Supercomputers (global market)
Directory Supercomputers (systems and projects) are available on TAdviser.
The main articles are:
2023: Spending on AI solutions for supercomputers in the world increased by 20% over the year to $4.64 billion
In 2023, the cost of solutions AI for supercomputers a global scale reached $4.64 billion. This is about 20% more compared to 2022, when expenses amounted to $3.89 billion. The industry is experiencing significant growth amid the rapid adoption of AI-based applications and services, as well as the increasing need for high performance computing (HPC). Indicators for the global market in early September 2024 were announced by Market Research Future.
The authors of the report point out that AI technologies are increasingly being used by companies and organizations to automate tasks, make informed decisions and gain competitive advantages. This increases the load on hardware platforms, which generates additional demand for AI solutions for supercomputers. Another important factor in the expansion of the market is the rapid progress in AI algorithms and equipment: such platforms are becoming more accurate and efficient, which opens up new opportunities for enterprises. In addition, the deployment of neural networks is facilitated by the emergence of specialized hardware solutions, in particular, tensor processors. Another factor in growth is government initiatives: various government entities around the world recognize the potential of AI and invest heavily in related research, development and infrastructure.
The global industry of AI solutions for supercomputers is segmented into hardware, software and services. At the end of 2023, costs in these areas amounted to $2.56 billion, $1.42 billion and $0.66 billion, respectively. AI equipment is actively acquired by hyperscalers and operators of large data centers. We are talking, in particular, about powerful accelerators based on GPUs and specialized accelerators.
By deployment model, the market is segmented into local, cloud, and hybrid solutions. By 2024, the local sector provides more than 60% of total revenue. At the same time, the cloud segment shows the highest growth rates, as more and more organizations transfer their loads to such sites. The hybrid segment also shows positive dynamics due to the fact that companies combine local resources with cloud services. Among the main areas of application of AI solutions for supercomputers are healthcare, manufacturing, financial services, retail and transport. In 2023, large enterprises generated the largest revenue thanks to significant R&D budgets and the need for HPC solutions. At the same time, the small and medium-sized businesses sector is growing at a faster pace.
North America accounted for the largest share of costs in 2023 - approximately 35.4%: this is due to the presence of advanced technological infrastructure and the concentration of major players. Next, in terms of costs, is Europe, and the Asia-Pacific region closes the top three, which shows the highest growth rates. Significant players in the industry are Google, Cray, AMD, Dell Technologies, Fujitsu, HPE, Atos, Cisco, IBM, Lenovo, Intel, Sugon, Inspur, Nvidia and NEC.
At the end of 2024, revenue in the market under consideration is estimated at $5.53 billion. Market Research Future analysts predict that in the future, the CAGR will be 19.18%. As a result, by 2032, the cost of AI solutions for supercomputers on a global scale will reach $22.5 billion.[1]
2020: Lowest number of new powerful supercomputers since 1993
On November 16, a new list of the top 15 most powerful supercomputers in the world was published, which is updated twice a year. The first place in it for the second time in a row was taken by the Japanese supercomputer Fugaku, the declared real performance of which is 442 Pflops, and the peak is about 537 Pflops. The machine is installed at the RIKEN Center for Computational Sciences.
The rest of the top ten follow him by a wide power margin. Thus, the real performance supercomputer of the Oakridge National Laboratory, USA which occupies the second line of the list, is - 148.6 Pflops, peak - about 200.8 Pflops.
Featured in the top 10, the German supercomputer installed at the Juelich Research Center and the supercomputer created at Saudi Aramco, Saudi Arabia's national oil company, are new to the list.
The compilers of the ranking state that although two new supercomputers were in the top ten systems this time, the total number of new systems on the list turned out to be the lowest since the beginning of its history in 1993. They do not explain what this is connected with, but, probably, it was not without the influence of the epidemic of the new coronavirus infection.
In addition to Japan and the United States, there are systems from China, Germany, Italy and Saudi Arabia in the top ten.
In terms of the number of systems in the entire list, China and the United States are in the lead, between which a serious supercomputer race has unfolded: of these countries, 212 and 113 supercomputers are in the ranking, respectively. Moreover, the US Department of Energy has repeatedly stated that their country must definitely win this race: "whoever leads the world in the field of high-performance computing will have a huge competitive advantage in all sectors of the economy."
