Alphabet's Google released new details about the supercomputers it uses to train its artificial intelligence models, saying the systems are both faster and more power-efficient than comparable systems from Nvidia. From a report: Google has designed its own custom chip called the Tensor Processing Unit, or TPU. It uses those chips for more than 90% of the company's work on artificial intelligence training, the process of feeding data through models to make them useful at tasks such as responding to queries with human-like text or generating images. The Google TPU is now in its fourth generation. Google on Tuesday published a scientific paper detailing how it has strung more than 4,000 of the chips together into a supercomputer using its own custom-developed optical switches to help connect individual machines.

Improving these connections has become a key point of competition among companies that build AI supercomputers because so-called large language models that power technologies like Google's Bard or OpenAI's ChatGPT have exploded in size, meaning they are far too large to store on a single chip. The models must instead be split across thousands of chips, which must then work together for weeks or more to train the model. Google's PaLM model - its largest publicly disclosed language model to date - was trained by splitting it across two of the 4,000-chip supercomputers over 50 days.

The US chip-maker Nvidia has said cryptocurrencies do not "bring anything useful for society" despite the company's powerful processors selling in huge quantities to the sector. From a report: Michael Kagan, its chief technology officer, said other uses of processing power such as the artificial intelligence chatbot ChatGPT were more worthwhile than mining crypto. Nvidia never embraced the crypto community with open arms. In 2021, the company even released software that artificially constrained the ability to use its graphics cards from being used to mine the popular Ethereum cryptocurrency, in an effort to ensure supply went to its preferred customers instead, who include AI researchers and gamers. Kagan said the decision was justified because of the limited value of using processing power to mine cryptocurrencies.

The first version ChatGPT was trained on a supercomputer made up of about 10,000 Nvidia graphics cards. "All this crypto stuff, it needed parallel processing, and [Nvidia] is the best, so people just programmed it to use for this purpose. They bought a lot of stuff, and then eventually it collapsed, because it doesn't bring anything useful for society. AI does," Kagan told the Guardian. "With ChatGPT, everybody can now create his own machine, his own programme: you just tell it what to do, and it will. And if it doesn't work the way you want it to, you tell it 'I want something different.'" Crypto, by contrast, was more like high-frequency trading, an industry that had led to a lot of business for Mellanox, the company Kagan founded before it was acquired by Nvidia. "We were heavily involved in also trading: people on Wall Street were buying our stuff to save a few nanoseconds on the wire, the banks were doing crazy things like pulling the fibres under the Hudson taut to make them a little bit shorter, to save a few nanoseconds between their datacentre and the stock exchange," he said. "I never believed that [crypto] is something that will do something good for humanity. You know, people do crazy things, but they buy your stuff, you sell them stuff. But you don't redirect the company to support whatever it is."

An anonymous reader writes: Last week, we learned that Microsoft spent hundreds of millions of dollars to buy tens of thousands of Nvidia A100 graphics chips so that partner OpenAI could train the large language models (LLMs) behind Bing's AI chatbot and ChatGPT.

Don't have access to all that capital or space for all that hardware for your own LLM project? Nvidia's DGX Cloud is an attempt to sell remote web access to the very same thing. Announced today at the company's 2023 GPU Technology Conference, the service rents virtual versions of its DGX Server boxes, each containing eight Nvidia H100 or A100 GPUs and 640GB of memory. The service includes interconnects that scale up to the neighborhood of 32,000 GPUs, storage, software, and "direct access to Nvidia AI experts who optimize your code," starting at $36,999 a month for the A100 tier.

Meanwhile, a physical DGX Server box can cost upwards of $200,000 for the same hardware if you're buying it outright, and that doesn't count the efforts companies like Microsoft say they made to build working data centers around the technology.

An anonymous reader quotes a report from The Guardian: The UK government is to invest 900 million pounds in a cutting-edge supercomputer as part of an artificial intelligence strategy that includes ensuring the country can build its own "BritGPT". The treasury outlined plans to spend around 900 million pounds on building an exascale computer, which would be several times more powerful than the UK's biggest computers, and establishing a new AI research body. An exascale computer can be used for training complex AI models, but also have other uses across science, industry and defense, including modeling weather forecasts and climate projections. The Treasury said the investment will "allow researchers to better understand climate change, power the discovery of new drugs and maximize our potential in AI.".

