Mediaboss Marketing

Sandia National Laboratory Computer Annex conducts the hourly walk-through of the Thunderbird supercomputer at 2 a.m.

WASHINGTON: The Pentagon recently completed a $68 million acquisition of two new supercomputing platforms and related technical services that rank among its most powerful supercomputers ever and will be among the top 100 performers globally.

These are significant assets, Kevin Newmeyer, deputy director of the Defense Departments High Performance Computing Modernization Program (HPCMP), told Breaking Defense. They bring to us an increase in our computing capacity and the latest advanced chips for artificial intelligence work and storage to support applications of both computational and machine learning concepts within the same computer that we hope will deliver products and services to the warfighter faster.

Its the HPCMPs job to give DoD military and civilian as well as defense contractor scientists, engineers, and technologists access to such supercomputers to solve some of the militarys most computationally complex problems.

The problems range from climate/weather/ocean modeling and simulation, space/astrophysical sciences, and acoustics to signal/image processing, data/decision analytics, and electronics, networks, and C4I systems. Newmeyer said the most common use case is computational fluid dynamics, which is required for making complicated calculations in areas such as aircraft and ship design and engineering.

For the latest acquisition, the Pentagon chose Penguin Computings TrueHPC supercomputing platform. The two new supercomputers, according to the company, will provide DoD with a combined total of over 365,000 cores, more than 775 terabytes of memory, and a total of 47 petabytes of high-performance storage, including over 5 petabytes of high-performance flash storage.

Thats about 150,000 computers all stacked together, operating as one thing, Newmeyer said. If you laid them end to end, you would work your way pretty much across the country.

What does all that compute power get you? An additional 17.6 petaFLOPS, in total. FLOPS or floating point operations per second are the standard measure of a supercomputers performance. FLOPS are determined by how many real numbers a computer can process per second while accounting for the trade-off between range and precision of calculations.

FLOPS are a measure of computational power for solving computer-based problems. Its the horsepower of a machine, Penguins Vice President of Federal Sales Tom Ireland told Breaking Defense.

PetaFLOPS number one quadrillion (1,000,000,000,000,000). To put that in perspective, HPCMP currently has a total capacity across all of its supercomputers of approximately 100 petaFLOPS, according to Newmeyer. That includes the Navys most powerful (known) supercomputer, Narwhal, which is capable of 12.8 petaFLOPS. The known part of the Air Forces most powerful supercomputer, Mustang, is capable of 4.87 petaFLOPS. (Part of Mustang is classified, Newmeyer noted.) Penguins two TrueHPC supercomputers expected to register at 8.5 petaFLOPS and 9 petaFLOPS will be two of HPCMPs most powerful computers ever, Ireland said.

According to the Top500 Project, the fastest supercomputer in the world, as of June 2021, is Japans Fugaku, which registered 442.01 petaFLOPS in November 2020, taking the top spot from IBMs Summit (148.6 petaFLOPS), which is housed at the Department of Energys Oak Ridge National Laboratory.

The Pentagons upgrade in supercomputing power comes amid an intense technological race against near-peer rival China. According to the Top500, China currently leads the world in the total number of supercomputers with 188, but when ranked by performance, the US has five of the top 10 most powerful supercomputers in the world, while China has two of the top 10. No other country has more than one in the top 10.

Ireland noted that Penguin, which has been building supercomputers for 20 years, has for years been running programs at the Department of Energy, which has the most powerful (known) supercomputers in the US. Fifteen of Penguins debuts over 20 years have made the Top500, and were DoD to run official benchmarks on these two new supercomputers, they would rank within the top 100 worldwide, Ireland said.

The Navys DoD Supercomputing Resource Center (DSRC) at Stennis Space Center in Mississippi will house one of the new platforms, while the other will go to the Air Force Research Labs DSRC at Wright-Patterson Air Force Base in Dayton, Ohio.

But first Penguin has to build, deploy, and integrate them into HPCMPs network, known as the Defense Research Engineering Network (DREN). Ireland said Penguins TrueHPC consists of about 1,500 nodes, which must be engineered to work as one, giant machine.

The trick with distributed computing meaning its taking what heretofore was done on a mainframe-style computer where its all on a board, and its broken up into separate, discrete servers is making sure that is an adequate platform for any given application, Penguins Chief Strategy Officer Matt Jacobs told Breaking Defense. To make sure that balance between the elements is right and theres an appropriate amount of compute to solve the problem.

Jacobs said some of the key elements include data patterns, network traffic, and storage capacity, which all must be brought together in a way that doesnt strand investment in any given element of those resources and that its an effective production platform for the workload application. Thats really the art, he added.

Jacobs said that Penguin generally builds these types of platforms in a couple of months, but like many companies worldwide, Penguin has encountered challenges in the global supply chain, especially around chips. Jacobs and Ireland said the supply chain hiccups are beyond the companys control, but said they still wouldnt significantly delay the project.

Notably, the platforms will include over 100 NVIDIA graphics processing units, or GPUs, to bolster DoDs AI and machine learning capabilities, Ireland said.

Ultimately, Ireland said, the project is about keeping the US warfighter equipped with state-of-the-art technologies to solve compute problems. Were keeping our warfighters current. You dont want them fighting wars with F-14s when theres F-22s.

Its unclear how long the era of supercomputers will last, as the US and China, among others, race ahead towards quantum computing, which uses quantum mechanics to make a technological leap in processing power. But Newmeyer said hes not concerned traditional supercomputing platforms will become obsolete anytime soon.

Youll still have a use for these types of machines, he said. Any quantum computer built in the near future is going to be highly expensive to operate, and [quantum computers] are only more useful for certain applications maybe in some stuff around hypersonics, certainly cryptology, navigation there quantum has a key role. But for general computation, [quantum] is an awful lot of money.

Read the rest here:
DoD Buys Two New Supercomputers That Rank Among Its Most Powerful Ever - Breaking Defense