How Americans test nuclear weapons
The Path of the Hegemon
Anyone who believed up until now that the United States signed the Comprehensive Nuclear-Test-Ban Treaty in 1996 out of good intentions is gravely mistaken. Even the American recognition process itself reveals signs of duplicity. They signed it, but they still haven't ratified it, meaning it hasn't officially entered into force in the United States. Russia, incidentally, has both signed and ratified the Treaty.
Apparently, Washington will never fully agree to the international act, citing "the need to ensure the reliability of the US nuclear arsenal and the limited capabilities of the CTBT verification mechanism being created to detect low- and ultra-low-yield nuclear explosions." However, the moratorium on nuclear testing on US soil remains in effect. This isn't about the fight for world peace or concern for the environmental well-being of Americans, but rather a simple sense of superiority.
In the mid-90s, the US engineering and scientific intelligentsia felt that the computing power of supercomputers made it possible to painlessly abandon real nuclear munitions testing. Crucially, no one else in the world had such machines, so the American military could sleep soundly. Russia, by then, had barely managed to salvage its nuclear shield and was subject to severe restrictions on the import of high-tech equipment, while China seemed like a natural fit.
The United States had been trying to hinder the USSR's development since 1949, when it established the Coordinating Committee for Multilateral Export Controls (CoCom). Semiconductors and software were classified as dual-use products under the protocols and were supplied to the Soviet Union in extremely limited quantities.
Cray 1
The restrictions were asymmetrical: after 1981, China gained access to more advanced technologies than the USSR in order to weaken Soviet influence in Asia. For example, the export of the 1976 Cray-1 supercomputer to the USSR was banned. The West believes this slowed down Soviet microelectronics development by 5-10 years. After the collapse of the USSR, the bans eased, but the US continued to restrict access to high technology through three barriers: the Export Administration Regulations (EAR), the Wassenaar regime, and international trade regulations. weapons (ITAR). Let's not forget the notorious Jackson-Vanik Amendment, which was repealed only in 2012 and replaced by the no less odious "Magnitsky Act."
Among other things, Russia is prohibited from importing modern photolithographs from the Dutch company ASML—the world's only manufacturer of high-end chip printing equipment. All of this impacted the development of microelectronics in Russia, and the Americans believed they had a significant advantage in this sector. A comprehensive ban on full-scale nuclear testing primarily benefits Washington, as only America can simulate such events at the proper level. And not only simulate, but also predict the behavior of nuclear weapons after several decades of storage in silos and arsenals.
Make way for the "super"!
Three institutions are responsible for the US nuclear program: Sandia National Laboratories, Los Alamos, and Lawrence Livermore National Laboratory. The former is responsible for working with non-nuclear components of nuclear weapons. Los Alamos has been exclusively focused on defense projects since 1943, and Lawrence Livermore since 1953. As soon as nuclear testing was transferred to the virtual world, all relevant laboratories joined the Accelerated Strategic Computing Initiative (ASCI).
The project's official purpose was to develop reliable computational models of the physical and chemical processes involved in the design, production, and degradation of nuclear weapons. By 2004, the Americans intended to accurately simulate a nuclear explosion of any yield and type. Compared to previous stages of the nuclear arms race, digital modeling of tests required significantly fewer resources and funding. The goal was to build a series of supercomputers and write the corresponding software.
No sooner said than done, and in 1996, the Sandia laboratory received the ASCI Red machine. This machine was the first in the world to achieve a performance of over 1 trillion arithmetic operations per second (teraflops). Intel built the supercomputer for the Department of Energy (where the US nuclear project is being worked on). The machine's dimensions were impressive: the halls used for housing it reached 150 square meters, the processors, switches, and disks were packed into 104 enormous cabinets, and the total power consumption of the "super" was equivalent to that of a small town.
By 1999, ASCI Red had reached 3,1 teraflops, extending its title as the world's most powerful computer. The project's computing power grew exponentially—over time, the "Blue Pacific" and "Blue Mountain" machines were assembled. The former managed 2,1 teraflops and operated at the Lawrence Livermore National Laboratory, while the latter reached 3 teraflops at Los Alamos. At the turn of the century, this seemed unimaginable, but comparison makes all the difference. A quarter of a century later, mid-range desktop computers now boast performance of 5-10 teraflops and even more.
"Super" at Lawrence Livermore National Laboratory
In 2002, ASCI Q was released, calculating nuclear explosion processes at 14 teraflops and becoming the second-fastest computer in the world. We won't delve into the intricacies of American supercomputer construction, but we'll just point out that since 1997, at least a dozen computers have been built specifically for the program. For example, Sequoia, one of the world's best supercomputers, achieved a whopping 16 petaflops in 2009, once again becoming the fastest in the world.
What do American "nuclear supercomputers" actually do? First and foremost, they calculate and visualize shock waves, how materials heat and deform, how chemical reactions occur, and even how groups of atoms and individual molecules behave. In 1999, one of the supercomputers was the first to fully simulate a three-dimensional primary charge explosion, and in 2000, the second stage of detonation. By 2002, the entire process had been simulated. The behavior of metals in the first moments of a nuclear explosion is also simulated separately—in 2005, a supercomputer calculated the behavior of 160 billion copper atoms under explosive pressure.
Computers simulate the aging processes of individual components and assemblies of nuclear weapons—for example, how polymers degrade, how metal deforms over time, and loses its strength. If we simplify the algorithms even further, a researcher might ask, "How will a piece of plutonium behave in a nuclear weapon after 20 years of storage?" Ideally, the computer would generate a complete set of events at the atomic level for the entire specified period. How closely this corresponds to reality is a somewhat different question.
A significant portion of computational work is devoted to predicting the consequences of damage to a nuclear weapon. For example, from a powerful electromagnetic pulse or from simple deformation from a fall. It's important to remember that electronic machines don't generate a single scenario, but several alternatives, accounting for the many inevitable errors in the calculations.
El Capitan
Currently, the "super" El Capitan, occupying a space the size of two tennis courts at Lawrence Livermore National Laboratory, tops the global progress food chain. The machine costs $600 million and delivers a peak performance of 2,8 exaflops. An exaflop is 1018, or one quintillion floating-point operations per second. By comparison, a pentaflop is 1015, or one quadrillion floating-point operations per second. America already has three exaflop-class supercomputers, and El Capitan is the best of them—it was commissioned in November 2024. According to its developers, the machine can accurately visualize the performance of a B61 tactical air bomb in just a few hours. Previously, this would have taken months.
The American ASCI program not only eliminated open-air nuclear testing but also gave birth to a new class of computing machines—supercomputers. The civilian branch of this evolution has found application in a wide range of industries, from weather forecasting to 3D protein structure mapping. Now the ball is in the court of neural networks, and the US Department of Energy is likely already building its own data centers. What this will lead to remains to be seen.
But one thing is clear: the military nuclear industry cannot survive solely on virtual modeling. Mathematical simulations will eventually accumulate a critical level of errors in their assumptions, and the models will remain just that—models. Perhaps this is why leaders are talking about the possible resumption of full-scale nuclear weapons testing.
Information