An exascale computer is one that can carry out more than one billion billion simple calculations a second, a metric known as an "exaflops". Only one such machine is known to exist, Frontier, which is housed at America's Oak Ridge National Laboratory and used for scientific research -- although supercomputers have such important military applications that it may be the case that others already exist but are not acknowledged by their owners. Frontier, which cost about 500 million pounds to produce and came online in 2022, is more than twice as powerful as the next fastest machine.

The Treasury said it would award a 1 million-pound prize every year for the next 10 years to the most groundbreaking AI research. The award will be called the Manchester Prize, in memory of the so-called Manchester Baby, a forerunner of the modern computer built at the University of Manchester in 1948. The government will also invest 2.5 billion pounds over the next decade in quantum technologies. Quantum computing is based on quantum physics -- which looks at how the subatomic particles that make up the universe work -- and quantum computers are capable of computing their way through vast numbers of different outcomes.

When Microsoft invested $1 billion in OpenAI in 2019, it agreed to build a massive, cutting-edge supercomputer for the artificial intelligence research startup. The only problem: Microsoft didn't have anything like what OpenAI needed and wasn't totally sure it could build something that big in its Azure cloud service without it breaking. From a report: OpenAI was trying to train an increasingly large set of artificial intelligence programs called models, which were ingesting greater volumes of data and learning more and more parameters, the variables the AI system has sussed out through training and retraining. That meant OpenAI needed access to powerful cloud computing services for long periods of time. To meet that challenge, Microsoft had to find ways to string together tens of thousands of Nvidia's A100 graphics chips -- the workhorse for training AI models -- and change how it positions servers on racks to prevent power outages. Scott Guthrie, the Microsoft executive vice president who oversees cloud and AI, wouldn't give a specific cost for the project, but said "it's probably larger" than several hundred million dollars. [...] Now Microsoft uses that same set of resources it built for OpenAI to train and run its own large artificial intelligence models, including the new Bing search bot introduced last month. It also sells the system to other customers. The software giant is already at work on the next generation of the AI supercomputer, part of an expanded deal with OpenAI in which Microsoft added $10 billion to its investment.

"IBM is the latest tech giant to unveil its own "AI supercomputer," this one composed of a bunch of virtual machines running within IBM Cloud," reports the Register: The system known as Vela, which the company claims has been online since May last year, is touted as IBM's first AI-optimized, cloud-native supercomputer, created with the aim of developing and training large-scale AI models. Before anyone rushes off to sign up for access, IBM stated that the platform is currently reserved for use by the IBM Research community. In fact, Vela has become the company's "go-to environment" for researchers creating advanced AI capabilities since May 2022, including work on foundation models, it said.

IBM states that it chose this architecture because it gives the company greater flexibility to scale up as required, and also the ability to deploy similar infrastructure into any IBM Cloud datacenter around the globe. But Vela is not running on any old standard IBM Cloud node hardware; each is a twin-socket system with 2nd Gen Xeon Scalable processors configured with 1.5TB of DRAM, and four 3.2TB NVMe flash drives, plus eight 80GB Nvidia A100 GPUs, the latter connected by NVLink and NVSwitch. This makes the Vela infrastructure closer to that of a high performance compute site than typical cloud infrastructure, despite IBM's insistence that it was taking a different path as "traditional supercomputers weren't designed for AI."

It is also notable that IBM chose to use x86 processors rather than its own Power 10 chips, especially as these were touted by Big Blue as being ideally suited for memory-intensive workloads such as large-model AI inferencing.
Thanks to Slashdot reader guest reader for sharing the story.

sciencehabit shares a report from Science Magazine: Fifty years ago today, astronauts aboard Apollo 17, NASA's last crewed mission to the Moon, took an iconic photograph of our planet. The image became known as the Blue Marble -- the first fully illuminated picture of Earth, in color, taken by a person. Now, scientists have re-created that image during a test run of a cutting-edge digital climate model. The model can simulate climatic phenomena, such as storms and ocean eddies, at 1-kilometer resolution, as much as 100 times sharper than typical global simulations.

