Skip to Main content Skip to Navigation
Reports

Decoupled Thermal Simulation

Surendra Guntur 1 Pierre Michaud 1
1 CAPS - Compilation, parallel architectures and system
IRISA - Institut de Recherche en Informatique et Systèmes Aléatoires, Inria Rennes – Bretagne Atlantique
Abstract : Small transistors and high clock frequency have resulted in high power density, which makes temperature a strong constraint in today's microprocessor design. For maximizing performance, the thermal design power must be set according to average, instead of worst case, conditions. Consequently, current processors feature temperature sensors and throt-tling mechanisms to keep the chip temperature at a safe level. To study future thermally-constrained processors and systems, researchers and engineers use cycle-accurate performance simulators modeling power consumption and temperature. Cycle-accurate simulators are relatively slow and make it difficult to study long-term thermal behaviors that may require to simulate several minutes or even hours of processor execution. Sampling or phase analysis cannot be applied directly in this case because temperature depends on all past energy events. We propose a partial solution to this problem, which consists in decoupling cycle-accurate simulations and thermal ones. Temperature-unaware cycle-accurate simulation is used to generate an energy trace representing the complete execution of an application. Phase analysis can be used to decrease the trace generation time and make compact traces. Temperature and thermal-throttling are simulated in a separate thermal simulator that reads energy traces. The thermal simulator is faster than the cycle-accurate one and can be used to explore, with the same energy trace, parameters that are not modeled in cycle-accurate simulation.
Complete list of metadata

Cited literature [50 references]  Display  Hide  Download

https://hal.inria.fr/inria-00192779
Contributor : Anne Jaigu <>
Submitted on : Thursday, November 29, 2007 - 1:33:40 PM
Last modification on : Tuesday, June 15, 2021 - 4:16:42 PM
Long-term archiving on: : Monday, April 12, 2010 - 5:30:44 AM

Files

PI-1871.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : inria-00192779, version 1

Citation

Surendra Guntur, Pierre Michaud. Decoupled Thermal Simulation. [Research Report] PI 1871, 2007, pp.31. ⟨inria-00192779⟩

Share

Metrics

Record views

645

Files downloads

170