Scaling up Servers
With massive projected growth attributed to the Internet of Things, how are semiconductor vendors keeping up with demands for capacity and performance from internet and data centre infrastructure? Sally Ward-Foxton finds out.
With more and more connected devices becoming part of the Internet of Things (IoT), the amounts of data being communicated between these devices is increasing at an extremely rapid rate. Coupled with this, evolution of IT systems which now place all or most data in remote servers and data centres (‘in the cloud’) means that new server technologies are having to develop quickly to keep up with demands for increased capacity and bandwidth. These requirements are passed on to semiconductor manufacturers, but how are they responding to demands for space saving, increased performance and better energy efficiency from the server market?
Curtis Pulley, Data Centre Business Development Manager at Xilinx, explains that heterogeneous computing, or using more than one type of processor in a multi-core system, is a rapidly accelerating trend in the data centre industry. The idea is that dissimilar processors have different strengths and therefore handle different types of task better, so performance can be improved if it is implemented correctly. This is leading to increased use of FPGAs in the server world; FPGAs can be programmed to perform specific algorithms in hardware, making them an effective hardware accelerator for number-crunching tasks.
“FPGAs’ parallel architecture is inherently optimised to accelerate specific workloads in data centres to achieve superior performance per Watt metrics for a given compute workload,” commented Pulley, adding: “Xilinx is investing in this area on multiple fronts to deliver to the exponentially growing compute needs in the data centre while delivering the optimally powered solution. The internet of things is seen as one of the drivers pushing these compute demands in the data centre.”
Pulley also points to the leading edge process technology used by semiconductor manufacturers to deliver more and more impressive power and compute features. For example, Xilinx’s 7 Series of FPGAs uses the 28nm process node, but forthcoming UltraScale products are currently ramping in 20nm, and will be in 16nm by the end of this year, he says. In general, continuing to move to lower process nodes when they become available means smaller and more cost-effective FPGAs, though this is sometimes at the expense of leakageLow DC leakage current.Low DC leakageLow DC leakage current. current. current and therefore power consumption. This trend is set to continue as new transistor architectures are invented to minimise leakageLow DC leakage current.Low DC leakageLow DC leakage current. current..
Getting the best possible performance and power efficiency out of FPGAs also depends on how you use them. Xilinx has also been working on tools which help maximise the devices’ utilisation and productivity.
“Critical to the IoT space, we announced the SDNet Design Environment last year,” said Pulley. “SDNet enables the easy creation of high performance packet processing systems, based on compiling high-level user defined specifications to optimise all-programmable FPGAs and SoCs. This capability enables the creation of solutions to manage the new network traffic flows being driven by the emerging IoT space.”
via Scaling up Servers.