To re-create the swirling winds of the Blue Marble -- including a cyclone over the Indian Ocean -- the researchers fed weather records from 1972 into the supercomputer-powered software. The resulting world captured distinctive features of the region, such as upwelling waters off the coast of Namibia and long, reedlike cloud coverage. Experts say the stunt highlights the growing sophistication of high-resolution climate models. Those are expected to form the core of the European Union's Destination Earth project, which aims to create a 'digital twin' of Earth to better forecast extreme weather and guide preparation plans.

The next wave of Moore's Law will rely on a developing concept called system technology co-optimization, Ann B. Kelleher, general manager of technology development at Intel told IEEE Spectrum in an interview ahead of her plenary talk at the 2022 IEEE Electron Device Meeting. From a report: "Moore's Law is about increasing the integration of functions," says Kelleher. "As we look forward into the next 10 to 20 years, there's a pipeline full of innovation" that will continue the cadence of improved products every two years. That path includes the usual continued improvements in semiconductor processes and design, but system technology co-optimization (STCO) will make the biggest difference. Kelleher calls it an "outside-in" manner of development. It starts with the workload a product needs to support and its software, then works down to system architecture, then what type of silicon must be within a package, and finally down to the semiconductor manufacturing process. "With system technology co-optimization, it means all the pieces are optimized together so that you're getting your best answer for the end product," she says.

STCO is an option now in large part because advanced packaging, such as 3D integration, is allowing the high-bandwidth connection of chiplets -- small, functional chips -- inside a single package. This means that what would once be functions on a single chip can be disaggregated onto dedicated chiplets, which can each then be made using the most optimal semiconductor process technology. For example, Kelleher points out in her plenary that high-performance computing demands a large amount of cache memory per processor core, but chipmaker's ability to shrink SRAM is not proceeding at the same pace as the scaling down of logic. So it makes sense to build SRAM caches and compute cores as separate chiplets using different process technology and then stitch them together using 3D integration. A key example of STCO in action, says Kelleher, is the Ponte Vecchio processor at the heart of the Aurora supercomputer. It's composed of 47 active chiplets (as well as 8 blanks for thermal conduction). These are stitched together using both advanced horizontal connections (2.5 packaging tech) and 3D stacking. "It brings together silicon from different fabs and enables them to come together so that the system is able to perform against the workload that it's designed for," she says.

Nvidia and Microsoft announced Wednesday a multi-year collaboration to build an AI supercomputer in the cloud, adding tens of thousands of Nvidia GPUs to Microsoft Azure. ZDNet: The new agreement makes Azure the first public cloud to incorporate Nvidia's full AI stack -- its GPUs, networking, and AI software. By beefing up Azure's infrastructure with Nvidia's full AI suite, more enterprises will be able to train, deploy, and scale AI -- including large, state-of-the-art models. "AI technology advances as well as industry adoption are accelerating," Manuvir Das, Nvidia's VP of enterprise computing, said in a statement. "The breakthrough of foundation models has triggered a tidal wave of research, fostered new startups, and enabled new enterprise applications."

Intel's latest plan to ward off rivals from high-performance computing workloads involves a CPU with large stacks of high-bandwidth memory and new kinds of accelerators, plus its long-awaited datacenter GPU that will go head-to-head against Nvidia's most powerful chips. From a report: After multiple delays, the x86 giant on Wednesday formally introduced the new Xeon CPU family formerly known as Sapphire Rapids HBM and its new datacenter GPU better known as Ponte Vecchio. Now you will know them as the Intel Xeon CPU Max Series and the Intel Data Center GPU Max Series, respectively, which were among the bevy of details shared by Intel today, including performance comparisons. These chips, set to arrive in early 2023 alongside the vanilla 4th generation Xeon Scalable CPUs, have been a source of curiosity within the HPC community for years because they will power the US Department of Energy's long-delayed Aurora supercomputer, which is expected to become the country's second exascale supercomputer and, consequently, one of the world's fastest.

