Intel shows off its Knights Corner one teraflop chip

Intel has developed an accelerator chip capable of running at speeds of one teraflop, equal to one trillion calculations per second.

The firm showed off the chip, dubbed Knights Corner, on a test machine at a supercomputing conference in Seattle.

Computer power on this scale is used to solve a range of problems in fields such as weather forecasting, molecular modelling and car crash simulations.

The chip pits Intel against rival add-on processors from Nvidia and AMD.

The Knights Corner chip acts as a co-processor - taking over some of the most complicated tasks from the computers central processing unit (CPU).

It packs more than 50 cores - or individual processors - onto a single piece of silicon.

The chip offers "double precision" processing which allows a greater amount of numbers to be represented at one time - resulting in faster calculations and more accurate forecasts.

Intel says the accelerator is also the first server processor to support full integration of the PCI Express 3.0 specification. The technology allows data to be transferred at up to 32 gigabytes per second to compatible devices - twice the speed of the previous generation.

"Collecting, analysing and sharing large amounts of information is critical to today's science activities and requires new levels of processor performance and technologies designed precisely for this purpose," said Rajeeb Hazra, Intel's general manager of technical computing.

"Having this performance now in a single chip... is a milestone that will once again be etched into HPC [high performance computing] history," he added.

GPUs allow you to get results more quickly but will take longer to program so there is an interesting trade-off”

While Intel's co-processor relies on the same instruction set architecture as its popular x86 processors, its rivals are taking a different approach.

They are offering chips known as graphic processing units (GPUs) which are designed to carry out the calculations necessary to draw, colour and shade objects on the screen at high speed.

They specialise in processes that can be broken down into several parts, where the output of one calculation does not affect the input of another. This makes them particularly well suited for other tasks such as speech recognition and image processing.

However, developers need to code their software in order to take best advantage of GPUs.

By contrast, Intel's accelerator should be able to run existing applications at high speed without the need for further software development.

"Traditional supercomputers were built by putting thousands of processors in a room but in the last few years there has been a shift toward graphic processors," said Martin Reynolds, a vice president at research firm Gartner.

"GPUs allow you to get results more quickly but will take longer to program so there is an interesting trade-off," he said.

Nvidia specialises in GPUs and the chief executive of Intel's rival, Jen Hsun Huang, talked up the technology at the Seattle conference.

He said that GPUs were less complex than other processors, wasting less energy in moving data across chips.

Computer power has come a long way since 1997 when Intel showed off its first 1 teraflop supercomputer, which required nearly 10,000 Pentium II chips and filled 72 computing cabinets. It cost $55m.

In 2008, IBM's Roadrunner achieved petaflop speed, equal to 1,000 trillion calculations per second.

Ten years on, in 2018, Intel hopes it will be able to deliver so-called exascale-level performance, which is more than 100 times faster than currently available.

Nvidia is not so sure. In his keynote, Mr Hsun Huang said he did not think exascale performance would happen before 2035.

Mr Reynolds thinks that Intel has the edge over Nvidia, at least for the next few years.

"Intel has a technology advantage because its manufacturing processes can make transistors half the size and more efficient but Nvidia will catch up," he said.