Hardware-Software Codesign 11. Thermal-Aware Design Iuliana Bacivarov & Lothar Thiele Swiss Federal Computer Engineering 11 - 1 Institute of Technology and Networks Laboratory
Contents Why is it important to consider temperature in system design? Power and temperature models Thermal simulation Thermal-aware scheduling Swiss Federal Computer Engineering 11 - 2 Institute of Technology and Networks Laboratory
Power/Thermal Wall Power/Thermal wall is recognized as the most significant barrier towards high performance Swiss Federal Computer Engineering 11 - 3 Institute of Technology and Networks Laboratory
Multi-Cores Face the Power/Thermal Wall Too [Loh: 3D-Stacked Memory Architectures for Multi-Core Processors, 2008] 72-Core Intel Xeon Phi platform Swiss Federal Computer Engineering 11 - 4 Institute of Technology and Networks Laboratory
Some Solutions VLSI design and cooling solutions Thermal-aware design, materials, reduce leakage and switching, ... Use better heat sinks, fans, air cooling, liquid cooling [source: Wikipedia] Thermal management Voltage/frequency scaling Stop-go execution completely TURN OFF components to allow for cooling Migration of tasks from hot to cool area Swiss Federal Computer Engineering [MJPEG decoder on 25-core processor] 11 - 5 Institute of Technology and Networks Laboratory
But scheduling of jobs and thermal management techniques affect both timing and thermal properties Thermal and performance objectives must be considered simultaneously during design Swiss Federal Computer Engineering 11 - 6 Institute of Technology and Networks Laboratory
Some Design Questions Thermal and performance objectives must be considered simultaneously during design How can we simultaneously consider during design both timing and temperature ? What is the worst case peak temperature of the chip? What is an optimal temperature-aware mapping scheme? What are temperature-aware scheduling techniques with low overhead (simple control, no temperature sensors)? Swiss Federal Computer Engineering 11 - 7 Institute of Technology and Networks Laboratory
Contents Why is it important to consider temperature in system design? Power and temperature models Thermal simulation Thermal-aware scheduling Swiss Federal Computer Engineering 11 - 8 Institute of Technology and Networks Laboratory
Single Source Power Model Frequently used power model for constant voltage Silicon chip Single power source Active processing Idle mode temperature a a a P ( t ) T ( t ) , for active processing = φ ⋅ + ψ P ( t ) { P ( t ) T ( t ) ( t ) = = φ ⋅ + ψ i i i P ( t ) T ( t ) = φ ⋅ + ψ , for idle mode power Just Including both leakage power dynamic and leakage power Swiss Federal Computer Engineering 11 - 9 Institute of Technology and Networks Laboratory
Active processing Idle mode Power Model a a a P ( t ) T ( t ) , for active processing = φ ⋅ + ψ P ( t ) { P ( t ) T ( t ) ( t ) = = φ ⋅ + ψ i i i P ( t ) T ( t ) = φ ⋅ + ψ , for idle mode Just Including both leakage power dynamic and leakage power Dynamic power Dynamic power Leakage power Leakage power - - Independent of temperature Independent of temperature - - Independent of the load Independent of the load - - Different power consumption for Different power consumption for - - Depends on the current temperature of Depends on the current temperature of every code segment every code segment the component the component - - Separate power consumption for Separate power consumption for - - Model [Skadron et al. 2004] Model [Skadron et al. 2004] each component (core, cache, each component (core, cache, 2 C / T P ~ T e − ⋅ Leak memory, …) memory, …) - - For the remaining lecture: We use a For the remaining lecture: We use a linear approximation (see above). linear approximation (see above). Swiss Federal Computer Engineering 11 - 10 Institute of Technology and Networks Laboratory
Static Power / Dynamic Power Ratio Swiss Federal Computer Engineering 11 - 11 Institute of Technology and Networks Laboratory
Static Power and Dynamic Power Dynamic power consumption: Between 0 and 1; quantifies switching activity Clock frequency Total capacity Supply voltage Static power consumption: - 20% or more in sub-micron era - Mostly leakage, i.e., the power dissipated by a transistor whose gate is intended to be off Swiss Federal Computer Engineering 11 - 12 Institute of Technology and Networks Laboratory