In a briefing with journalists, Jeff McVeigh, the head of Intel's Super Compute Group, said the Max name represents the company's desire to maximize the bandwidth, compute and other capabilities for a wide range of HPC applications, whose primary users include governments, research labs, and corporations. McVeigh did admit that Intel has fumbled in how long it took the company to commercialize these chips, but he tried to spin the blunders into a higher purpose. "We're always going to be pushing the envelope. Sometimes that causes us to maybe not achieve it, but we're doing that in service of helping our developers, helping the ecosystem to help solve [the world's] biggest challenges," he said. [...] The Xeon Max Series will pack up to 56 performance cores, which are based on the same Golden Cove microarchitecture features as Intel's 12th-Gen Core CPUs, which debuted last year. Like the vanilla Sapphire Rapids chips coming next year, these chips will support DDR5, PCIe 5.0 and Compute Express Link (CXL) 1.1, which will enable memory to be directly attached to the CPU over PCIe 5.0.

The NOEMA radio telescope, located on the Plateau de Bure in the French Alps, is now equipped with twelve antennas, making it the most powerful radio telescope of its kind in the northern hemisphere. Phys.Org reports: Eight years after the inauguration of the first NOEMA antenna in 2014, the large-scale European project is now complete. Thanks to its twelve 15-meter antennas, which can be moved back and forth on a specially developed rail system up to a distance of 1.7 kilometers long, NOEMA is a unique instrument for astronomical research. The telescope is equipped with highly sensitive receiving systems that operate close at the quantum limit. During observations, the observatory's twelve antennas act as a single telescope -- a technique called interferometry. After all the antennas have been pointed towards one and the same region of space, the signals they receive are combined with the help of a supercomputer. Their detailed resolution then corresponds to that of a huge telescope whose diameter is equal to the distance between the outermost antennas.

The respective arrangement of the antennas can extend over distances from a few hundred meters to 1.7 kilometers. The network thus functions like a camera with a variable lens. The further apart the antennas are, the more powerful is the zoom: the maximum spatial resolution of NOEMA is so high that it would be able to detect a mobile phone at a distance of over 500 kilometers. NOEMA is one of the few radio observatories worldwide that can simultaneously detect and measure a large number of signatures -- i.e., "fingerprints" of molecules and atoms. Thanks to these so-called multi-line observations, combined with high sensitivity, NOEMA is a unique instrument for investigating the complexity of cold matter in interstellar space as well as the building blocks of the university. With NOEMA, over 5,000 researchers from all over the world study the composition and dynamics of galaxies as well as the birth and death of stars, comets in our solar system or the environment of black holes. The observatory captures light from cosmic objects that has traveled to Earth for more than 13 billion years. NOEMA has "observed the most distant known galaxy, which formed shortly after the Big Bang," notes the report. It also "measured the temperature of the cosmic background radiation at a very early stage of the universe, a scientific first that should make it possible to trace the effects of dark energy driving the universe apart."

One piece of news from Tesla's AI Day presentation on Friday that was overshadowed by the company's humanoid "Optimus" robot and Dojo supercomputer was the improvements to Tesla's Full Self Driving software. According to Autopilot director Ashok Elluswamy, "there are now 160,000 customers running the beta software, compared to 2,000 from this time last year," reports The Verge. From the report: In total, Tesla says there have been 35 software releases of FSD. In a Q&A at the end of the presentation, Musk made another prediction -- he's made a few before -- that the technology would be ready for a worldwide rollout by the end of this year but acknowledged the regulatory and testing hurdles that remained before that happens. Afterward, Tesla's tech lead for Autopilot motion planning, Paril Jain, showed how FSD has improved in specific interactions and can make "human-like" decisions. For example, when a Tesla makes a left turn into an intersection, it can choose a trajectory that doesn't make close calls with obstacles like people crossing the street.

