Haaretz.com By Israel News
Israel’s official national supercomputer project is launched – at a cost of almost 300 million shekels. There’s only one problem: Those who are supposed to use it say it may be redundant
The first Israeli supercomputer is officially on its way. At least that’s the plan, according to Israel’s so-called “Arrangements Law” – the legislation that serves as the country’s de facto budget.
The law regulates Knesset operations and its passing includes an approval of the first stage of a national program dedicated to artificial intelligence. The AI plan also calls for the creation of an HPC (High Performance Computing) system for Israel. The project is not cheap and is expected to cost the public roughly 290 million shekels (just shy of $88 million).
This is an ambitious project with the potential to advance technological and scientific research and development – but it is also a contentious one. Many people in the academic and tech worlds say such a supercomputer is unnecessary, with more efficient alternatives already available.
The building of a supercomputer is one of the recommendations laid out by a special committee set up on the issue of AI. The committee, headed by computer scientist and entrepreneur Dr. Orna Berry, was established in 2020 by the National Infrastructure Forum for Research and Development, which includes the Planning and Budgeting Committee of the Council for Higher Education, the Innovation Authority, the ministries of defense, science and finance, as well as the National Academy of Science.
According to the plan, these agencies will fund the project. The planned computing infrastructure will have powerful data-crunching capabilities intended mainly for research and development, and will be made available to academicians, the high-tech industry, the defense establishment and the public sector. There are hundreds of such computers around the world, but this is the first time such an initiative has been launched in Israel.
A tender for Israeli company will soon be published and the winner will be charged with managing the project and actually assembling the supercomputer.
Despite its name, a supercomputer is not really a single powerful computer. Rather it is a type of server farm in which all of the different computers serve as components that can act together in synchrony as a single computing unit. Estimates are that it will span 400-500 square meters (4300-5400 square feet) and will be built at one of the server farms in Israel. The committee set a minimal standard of 88,000 CPUs (computational power) and 1,000 GPUs (graphic processing power) – both required for speeding up AI-related tasks. Furthermore, the supercomputer will require powerful servers and communications components.
Giant corporations such as Intel, Nvidia (and Mellanox which it owns), HPE, IBM and others are prominent suppliers of such components for supercomputers around the world. Aviv Zeevi, VP of Technological Infrastructure at the Innovation Authority, estimates that if all progresses as planned, the supercomputer can be up and running by January 2022.
Operating the system will be a company receiving a franchise from the state, a government steering committee and a director of the national plan for artificial intelligence, expected to be nominated soon. At first, the cost of using the supercomputer will be token, followed subsequently by much higher fees which will enable its operators to make money. Over the years, the supercomputer is expected to have its hardware upgraded, covered by current budgets and user fees.
Why does Israel need a supercomputer?
According to the plan’s formulators, a supercomputing infrastructure will enable heavy-duty tasks that cannot be performed by private or local servers to be carried out – or, at least those that can be run on cloud services such as those provided by Amazon, Microsoft and Google – but for a very high cost.
The budget plan says that the supercomputer will support R&D and an innovation environment in Israel which will enable local universities, industry and the defense establishment to work together, running trials for experimental hardware and software components independently of the public cloud systems run by international companies. Israel announced recently that Google and Amazon will operate its official state cloud, known as Project Nimbus.
The computing network will be powerful enough to enable a large number of researchers and developers to run different tasks in tandem, including those that demand especially large volumes of data (such as simulations, computerized models, training of AI systems, genomic analysis, etc).
According to Zeevi, the supercomputer will be able to create a model of artificial intelligence capable of processing and understanding Hebrew as what is termed a “natural language.” Natural language processing, or NLP, is something with a large demand in almost all areas of tech today – and Israel is no different.
“In order to study and understand a language, an enormous amount of texts must be analyzed first. The academic world needs this tool for research, the Defense Ministry will use it for gaining insights from intelligence reports and Israeli companies will use it for constructing smart chatbots,” he says.