Single Power Source Model Single power source I ≅ P Silicon chip V ≅ T Cooling G C V 0 ≅ T amb P ( t ) T ( t ) ( t ) = φ ⋅ + ψ Power parameters Thermal Environment Thermal conductance temperature capacity Swiss Federal Computer Engineering 11 - 13 Institute of Technology and Networks Laboratory
Solution of the Thermal Equation Explicit solution Steady state temperature Swiss Federal Computer Engineering 11 - 14 Institute of Technology and Networks Laboratory
Temperature Profile Active state : dynamic and static (leakage) power Task execution schedule Temperature increase : based on linear thermal model Swiss Federal Computer Engineering 11 - 15 Institute of Technology and Networks Laboratory
Temperature Profile Idle state : static (leakage) power Task execution schedule Temperature decrease : based on linear thermal model Swiss Federal Computer Engineering 11 - 16 Institute of Technology and Networks Laboratory
Temperature Profile Task execution schedule Peak temperature Swiss Federal Computer Engineering 11 - 17 Institute of Technology and Networks Laboratory
Multi Source Models [ Barcella et. Al., U. Virginia ] Layout RC equivalent model - A and B are matrixes - T is an N-dimensional temperature vector - u is the input vector Swiss Federal Computer Engineering 11 - 18 Institute of Technology and Networks Laboratory
Multi Source Models – Solution Explicit solution: Impulse response matrix A, H, B, are matrixes T is an N-dimensional temperature vector H ij (t) is the impulse response between power injected at source j and temperature variation at location i Swiss Federal Computer Engineering 11 - 19 Institute of Technology and Networks Laboratory
Multi-Core Effect Self-heating effect Neighboring effect#2 Neighboring effect #1 temperature temperature temperature no delay delayed delayed Neighboring effect#3 temperature delayed Swiss Federal Computer Engineering 11 - 20 Institute of Technology and Networks Laboratory
The Impulse Response Temperature rises with power Temperature rises with power at same location (without delay) at some other location after delay Swiss Federal Computer Engineering 11 - 21 Institute of Technology and Networks Laboratory
Multi-Core Effect – Heat Transfer (I) C = thermal capacitance matrix G =thermal conductance matrix K = thermal ground conductance matrix P = power dissipation vector T amb = ambient temperature vector T amb = T amb [1, . . . , 1]’ Power dissipated by component l l in ‘active’ ( a ) and ‘idle’ ( i ) processing modes Thermal model u (t) =input vector Swiss Federal Computer Engineering 11 - 22 Institute of Technology and Networks Laboratory
Multi-Core Effect – Heat Transfer (II) closed-form solution of the temperature = self-impulse response = impulse response between nodes l l and k Swiss Federal Computer Engineering 11 - 23 Institute of Technology and Networks Laboratory
Multi-Core Effect – Heat Transfer (III) Closed-form solution of the temperature Convolution between the impulse Temperature of node k response H kl and the input u l Workload of component l Swiss Federal Computer Engineering 11 - 24 Institute of Technology and Networks Laboratory
Multi-Core Effect self-impulse response H kk (τ-t) Swiss Federal Computer Engineering 11 - 25 Institute of Technology and Networks Laboratory
Solving the Differential Equations What’s happening in numerical simulations? T ( t ) dT ( t ) Δ A T ( t ) B u ( t ) A T ( t ) B u ( t ) = ⋅ + ⋅ = ⋅ + ⋅ k 1 k 1 − − t dt Δ T ( 0 t ) T T ( t ) T ( t ) T ( t ) = Δ = − amb k k 1 − Temperature of Constant time interval interest [ ] T ( t ) T ( t ) A T ( t ) B u ( t ) t = + ⋅ + ⋅ ⋅ Δ k k 1 k 1 k 1 − − − With P(t) = P = const. for 0 ≤ t ≤ Δt, (and therefore u(t) = const.) Simulators HotSpot http://lava.cs.virginia.edu/HotSpot/index.htm 3D-Ice http://esl.epfl.ch/3d-ice.html Swiss Federal Computer Engineering 11 - 26 Institute of Technology and Networks Laboratory
Contents Why is it important to consider temperature in system design? Power and temperature models Thermal simulation Thermal-aware scheduling Swiss Federal Computer Engineering 11 - 27 Institute of Technology and Networks Laboratory
Recommend
More recommend