About 63% of all systems in the November list are involved in research, according to the summary statistics.
There are two Russian systems in the ranking: the Christofari supercomputer of Sberbank, whose real performance is about 6.7 Pflops, and the Lomonosov-2 supercomputer of Moscow State University with a capacity of about 2.5 Pflops. They are ranked 40th and 158th on the list, respectively.
Over the past few years, Russia has been on the list with 2-5 systems. A year ago, there were three systems on the list from the country due to the Roshydromet supercomputer, whose real capacity is 1.29 Pflops. In November 2020, this was not enough to enter the ranking: the threshold value has since grown and amounted to 1.32 Pflops.
In February 2020, President Vladimir Putin set the government the task of submitting proposals to increase the capacity of Russian supercomputer centers, which should then be included in the national project "Science" and the national program "Digital Economy." And in the spring of 2020, Russian scientists, together with the Ministry of Education and Science, prepared the concept of supercomputer infrastructure in the country. However, as of the end of the year, the development strategy for this area remains unclear.
2018
Lenovo has become the world's largest manufacturer of supercomputers
In June 2018, Lenovo became the world's largest manufacturer of supercomputers. The Chinese company received this status due to the fact that its systems took the most places in the Top500 ranking. Read more here.
Europe to invest €1bn in supercomputers
On January 11, 2017, the European Commission (EC) announced plans to spend 1 billion euros on the development of supercomputers. These investments point to European attempts to catch up with China, the United States and Japan, which dominate the development of high-performance computing systems, Bloomberg reports.
The project involves the development of the financial and legal structure of EuroHPC, within the framework of which the infrastructure of calculations of a high degree of efficiency will be deployed in the EU. EuroHPC will also support a program of research and innovation to develop technology and information technology, as well as software for supercomputers.
 | Through the EuroHPC project, we want to provide European researchers and companies with leading supercomputing capacity by 2020 to develop technologies such as artificial intelligence, as well as to create everyday applications of the future in areas such as healthcare, security and engineering, the European Commission said in a statement. |  |
It is assumed that the EC invests about 486 million euros in the development of supercomputers. About the same amount will be invested by EU member states. In addition, companies can join the initiative and make their "contribution in a non-monetary form."
According to the EC, the infrastructure of supercomputers will not only create new jobs in the EU, but will also allow states to more effectively conserve energy and fight climate change.
The goal of the project is to deploy two world-class supercomputers in Europe by 2020 with a capacity of at least 100 million billion calculations per second, as well as more than two mid-range systems capable of producing tens of millions of billions of calculations per second. The EC called this initiative "critical for EU competitiveness and independence in the data economy."
By early 2018, the two most powerful supercomputers are in China. In addition, the top ten leaders include mainly systems from Japan and the United States.[2]
2017
Historical stage: How the principles of building the largest supercomputers have changed
A new generation of supercomputers focused on working with big data and artificial intelligence is created according to principles that are significantly different from those that were laid down in high-performance cluster systems until 2017.
Detailed in TAdviser review: Supercomputers for artificial intelligence (AI supercomputers)
IBM, Intel, Cray will give out a quarter of a billion dollars from the treasury
Minister power engineering specialists USA Rick Perry announced government plans to invest $258 million in the creation of exaflops (supercomputers computers with a performance of more than 1,000. Pflops or more than 1018 floating point operations per second)[3]
The PathForward program plans to distribute these funds to six of the world's leading technology companies by sponsoring their developments. According to Digital Trends, six companies are now on the list of applicants for state support - AMD, Cray Inc. (specializing in the development of supercomputers, part of Tera Computer), Hewlett Packard Enterprise, IBM, Intel and Nvidia.
The distribution of funds will be carried out through a special division of the Ministry of Energy - Exascale Computing Project. Finance will be allocated to each participating company for three years, the amount of funds allocated should not exceed 40% of the total cost of the project developed by the company.
Research funding is planned primarily in three key areas: hardware technology, programming technology and application development. According to the calculations of American officials, the result of the investment should be the development of an exaflops computer by 2021.
