Ames Research Center/NAS
79		Raven	ThinkSystem SD650 V2, Xeon Platinum 8360Y 36C 2.4GHz, InfiniBand HDR 100	Max-Planck-Gesellschaft MPI/IPP
80	54	MAHTI	Bull Sequana XH2000 , AMD Rome 7H12 64C 2.6GHz, Mellanox HDR100	CSC (Center for Scientific Computing)
81	56	Pangea	SGI ICE X, Xeon Xeon E5-2670/ E5-2680v3 12C 2.5GHz, Infiniband FDR	Total Exploration Production
82		Berzelius	NVIDIA DGX A100, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100, Infiniband HDR	National Supercomputer Centre (NSC)
83	57	TX-GAIA (Green AI Accelerator)	Apollo 2000, Xeon Gold 6248 20C 2.5GHz, NVIDIA Tesla V100, Intel Omni-Path	MIT Lincoln Laboratory Supercomputing Center
84		BioHive-1	NVIDIA DGX A100, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100, Mellanox HDR Infiniband	Recursion
85	58	Flow Type II subsystem	PRIMERGY CX2570 M5, Xeon Gold 6230 20C 2.1GHz, NVIDIA Tesla V100 SXM2, Infiniband EDR	Information Technology Center, Nagoya University
86	59	Eagle	HPE SGI 8600, Xeon Gold 6154 18C 3GHz, EDR Infiniband/ 8D Hypercube	National Renewable Energy Laboratory
87	60	Cheyenne	SGI ICE XA, Xeon E5-2697v4 18C 2.3GHz, Infiniband EDR	National Center for Atmospheric Research (NCAR)
88	61	Betzy	Bull Sequana XH2000 , AMD EPYC 7742 64C 2.25GHz, Mellanox HDR Infiniband	UNINETT Sigma2 AS
89	62	PARAM Siddhi-AI	NVIDIA DGX A100, AMD EPYC 7742 64C 2.25GHz, NVIDIA A100, Mellanox HDR Infiniband	Center for Development of Advanced Computing (C-DAC)
90	63	ITO - Subsystem A	Fujitsu PRIMERGY CX2550/CX2560 M4, Xeon Gold 6154 18C 3GHz, Mellanox InfiniBand EDR	Research Institute for Information Technology, Kyushu University
91		Discoverer	BullSequana XH2000, AMD EPYC 7H12 64C 2.6GHz, Infiniband HDR	Consortium Petascale Supercomputer Bulgaria
92	64	Jean Zay	HPE SGI 8600, Xeon Gold 6248 20C 2.5GHz, NVIDIA Tesla V100 SXM2, Intel Omni-Path	CNRS/IDRIS-GENCI
93		Wisteria/BDEC-01 (Aquarius)	PRIMERGY GX2570 M6, Xeon Platinum 8360Y 36C 2.4GHz, NVIDIA A100 SXM4 40 GB, Infiniband HDR	Information Technology Center, The University of Tokyo
94	65	Atlas	Bull 4029GP-TVRT, Xeon Gold 6240 18C 2.6GHz, NVIDIA Tesla V100, Infiniband EDR	Petróleo Brasileiro S.A
95	66	Advanced Computing System(PreE)	Sugon TC8600, Hygon Dhyana 32C 2GHz, Deep Computing Processor, 200Gb 6D-Torus	Sugon
96	67	CRONOS	BullSequana X, Xeon Platinum 8260 24C 2.4GHz, Infiniband HDR	EDF
97	68	Oakbridge-CX	Fujitsu PRIMERGY CX2550/CX2560 M5, Xeon Platinum 8280 28C 2.7GHz, Intel Omni-Path	Information Technology Center, The University of Tokyo
98	69	SNL/NNSA CTS-1 Manzano	Tundra Extreme Scale, Xeon Platinum 8268 24C 2.9GHz, Intel Omni-Path	Sandia National Laboratories
99	70		Apollo 6500 XL270d Gen10, Xeon Gold 6140 18C 2.3GHz, Mellanox InfiniBand EDR, NVIDIA Tesla V100 SXM2	Research Institute
100		Wilkes-3	PowerEdge XE8545, AMD EPYC 7763 64C 2.45GHz, NVIDIA A100 80GB, Infiniband HDR200 dual rail	University of Cambridge

▲“富岳”超级计算机

广告声明：文内含有的对外跳转链接（包括不限于超链接、二维码、口令等形式），用于传递更多信息，节省甄选时间，结果仅供参考，IT之家所有文章均包含本声明。