Power/Performance Bits: Aug. 30

Scalable data center chip

Princeton University researchers designed a new scalable chip specifically for data centers and massive computing systems.

The team believes the chip, called Piton, can substantially increase processing speed while slashing energy needs. The chip architecture is scalable; designs can be built that go from a dozen cores to several thousand. Also, the architecture enables thousands of chips to be connected together into a single system containing millions of cores.

“With Piton, we really sat down and rethought computer architecture in order to build a chip specifically for data centers and the cloud,” said David Wentzlaff, an assistant professor of electrical engineering and associated faculty in the Department of Computer Science at Princeton University. “The chip we’ve made is among the largest chips ever built in academia and it shows how servers could run far more efficiently and cheaply.”

The current version of the Piton chip measures six by six millimeters and was taped out on IBM’s 32nm SOI process. The chip has 460 million transistors and 25 cores.

An annotated CAD tool layout of the Princeton Piton Processor showing 25 cores. (Source: Princeton University)

The Piton chip’s design focuses on exploiting commonality among programs running simultaneously on the same chip through execution drafting. At a data center, multiple users often run programs that rely on similar operations at the processor level. The chip’s cores can recognize these instances and execute identical instructions consecutively, so that they flow one after another. Doing so can increase energy efficiency by about 20% compared to a standard core, the researchers said.

The chip parcels out when competing programs can access computer memory that exists off of the chip. The team said this approach can yield an 18% performance jump compared to conventional allocation.

The Piton chip also gains efficiency by its management of cache memory, which assigns areas of the cache and specific cores to dedicated applications. The researchers say the system can increase efficiency by 29% when applied to a 1,024-core architecture. They estimate that this savings would multiply as the system is deployed across millions of cores in a data center.

The researchers have made the design open source, available at http://www.openpiton.org. You can read more about it at the team’s Hot Chips presentation.

Solar supercomputing

The University of Texas at Austin, along with Japan’s New Energy and Industrial Technology Development Organization, completed an effort to build a data center which runs primarily on solar power.

The supercomputer, a 432-node HPE Apollo 8000 with peak performance of 400 teraflops called “Hikari,” is primarily used for health and disease research involving sensitive patient data.

In particular, Hikari highlights the potential of High Voltage Direct Current (HVDC) for high-performance computers. Operating on a 380-volt DC distribution system powered by 240 kilowatts from a solar farm the size of 55 parking spaces, the system also includes an HVDC UPS battery system and HVDC CRAC (Computer Room Air Conditioner) units, as well as the power distribution equipment.

The Hikari system. (Source: Jorge Salazar)

Running all DC from the solar source through the UPS and directly to the racks saves four AC/DC conversions that would happen in a normal datacenter. When solar is unavailable, the system will use grid power.

According to the Texas Advanced Computing Center (TACC), “In addition to the HVDC components, the HPE Apollo 8000 has a number of other “green” features that should make this our most efficient system yet. The system continues the evolution of bringing water ever closer to the racks with self-contained water-cooled racks (that don’t reject any heat into the datacenter air). Separate water loops move air directly across the processors – these loops are at near-vacuum pressures so the water will boil at low temperatures (around 105F or 40C) and convect to the sides of each blade, where it hits a heat exchanger, condenses, and is pushed back over the processors without the need for active pumps in each blade.”

Stretching supercapacitors

Hoping to push forward the development of wearable electronics, researchers at the Nanyang Technological University in Singapore developed a stretchy micro-supercapacitor using ribbons of graphene.

“Most power sources, such as phone batteries, are not stretchable. They are very rigid,” said Xiaodong Chen, associate professor of materials science and engineering at Nanyang. “My team has made stretchable electrodes, and we have integrated them into a supercapacitor.”

Graphene is renowned for its thinness, strength and conductivity. “Graphene can be flexible and foldable, but it cannot be stretched,” he said. To fix that, Chen’s team took a cue from skin. Skin has a wave-like microstructure, Chen said. “We started to think of how we could make graphene more like a wave.”

The researchers’ first step was to make graphene micro-ribbons. Most graphene is produced with physical methods, but the team used chemistry to build the material. “We have more control over the graphene’s structure and thickness that way,” Chen explained. “It’s very difficult to control that with the physical approach. Thickness can really affect the conductivity of the electrodes and how much energy the supercapacitor overall can hold.”

The stretchy supercapacitor made from graphene. (Source: Xiaodong Chen, Ph.D.)

The next step was to create the stretchable polymer chip with a series of pyramidal ridges. The researchers placed the graphene ribbons across the ridges, creating the wave-like structure. The design allowed the material to stretch without the graphene electrodes of the superconductor detaching, cracking or deforming. In addition, the team developed kirigami structures, which are variations of origami folds, to make the supercapacitors 500% more flexible without decaying their electrochemical performance. As a final test, the researchers powered an LCD from a calculator with the stretchy graphene-based micro-supercapacitor. Similarly, such stretchy supercapacitors could be used in pressure or chemical sensors.

In future experiments, the researchers hope to increase the electrode’s surface area so it can hold even more energy. The current version only stores enough energy to power LCD devices for a minute.