The World’s Most Powerful Supercomputer Is an Absolute Beast

The World’s Most Powerful Supercomputer Is an Absolute Beast

https://ift.tt/2HwP8nb

A row of Summit’s server racks.
Photo: Oak Ridge National Laboratory

Behold Summit, a new supercomputer capable of making 200 million billion calculations per second. It marks the first time in five years that a machine from the United States has been ranked as the world’s most powerful.

The specs for this $200 million machine defy comprehension. Built by IBM and Nvidia for the US Department of Energy’s Oak Ridge National Laboratory, Summit is a 200 petaflop machine, meaning it can perform 20 quadrillion calculations per second. That’s about a million times faster than a typical laptop computer. As the the New York Times put it, a human would require 63 billion years to do what Summit can do in a single second. Or as stated by MIT Technology Review, “everyone on Earth would have to do a calculation every second of every day for 305 days to crunch what the new machine can do in the blink of an eye.”

The machine, with its 4,608 servers, 9,216 central processing chips, and 27,648 graphics processors, weighs 340 tons. The system is housed in a 9,250 square-foot room at Oak Ridge National Laboratory’s facility in Tennessee. To keep this machine cool, 4,000 gallons of water are pumped through the system. The 13 megawatts of energy required to power this behemoth could light up over 8,000 US homes.

Summit is now the world’s most powerful supercomputer, and it is 60 percent faster than the previous title holder, China’s Sunway TaihuLight. It’s the first time since 2013 that a US-built computer has held the title, showing the US is keeping up with its main rival in this area, China. Summit is eight times more powerful that Titan, America’s other top-ranked system.

Photo: Oak Ridge National Laboratory

As MIT Technology Review explains, Summit is the first supercomputer specifically designed to handle AI-specific applications, such as machine learning and neural networks. Its thousands of AI-optimized chips, produced by Nvidia and IBM, allow the machine to crunch through hideous amounts of data in search of patterns imperceptible to humans. As noted in an Energy.gov release, “Summit will enable scientific discoveries that were previously impractical or impossible.”

Summit and machines like it can be used for all sorts of processor-heavy applications, such as designing new aircraft, climate modeling, simulating nuclear explosions, creating new materials, and finding causes of disease. Indeed, its potential to help with drug discovery is huge; Summit, for example, could be used to hunt for relationships between millions of genes and cancer. It could also help with precision medicine, in which drugs and treatments are tailored to individual patients.

From here, we can look forward to the next generation of computers, so-called “exascale” computers capable of executing a billion billion (or one quintillion) calculations per second. And we may not have to wait long: The first exascale computers may arrive by the early 2020s.

[Energy.gov, New York Times, MIT Technology Review]

Tech

via Gizmodo http://gizmodo.com

June 8, 2018 at 03:57PM

Surprise, Facebook Reportedly Gave Companies Your Friends’ Data After It Said It Wouldn’t

Surprise, Facebook Reportedly Gave Companies Your Friends’ Data After It Said It Wouldn’t

https://ift.tt/2JvqFjR

Stop me if you’ve heard this one before, but it looks like Facebook may have been sharing more of your data than you thought it was. The Wall Street Journal reported Friday that the social network cut deals with a number of companies to provide access to user records and friend data even after its policy change that prevented apps from scraping that very information.

According to the report, Facebook reached agreements with a number of major corporations to provide data about the friends of its users. The information handed over to companies included details like phone numbers and a metric called “friend link” that determined how much communication and connectivity there was between users and their friends.

Disclosure of the deals punctures a hole in the picture Facebook has tried to paint as a suddenly user-friendly, privacy-minded company after 2014—not that anyone was buying that image anyway. That year, Facebook cut off the significant amount of access app developers could pull from users on the platform, including completely restricting the ability to scrape data from a user’s friends without their consent.

Prior to 2014, Facebook allowed app makers to suck up a significant amount of data from people without their permission—a policy the company has since had to pay for while facing scrutiny over the Cambridge Analytica scandal. The UK-based political data firm acquired information on more than 87 million people collected without consent through a Facebook app.

Facebook supposedly realized the error of its ways and turned off the spigot that allowed friend data to be collected—but as it turns out you could just buy your way back in if you really wanted. Citing “court documents, company officials and people familiar with the matter,” the Wall Street Journal reported that Facebook made deals with Royal Bank of Canada, Nissan—both advertisers on the platform—and Nuance Communications, which was working with Fiat at the time. The agreements were also completely separate from Facebook’s data sharing program with device manufacturers, which the company owned up to earlier this week.

Facebook has copped to the agreements, which it said lasted between just weeks to several months, but tried to minimize them as much as possible. “As we were winding down over the year, there was a small number of companies that asked for short-term extensions, and that, we worked through with them,” Ime Archibong, Facebook’s vice president of product partnerships told WSJ. “But other than that, things were shut down.”

Per the Journal, Facebook internally called the deals “whitelists,” which may be a little bit of insight into how the company viewed the arrangements. Whitelisting someone typically means providing them access that they otherwise wouldn’t have. At the time, Facebook had cut off access to friend data that was previously accessible to app developers using Facebook. These companies were effectively whitelisted to use that data. Access is never truly cut off if you have some money to throw around.

[Wall Street Journal]

Tech

via Gizmodo http://gizmodo.com

June 8, 2018 at 09:27PM