It's known that every Tesla can provide datasets to build the models that FSD uses, and according to Tesla's engineering manager Phil Duan, now Tesla will start building and processing detailed 3D structures from that data. They said the cars are also improving decision-making in different environmental situations, like night, fog, and rain. Tesla trains the company's AI software on its supercomputer, then feeds the results to customers' vehicles via over-the-air software updates. To do this, it processes video feeds from Tesla's fleet of over 1 million camera-equipped vehicles on the road today and has a simulator built in Unreal Engine that is used to improve Autopilot.

An anonymous reader quotes a report from Electrek: Tesla has unveiled its latest version of its Dojo supercomputer and it's apparently so powerful that it tripped the power grid in Palo Alto. Dojo is Tesla's own custom supercomputer platform built from the ground up for AI machine learning and more specifically for video training using the video data coming from its fleet of vehicles. [...] Last year, at Tesla's AI Day, the company unveiled its Dojo supercomputer, but the company was still ramping up its effort at the time. It only had its first chip and training tiles, and it was still working on building a full Dojo cabinet and cluster or "Exapod." Now Tesla has unveiled the progress made with the Dojo program over the last year during its AI Day 2022 last night.

The company confirmed that it managed to go from a chip and tile to now a system tray and a full cabinet. Tesla claims it can replace 6 GPU boxes with a single Dojo tile, which the company claims costs less than one GPU box. There are 6 of those tiles per tray. Tesla says that a single tray is the equivalent of "3 to 4 fully-loaded supercomputer racks." The company is integrating its host interface directly on the system tray to create a big full host assembly. Tesla can fit two of these system trays with host assembly into a single Dojo cabinet. That's pretty much where Tesla is right now as the automaker is still developing and testing the infrastructure needed to put a few cabinets together to create the first "Dojo Exapod."

Bill Chang, Tesla's Principal System Engineer for Dojo, said: "We knew that we had to reexamine every aspect of the data center infrastructure in order to support our unprecedented cooling and power density." They had to develop their own high-powered cooling and power system to power the Dojo cabinets. Chang said that Tesla tripped their local electric grid's substation when testing the infrastructure earlier this year: "Earlier this year, we started load testing our power and cooling infrastructure and we were able to push it over 2 MW before we tripped our substation and got a call from the city." Tesla released the main specs of a Dojo Exapod: 1.1 EFLOP, 1.3 TB SRAM, and 13 TB high-bandwidth DRAM.

High-profile universities and state-run research institutes in China have been relying on a U.S. computing chip to power their artificial intelligence (AI) technology but whose export to the country Washington has now restricted, a Reuters review showed. From the report: U.S. chip designer Nvidia last week said U.S. government officials have ordered it to stop exporting its A100 and H100 chips to China. Local peer Advanced Micro Devices also said new licence requirements now prevent export to China of its advanced AI chip MI250. The development signalled a major escalation of a U.S. campaign to stymie China's technological capability as tension bubbles over the fate of Taiwan, where chips for Nvidia and almost every other major chip firm are manufactured.

China views Taiwan as a rogue province and has not ruled out force to bring the democratically governed island under its control. Responding to the restrictions, China branded them a futile attempt to impose a technology blockade on a rival. A Reuters review of more than a dozen publicly available government tenders over the past two years indicated that among some of China's most strategically important research institutes, there is high demand - and need - for Nvidia's signature A100 chips. Tsinghua University, China's highest-ranked higher education institution globally, spent over $400,000 last October on two Nvidia AI supercomputers, each powered by four A100 chips, one of the tenders showed. In the same month, the Institute of Computing Technology, part of top research group, the Chinese Academy of Sciences (CAS), spent around $250,000 on A100 chips. The school of artificial intelligence at a CAS university in July this year also spent about $200,000 on high-tech equipment including a server partly powered by A100 chips. In November, the cybersecurity college of Guangdong-based Jinan University spent over $93,000 on an Nvidia AI supercomputer, while its school of intelligent systems science and engineering spent almost $100,000 on eight A100 chips just last month. Less well-known institutes and universities supported by municipal and provincial governments, such as in Shandong, Henan and Chongqing, also bought A100 chips, the tenders showed.