Even though consultations were held with a wide range of experts and companies in this area over the last year or two, the company most actively promoting the construction of this computer is Mellanox. Senior company officials have been trying to promote such a project for years and they are now very involved in the process, providing information and active counselling to the Innovation Authority and other involved players. Mellanox and Nvidia are very prominent global suppliers of components for supercomputers. It is reasonable to assume that their components will be integrated into Israel’s supercomputer, making its construction lucrative for them.
Gilad Shiner, senior VP for computing at Nvidia, tells Haaretz that the reasons behind the company’s heavy involvement in the project are not commercial. “We’ve been dealing with supercomputers for years, seeing the research, discoveries and product planning they can accomplish. The entire industry is moving to supercomputers,” he says.
He claims, for example, that the analysis of MRI images, the analysis of the structure of the coronavirus and the prediction of disasters, as well as tasks related to national security, were all successful thanks to supercomputers. “Why don’t we develop drugs in Israel? Because we don’t have a supercomputer, while others do,” adds Shiner.
Researchers vs supercomputers
Nevertheless, not everyone is convinced that a supercomputer is needed in Israel. Some claim that it’s redundant as many companies and researchers currently make do with public cloud services, such as those of Google, Amazon, Microsoft and Oracle, which have capabilities comparable to those of supercomputers. In fact, these companies are expected to build server farms (data centers) in Israel in the coming years (as part of Project Nimbus).
Zeevi argues that cloud infrastructure is not a realistic alternative to a supercomputer due to the high costs of using it, and because a supercomputer provides additional capabilities not provided by cloud services. However, the AI committee’s report from last December includes a recommendation to buy advanced cloud services in addition to the building of a supercomputer, which will be made available to researchers at universities, in industry and in the defense establishment.
If so, then why does the building of a supercomputer raise concerns among academic researchers?
Currently, such researchers use private computing infrastructure located in universities, usually built according to their specific requirements. They worry that in the future, they will not get funding for the establishment or upgrading of this private infrastructure, and will be directed instead to the advanced computing services of the national supercomputer that replaces it. What worries them is the bureaucracy involved in scheduling the time-sharing of the supercomputer, since many organizations will be using the supercomputer in parallel. This includes requests for resources every time they will want to run heavy-duty tasks.
An AI researcher with a background in high-tech and the academic world says that this “supercomputer will impinge on people’s independence and undermine their ability to be dynamic. I once worked in AI labs in the U.S., where we had access to a supercomputer. No one ran anything on it since the processors were outdated, and no one wanted to deal with the bureaucracy of managing the time-sharing for running tasks on it.
“Compelling researchers to work with supercomputers will be disastrous. You can’t depend on an external agent as opposed to your own lab. You just lose all the flexibility over scheduling issues just for accessing the supercomputer. For example, you sometimes need the servers to run continuously for a month. It’s not clear how one can synchronize all the researchers and constraints in the country.”
A further worry raised by people in this field is that supercomputers quickly become outdated. Hardware now considered new and the most advanced there is quickly becomes outdated within a year. Even if the Israeli supercomputer is upgraded and undergoes complete replacement every three years, there will still be a gap that may deter researchers and developers, harming R&D capabilities. In comparison to Chinese universities and tech giants with advanced computing resources, the maintenance may not be worth it. Thus, there are people claiming that it’s best to invest the money elsewhere altogether. This could include the upgrading of local computing systems in universities and research institutes, or providing funding for start-ups and researchers for using public cloud services run by the tech giants.
The people promoting this project are aware of these arguments, admitting that there are discussions and debates around the model of operating this project. Despite this, they claim that concerns are overblown, since ultimately, the supercomputer will serve the R&D needs of universities and industry, as well as other national needs.
“There are dozens of countries with supercomputers. You can’t argue that the need is baseless,” says Zeevi. “Claims that industry can do without it and that there is a misunderstanding of the market here may be true in the short run, but not if you want to create a revolution.”
