Intel readies chip to rival NVIDIA for machine learning

After abandoning its own GPU for supercomputers, machine learning, and video games in 2009, Intel has returned to the market with a new 72-core Xeon Phi, to compete with NVIDIA’s growing portfolio of GPUs.

The Xeon Phi ‘Knights Landing’ chip, announced at the International Supercomputing Conference in Frankfurt, Germany last week, is Intel’s most powerful and expensive chip to date and is aimed at machine learning and supercomputers, two areas where Nvidia’s GPUshave flourished.

Inside the chip there is 72-cores running at 1.5GHz, alongside 16GB of integrated stacked memory. The chip supports up to 384GB of DDR4 memory, making it immensely scalable for machine learning programs.

At the conference, Intel mentioned some of the issues with GPUs for complex machine learning programs. It believes that the Xeon Phi, which is a byproduct of the failed GPU in 2009 called Larrabee, is an answer to some of those problems.

Intel has already deployed the new Xeon Phi chip to several supercomputers including the Stampede 2, an 18-petaflop supercomputer that ranks in the top ten worldwide.

While supercomputers are definitely the focus, Intel does see the Xeon Phi chips being utilized by machine learning and artificial intelligence developers. Servers are another area where Intel is looking into, though the $6,294 price tag may put some companies off.

Intel could have an edge over Google

Google has built its own chip, a Tensor Processing Unit (TPU), specifically for machine learning and deep neural networks. While it may be hard for a general chip that doesn’t offer specific functionality to compete with a chip designed for machine learning, some developers may choose Intel, a neutral company, over Google who may be a competitor in the machine learning and AI markets.

Intel has plans to launch an even faster version of the Xeon Phi, clearly showing a commitment to the market that they didn’t have in 2009. Hopefully, if the first Xeon Phi doesn’t succeed, it won’t back away from those plans and give the market entirely to NVIDIA.

China made the best new supercomputer and it uses Chinese chips

There's a new supercomputer in town, and it has topped every other supercomputer on the planet.

The new Chinese supercomputer, dubbed the Sunway TaihuLight, was in the first slot on the Top500 list of supercomputers Monday.

The fact that the Sunway TaihuLight is in China isn't anything special — the previous and now second-best supercomputer on the list is also Chinese. But that one, the Tianhe-2, used chips from Intel, a U.S. company. 

The Sunway TaihuLight breaks the mold by using all-Chinese architecture, and it hits some impressive benchmarks in doing so.

As a basis for comparison, Intel's new 10-core processor for home computing sounds positively puny next to the TaihuLight's 10.65 million cores, which is considerably higher than the 560,000 cores of the U.S.'s top supercomputer — Titan, a Cray XK7 system. The Titan is located at the Department of Energy’s Oak Ridge National Laboratory and is currently ranked third in the Top500. 

With 10.65 million cores, the new biggest supercomputer in town hits 93 petaflops, or 93 quadrillion calculations per second.

And with that kind of power, the TaihuLight — on its own — comprises 16.4 percent of the total computing performance on the list, against all 499 other entrants.

Not only does China's new supercomputer have impressive processing power, it's also surprisingly energy-efficient. A lot of power is expended on RAM, and the TaihuLight uses a smaller amount of RAM than less-powerful computers.

This is also the first time that the U.S. does not have the most supercomputers on the Top500 list. China now leads the list with 167 entries, including the top two, while the U.S. is in second place with 165 computers.

It's possible that the U.S. could bounce back, as Intel is expected to announce a new high-level chip at the 2016 International Supercomputer Conference this week.