Towards Near-Threshold Server Processors

Abstract

Abstract-The popularity of cloud computing has led to a dramatic increase in the number of data centers in the world. The ever-increasing computational demands along with the slowdown in technology scaling has ushered an era of power-limited servers. Techniques such as near-threshold computing (NTC) can be used to improve energy efficiency in the post-Dennard scaling era. This paper describes an architecture based on the FD-SOI process technology for near-threshold operation in servers. Our work explores the trade-offs in energy and performance when running a wide range of applications found in private and public clouds, ranging from traditional scale-out applications, such as web search or media streaming, to virtualized banking applications. Our study demonstrates the benefits of near-threshold operation and proposes several directions to synergistically increase the energy proportionality of a near-threshold server.

Energy proportionality in near-threshold computing servers and cloud data centers: Consolidating or Not?

“…3) Trade-offs Discussion: As shown by [3], workloads can tolerate low frequencies if only core power is considered, thus enabling NTC operation to reduce core power consumption. However, not all server components scale with the core voltage, shifting the most energy-efficient point to a higher frequency.…”

“…However, other recent works on processors in FD-SOI demonstrated the near-threshold capabilities of the technology, capable to run a dual-core CortexA9 processor at 1 GHz at the supply voltage of 0.6V [4]. The work presented in [3] was the first one proposing the usage of NTC servers in UTBB FD-SOI technology. Nonetheless, the power model proposed in that work did not include a detailed characterization of the uncore components.…”

“…1) Cores: Similarly to [3] we combine the 28nm FD-SOI power and performance model of a recent Cortex A9 implementation of STM in 28nm bulk and FD-SOI, considering the differences in pipeline length ratio and critical path between Cortex A57 and Cortex A9. These parameters are extracted by comparing the different voltage to frequency ratio (extracted via the CPUFreq Linux driver) present in the Samsung Exynos processor family.…”

“…291125). the trade-off between performance and power, emerging as a promising approach to overcome the power-wall [3].…”

Microprocessors optimal power dissipation using combined threshold hopping and voltage scaling

“…The theoretical concept and design issues of optimal low power limits has strong face validity in recent publications such as; thermal aware dynamic voltage frequency scaling for energy consumption reduction [4], optimal combination of supply-body bias voltage to the core and memory for a real micro-controller chip [5], determining a set of supply-body bias voltage combinations to achieve the minimum energy consumption for a target frequency [6], analysis and determination of the best switching management strategy for dynamic set of operating environment in terms of process choices, circuit activity, and temperatures [7], an architecture based on the fully depleted silicon on insulator (FD-SOI) process technology for server operation near threshold region [8], standby leakage power consumption using the body bias and pin reordering technique for nanometer-scale CMOS circuits [9], finally, Adaptive supply and body voltage control for ultra-low power microprocessors using 22-nm technology model with considerable improvement in power and temperatures [10].…”