See also Nvidia will receive funding from the US Department of Energy on HPC
2015: Ban on the export of super-PCs with a capacity of 8 teraflops from China
In early August 2015, it became known about the restrictions imposed by the Chinese authorities on the export of supercomputers for reasons of national security.
According to Reuters, citing state news agency Xinhua, from August 15, 2015, it will be prohibited to export computer systems with a computing capacity of more than 8 teraflops from the PRC, which roughly corresponds to the performance of four Sony PlayStation 4 game consoles. For comparison, the Chinese supercomputer Tianhe-2, which by August 2015 is considered the most powerful in the world, is capable of processing 34 trillion operations per second (34 teraflops).
Companies supplying powerful computers abroad will have to receive special licenses that spell out new requirements for exported equipment.
Restrictions on the export of computing solutions in China came after the United States imposed a number of bans on the use of Chinese IT products by American companies. In addition, in April 2015, the US authorities imposed an embargo on the supply of Intel technologies to Chinese firms that ensure the operation of Tianhe-2.
In addition to supercomputers, China has banned the export of unmanned aerial vehicles. From mid-August 2015, drones capable of rising to a height of more than 15 thousand meters and flying for more than one hour will not be taken out of the Celestial Empire. This does not apply to consumer drones with a low flight range.
China fears that remotely controlled high-tech aircraft will fall into the hands of terrorists. Another reason for restricting the removal of drones is the protection of intellectual property rights.[4]
According to Gene Cao, an analyst at Forrester Research, the ban on the export of supercomputers confirms China's global leadership in the production of this type of computer technology and is of great importance to the industry.
2013: HPC market to reach $14 billion in 2015 - IDC
In March 2013, IDC analysts predicted that in 2011-2016. the HPC market as a whole will grow by 7.3% and in 2015 its volume will exceed $14 billion.
2012
Sales of HPC servers - $11.6 billion, growth - 7%
In 2012, supercomputers contributed to record revenues in the high-performance computing (HPC) market, when companies and countries around the world turned to these powerful systems in the hope of gaining computational and economic advantages, IDC analysts say.
According to IDC data published on March 21, 2013, in 2012, revenues from the sale of servers for HPC increased by 7.7% and reached $11.6 billion compared to $10.3 billion in 2011. This happened despite a decrease in sales in unit terms by 6.8%, which indicates an increase in the average sales price, analysts say.
This can largely be attributed to the demand for supercomputers, which are sold at prices ranging from $0.5 million or more. According to IDC, in 2012, revenues from the sale of supercomputers jumped by 29.3%, reaching $5.6 billion. In general, the supercomputer segment accounted for 50.9% of revenues in 2012, and the demand for very large systems continues to grow.
According to IDC, IBM and HP account for over 60% of revenue in the HPC server segment, while Fujitsu and Cray have sharply increased sales.
IDC points to the Fujitsu K system installed at the RIKEN Advanced Institute for Computational Science in Kobe, Japan, which generated over $0.5 billion in total supercomputer revenues. This system, with 705,024 cores in Fujitsu's SPARC64 processors, was the first to cross the 10 Pflops milestone. She was the fastest supercomputer in the world until June 2012, when this title was taken from her by a system called Sequoia, created at IBM.
Six months later, this system was replaced as' number one'by the Cray Titan, a powerful supercomputer XK7 powered by Opteron processors of Advanced Micro Devices Corporation and graphics accelerators manufactured by Nvidia. It is installed at Oakridge National Laboratory (USA, pc. Tennessee).
'HPC servers, especially supercomputers, are closely associated not only with the progress of science, but also with industrial innovation and economic competitiveness, 'said Earl Joseph, IDC Software Vice President of Technical Computing, in a statement. 'For this reason, states and regions around the world are increasing investment in supercomputers even in today's difficult economic conditions'.
However, while 2012 was a very good year for the supercomputer segment, the rate of revenue growth here is likely not to hold steady in the future, IDC Tech Computing Research Vice President Steve Conway said in a statement.