Long-time Slashdot reader swell shared Scientific American's report on the quest for neuromorphic chips: The human brain is an amazing computing machine. Weighing only three pounds or so, it can process information a thousand times faster than the fastest supercomputer, store a thousand times more information than a powerful laptop, and do it all using no more energy than a 20-watt lightbulb. Researchers are trying to replicate this success using soft, flexible organic materials that can operate like biological neurons and someday might even be able to interconnect with them. Eventually, soft "neuromorphic" computer chips could be implanted directly into the brain, allowing people to control an artificial arm or a computer monitor simply by thinking about it.

Like real neurons — but unlike conventional computer chips — these new devices can send and receive both chemical and electrical signals. "Your brain works with chemicals, with neurotransmitters like dopamine and serotonin. Our materials are able to interact electrochemically with them," says Alberto Salleo, a materials scientist at Stanford University who wrote about the potential for organic neuromorphic devices in the 2021 Annual Review of Materials Research. Salleo and other researchers have created electronic devices using these soft organic materials that can act like transistors (which amplify and switch electrical signals) and memory cells (which store information) and other basic electronic components.

The work grows out of an increasing interest in neuromorphic computer circuits that mimic how human neural connections, or synapses, work. These circuits, whether made of silicon, metal or organic materials, work less like those in digital computers and more like the networks of neurons in the human brain.... An individual neuron receives signals from many other neurons, and all these signals together add up to affect the electrical state of the receiving neuron. In effect, each neuron serves as both a calculating device — integrating the value of all the signals it has received — and a memory device: storing the value of all of those combined signals as an infinitely variable analog value, rather than the zero-or-one of digital computers.

For decades, you could test a computer chip's mettle by how small and tightly packed its electronic circuitry was. Now Intel believes another dimension is as big a deal: how artfully a group of such chips can be packaged into a single, more powerful processor. From a report: At the Hot Chips conference Monday, Intel Chief Executive Pat Gelsinger will shine a spotlight on the company's packaging prowess. It's a crucial element to two new processors: Meteor Lake, a next-generation Core processor family member that'll power PCs in 2023, and Ponte Vecchio, the brains of what's expected to be the world's fastest supercomputer, Aurora.

"Meteor Lake will be a huge technical innovation," thanks to how it packages, said Real World Tech analyst David Kanter. For decades, staying on the cutting edge of chip progress meant miniaturizing chip circuitry. Chipmakers make that circuitry with a process called photolithography, using patterns of light to etch tiny on-off switches called transistors onto silicon wafers. The smaller the transistors, the more designers can add for new features like accelerators for graphics or artificial intelligence chores. Now Intel believes building these chiplets into a package will bring the same processing power boost as the traditional photolithography technique.

Google has been challenged by an algorithm that could solve a problem faster than its Sycamore quantum computer, which it used in 2019 to claim the first example of "quantum supremacy" -- the point at which a quantum computer can complete a task that would be impossible for ordinary computers. Google concedes that its 2019 record won't stand, but says that quantum computers will win out in the end. From a report: Sycamore achieved quantum supremacy in a task that involves verifying that a sample of numbers output by a quantum circuit have a truly random distribution, which it was able to complete in 3 minutes and 20 seconds. The Google team said that even the world's most powerful supercomputer at the time, IBM's Summit, would take 10,000 years to achieve the same result. Now, Pan Zhang at the Chinese Academy of Sciences in Beijing and his colleagues have created an improved algorithm for a non-quantum computer that can solve the random sampling problem much faster, challenging Google's claim that a quantum computer is the only practical way to do it. The researchers found that they could skip some of the calculations without affecting the final output, which dramatically reduces the computational requirements compared with the previous best algorithms. The researchers ran their algorithm on a cluster of 512 GPUs, completing the task in around 15 hours. While this is significantly longer than Sycamore, they say it shows that a classical computer approach remains practical.

"A new supercomputer called Frontier has been widely touted as the world's first exascale machine — but was it really?"

