Fujitsu's New Circuit Will Boost Server, Supercomputer Performance

Researchers have developed a circuit that will double the communication speed between CPUs, which is important in cloud computing.

data center

Fujitsu Laboratories is introducing a circuit that officials said could significantly speed up the performance of servers and supercomputers that are under increasing pressure from cloud computing environments to process data more quickly.

The circuit will be able to move data between CPUs in the system at 56 gigabits per second, leveraging a new architecture Fujitsu officials have called "look ahead," which addresses the quality degradation in incoming signals, parallels the processing and increases the operating frequency, enabling the circuit to double its speed.

Fujitsu Laboratories researchers presented its findings on the new circuit June 13 at the 2014 Symposia on VLSI Technology and Circuits in Honolulu.

Standard practice for speeding up the data processing in servers has been through improving the performance of the CPU, including cranking up the speed of the data communications between the chips, company officials said. A challenge has been the degradation of waveforms of incoming signals, something circuits have had a difficult time addressing.

The "look ahead" architecture uses a decision feedback equalizer (DFE) to compensate for the degraded incoming data signals by correcting the input signal based on the bit-values within the signal. There are other circuit designs that run at 56G bps, but they use four 14G-bps circuits running in parallel. Fujitsu Laboratories researchers said their design reaches 56G bps with a single circuit.

Increasing the speed of communications between the CPUs in both a single chassis and between separate chassis is becoming increasingly important with the continued growth of cloud computing, according to Fujitsu officials. Cloud computing is putting increased demands on organizations to improve the performance of their data centers by using servers that process data much faster than traditional systems.

This is done through both using faster CPUs and through the use of large-scale servers that tie together many CPUs, they said. The amount of data being processed by these systems is growing fast and will continue to increase quickly going well into the future. With that in mind, there is a push on to double the current level of data transmission to 56G bps, and the Optical Internetworking Forum is pushing forward on the drive to standardize 56G bps for optical modules used for optical transmission between chassis, they said.

Fujitsu Laboratories researchers are continuing to work on the circuit technology, with plans to apply it to interfaces of CPUs and optical modules. The first practical implementations will come in late 2016 or early 2017. While the technology is aimed at high-end servers and supercomputers, researchers reportedly have said that it could find its way into consumer devices as well.