Intel says its supercomputing chips won't suffer Larrabee's fate
- 17 November, 2014 23:03
Intel introduced its latest Xeon Phi chip Monday, in what seems to be an effort to prove that its supercomputing chips aren't just a flash in the pan.
The chip, code-named Knights Hill, will go into some of the fastest supercomputers in the world. Intel wasn't forthcoming on more product details or release dates, but the seemingly premature announcement was triggered by speculation about Xeon Phi's future.
"We've got to prove it's not a one-generation or two-generation [chip], but a long-term commitment to Knights [products]," said Charlie Wuischpard, vice president and general manager of Intel's workstations and high-performance computing group.
Xeon Phi chips are Intel's fastest, using some of the company's latest technologies. The chip was a by-product of the disastrous Larrabee chip, which was meant to be Intel's first graphics product but was ultimately scrapped in 2009 after multiple delays.
Intel has dealt with chip troubles in the past, but has been reluctant to admit defeat. One such chip was the recent Broadwell, which is on its way to a shorter-than-expected life after production delays. Another chip likely headed for the chopping block is Itanium, which has limited visibility, and is now being phased out by server and software makers.
Knights Hill will be Intel's third supercomputing chip to appear in commercial supercomputers. Intel's first supercomputing chip was a prototype called Knights Ferry, which was released in 2010 to some server makers and academic institutions for testing. That was followed by the first Xeon Phi processor code-named Knights Corner, which shipped in 2012 and appeared in petaflop supercomputers. The second Xeon Phi chip, code-named Knights Landing, will appear in supercomputers starting next year.
Compared to its predecessors, Knights Hill has more cores, cache and threads to run applications faster, Wuischpard said.
The chip will be made using the 10-nanometer process and will perform better while drawing less power, Wuischpard said. Intel's first 10-nm chips will be out next year or in 2016. They will be used first in PCs and mobile devices before making their way into servers and supercomputers.
Knights Hill solves a major supercomputing problem of trying to scale performance while reducing power consumption. Intel has been making advances in bandwidth, storage and memory to balance computing resources, with those advancements continuing in Knights Hill, Wuischpard said.
"It's trying to do a lot more of the same," Wuischpard said, reflecting back to advances in previous Knights chips, which had cutting-edge technologies and were made using the latest manufacturing processes.
The company has made selective Knights Hill disclosures.
"There are a number of projects under discussion with partners regarding Knights Hill," Wuischpard said.
Also, Knights Hill will help Intel reach the milestone of making an exascale supercomputer by 2020 to 2022, Wuischpard said.
Beyond processors, Intel is making other advances in high-performance computing. Intel will facilitate faster communication between computing units and storage in supercomputers by using light as a data-transfer mechanism. Intel will also use new forms of memory such as DDR4 and Hybrid Memory Cube to speed up task execution and data throughput.
Knights chips compete with graphics processors from Nvidia, which are also used in supercomputers to speed up scientific and math applications.