That's the question that long-time Slashdot reader MattSparkes explores in a new article at New Scientist... Although Frontier, which was built by the Oak Ridge National Laboratory in Tennessee, topped what is generally seen as the definitive list of supercomputers, others may already have achieved the milestone in secret....

The definitive list of supercomputers is the Top500, which is based on a single measurement: how fast a machine can solve vast numbers of equations by running software called the LINPACK benchmark. This gives a value in float-point operations per second, or FLOPS. But even Jack Dongarra at Top500 admits that not all supercomputers are listed, and will only feature if its owner runs the benchmark and submits a result. "If they don't send it in it doesn't get entered," he says. "I can't force them."

Some owners prefer not to release a benchmark figure, or even publicly reveal a machine's existence. Simon McIntosh-Smith at the University of Bristol, UK points out that not only do intelligence agencies and certain companies have an incentive to keep their machines secret, but some purely academic machines like Blue Waters, operated by the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign, are also just never entered.... Dongarra says that the consensus among supercomputer experts is that China has had at least two exascale machines running since 2021, known as OceanLight and Tianhe-3, and is working on an even larger third called Sugon. Scientific papers on unconnected research have revealed evidence of these machines when describing calculations carried out on them.

McIntosh-Smith also believes that intelligence agencies would rank well, if allowed. "Certainly in the [US], some of the security forces have things that would put them at the top," he says. "There are definitely groups who obviously wouldn't want this on the list."

The United States is on top of the supercomputing world in the Top500 ranking of the most powerful systems. From a report: The Frontier system from Oak Ridge National Laboratory (ORNL) running on AMD EPYC CPUs took first place from last year's champ, Japan's ARM A64X Fugaku system. It's still in the integration and testing process at the ORNL in Tennessee, but will eventually be operated by the US Air Force and US Department of Energy. Frontier, powered by Hewlett Packard Enterprise's (HPE) Cray EX platform, was the top machine by a wide margin, too. It's the first (known) true exascale system, hitting a peak 1.1 exaflops on the Linmark benchmark. Fugaku, meanwhile, managed less than half that at 442 petaflops, which was still enough to keep it in first place for the previous two years. Frontier was also the most efficient supercomputer, too. Running at just 52.23 gigaflops per watt, it beat out Japan's MN-3 system to grab first place on the Green500 list. "The fact that the world's fastest machine is also the most energy efficient is just simply amazing," ORNL lab director Thomas Zacharia said at a press conference.

Russia is adapting to a world where it no longer has access to many technologies abroad with the development of a new supercomputer platform that can use foreign x86 processors such as Intel's in combination with the country's homegrown Elbrus processors. The Register reports: The new supercomputer reference system, dubbed "RSK Tornado," was developed on behalf of the Russian government by HPC system integrator RSC Group, according to an English translation of a Russian-language press release published March 30. RSC said it created RSK Tornado as a "unified interoperable" platform to "accelerate the pace of important substitution" for HPC systems, data processing centers and data storage systems in Russia. In other words, the HPC system architecture is meant to help Russia quickly adjust to the fact that major chip companies such as Intel, AMD and TSMC -- plus several other technology vendors, like Dell and Lenovo -- have suspended product shipments to the country as a result of sanctions by the US and other countries in reaction to Russia's invasion of Ukraine.

RSK Tornado supports up to 104 servers in a rack, with the idea being to support foreign x86 processors (should they come available) as well as Russia's Elbrus processors, which debuted in 2015. The hope appears to be the ability for Russian developers to port HPC, AI and big data applications from x86 architectures to the Elbrus architecture, which, in theory, will make it easier for Russia to rely on its own supply chain and better cope with continued sanctions from abroad. RSK Tornado systems software is RSC proprietary and is currently used to orchestrate supercomputer resources at the Interdepartmental Supercomputer Center of the Russian Academy of Sciences, St Petersburg Polytechnic University and the Joint Institute for Nuclear Research. RSC claims to have also developed its own liquid-cooling system for supercomputers and data storage systems, the latter of which can use Elbrus CPUs too.