1. System Simulation Of 1000-cores Heterogeneous SoCs Shivani Raghav Embedded System Laboratory (ESL) Ecole Polytechnique Federale de Lausanne (EPFL)

2. ESL Work on Energy-Aware Datacenter Design 2 System Simulation for many-core

3. Emerging Data-Intensive Workloads Cloud Servers Molecular Dynamics Monte Carlo Simulations Gene Sequencing Online Gaming Services Financial Simulations Medical Imaging

4. Demand for Hardware Acceleration Tile based Manycores Intel SCC, Tile 64 (Integrated) GPU Clusters (off –chip Accelerators) Hybrid Cores AMD Fusion (on-chip)

5. Urgent Need for Simulation of Heterogeneous SoCs Thermal & Power Evaluations Benchmarking Profiling Debugging Design Space Exploration Early Software Development Simulation

6. How to Design a Fast and Scalable Many-Core Simulator? Parallel Target Parallel Simulator Parallel Host

7. Simulating Parallel Target on Parallel Host is an Old Technology… FPGA GPGPU Flexus RAMP Opportunity WWT II Graphite Cotson, OVPSim Large Parallel Systems

8. Target Architecture Data-Parallel Coprocessors Simple In-order Cores 1000s of cores in a tile network Fine grain parallelism Core Caches Memory Switch

9. Solution – Accelerating Simulation using GPGPUs Target ArchitectureHost Platform A Perfect Match

10. Outline Problem Overview Simulation of Heterogeneous SoCs Solution SIMinG-1k (GPU accelerated simulator) Evaluation Summary

11. Outline Problem Overview Simulation of Heterogeneous SoCs Solution SIMinG-1k: A GPU accelerated simulator Evaluation Summary

12. Overall Simulation Framework Host Platform Sequential Code Data Parallel Code Simulator Target Architecture General Purpose CPU General Purpose CPU Many-Core Accelerator Application

13. SIMinG-1k - Features Instruction Accurate Inexpensive and Easily Available Fast Development Cycle Equation Performance Model Portability (Target Independent) Interpretation based core-simulation

14. Challenges of using GPU as a host SIMT (Single inst multiple threads) Divergent Code is a problem Synchronization outside thread block Slow CPU-GPU communication Global Memory is slow and limited

15. Outline Problem Overview Simulation of Heterogeneous SoCs Solution SIMinG-1k (GPU accelerated simulator) Evaluation Summary

16. Results – Architecture 1 MIPS - Number of simulated instruction in host wall clock time ARM ISA Data Scratchpad Single tile of target Accelerator Inst Scratchpad

17. Speed Up – Architecture 1 Speedup compared to simulation on OVPSim (thousands of ARM cores)

18. Single tile of Data-parallel Accelerator (cores, caches, on-chip interconnect) Results – Architecture 2 Core Caches Memory Switch

19. Speed Up – Architecture 2 Speedup compared to serial simulation on QEMU

20. Outline Problem Overview Simulation of Heterogeneous SoCs Solution SIMinG-1k (GPU accelerated simulator) Evaluation Summary

21. Conclusion  Challenge Fast and parallel simulator for heterogeneous SoCs  Solution Parallelize 1000 core simulation using GPUs  Design Full System Simulation using QEMU and SIMinG-1k  Results High Scalability and speedup upto 4096 cores  Extend the simulator for thermal and power evaluations  Complete simulation of Cloud Data Centers Future Work

22. Thanks! Questions?