With the growing demand for supercomputers, the interest of manufacturers in developing new technologies for them is also growing. This is evident in everything from the IBM Watson supercomputer to the new Xeon Phi coprocessors, which Intel offers as an alternative to AMD and Nvidia (Nvidia) graphics accelerators used in organizations to improve system performance.
If revenues are growing rapidly in the supercomputer segment, then this cannot be said about other server segments for HPC. In the Division segment (systems valued at 250-499 thousand dollars), revenues in 2012 decreased by 2.2% compared to 2011 and amounted to 1.2 billion dollars. This is 10.9% of revenues from the sale of servers for HPC over the past year.
In the Departmental segment ($100-250 thousand), revenues decreased from $3.5 billion in 2011 to $3 billion in 2012 and accounted for 27% of revenues from the sale of servers for HPC. In the Workgroup segment (up to $100 thousand), revenues increased by 1.2%, to $1.24 billion. According to IDC analysts, this segment was most influenced by the global economic downturn, due to which organizations postponed or canceled purchases.
Systems worth less than $100,000 are purchased using short trading cycles and are not vital. In addition, in some organizations, working groups combine funds allocated to HPC and buy more powerful systems. Nevertheless, analysts believe that sales growth will continue in this segment.
In general, the largest producer is:
- IBM, which accounts for 32% of server revenue for HPC.
- HP's share was 30.8%.
- It is followed by Dell at 13.5%.
- According to IDC, Fujitsu's 2012 revenue grew by 470.5% thanks to the sale of the K system in Japan.
- Cray's revenue jumped 127.3%, and in the IV quarter, Cray's 2012 takeover of Appro[5] had a positive effect on its revenue[5].
'App supermarkets' emerge to access supercomputers in US
In 2012 in the USA a number of initiatives of "democratization" of access to supercomputers is carried out. These projects will give an opportunity to small business companies to resort to high-performance calculations at design of products. In particular, the concept of supercomputer mini-applications (apps) similar mobile appeared, each of which solves separate small problems, for example, flow simulation of liquid in a pipe.
The equipment and ON highly productive calculations can manage in the huge sums. But now it is not required to acquire the universal program system any more. In the Procter & Gamble company which provides the code of some developments in the form of supercomputer mini-applications, are convinced that this concept will inhale new life in the market ON for HPC-SYSTEMS on which the last ten years there is a calm.
The model of supercomputer mini-applications is already used, in particular, on the NanoHub.org portal created by National scientific fund of the USA for developers of nanotechnologies. There now about 250 such programs, mostly public, but with the closed code. Sozdaetsya another analogichnyy portal, ManufacturingHub.com, for proizvodstvennykh companies.
2010: HP and IBM dominate the supercomputer market
World supercomputer manufacturers according to Top500.org. Criterion: number of models produced
World supercomputer manufacturers according to Top500.org. Criterion: performance
The IBM Roadrunner supercomputer continues to be the most powerful supercomputer in the world. IBM built this computer for the US Department of Energy. It is built on a hybrid scheme of 6,500 AMD Opteron dual-core processors and almost 13,000 IBM Cell 8i processors in special TriBlades racks connected using Infiniband. The Roadrunner supercomputer runs Linux and is controlled by xCAT. It covers approximately 1,100 square meters, and weighs 226 tons. Power consumption is 3.9 megawatts. It entered service in June 2008. IBM Roadrunner cost $133 million.
The Roadrunner supercomputer IBM is used to simulate military operations and explosions by the Department of Defense. USA
Supercomputers of the world according to Top500.org
Notes
- ↑ Ai In Supercomputer Market Research Report
- ↑ EU Launches $1.2 Billion Supercomputing Plan, Without U.K
- ↑ IBM, Intel, Cray will give out a quarter of a billion dollars from the treasury for next-generation supercomputers.
- ↑ China to restrict drone, supercomputer exports
- ↑ 5,0 5,1 [http://www.pcweek.ru/themes/detail.php?ID=148834 IDC
| The site content is translated by machine translation software powered by PROMT. The machine-translated articles are not always perfect and may contain errors in vocabulary, syntax or grammar. Read original article If you find inaccuracies or errors in the results of machine translation, please write to editor@tadviser.ru. We will make every effort to correct them as soon as possible. |