View on Github

Accel-Sim v1.1.0 has officially been released!
Accel-Sim is a simulation framework for simulating and validating programmable accelerators like GPUs. For full details, please see our recent ISCA 2020 paper and download slides from here. To keep you up-to-date with the recent news, please join our Google group here!

If you use GPGPU-Sim 4.x, trace-driven simulation, or any of the Accel-Sim components in your research, please cite:

  • Mahmoud Khairy, Jason Shen, Tor M. Aamodt, and Timothy G. Rogers "Accel-Sim: An Extensible Simulation Framework for Validated GPU Modeling", In The 47th International Symposium on Computer Architecture, May 2020
  • Extensively Validated

    Using NVIDIA's machine ISA (SASS), integrated into GPGPU-Sim 4.0's performance model, Accel-Sim is highly correlated to contemporary NVIDIA architectures. Accel-Sim provides a streamlined workflow to generate detailed correlation information with little effort.

    Simulate SASS for any CUDA App

    No functional implementation required. Trace any CUDA binary, including those using cuBLAS, cuDNN, pyTorch, etc. If it runs in silicon: you can simulate it with mininal effort in Accel-Sim.

    Highly Extensible

    Building an extensible simulation for rapidly-evolving GPU architecture is challenging. Accel-Sim is built to ensure it is up-to-date with industrial designs and reduces the simulation accuracy gap between academia and industry on an ongoing basis.


    Accel-Sim Overview

    Accel-Sim consists of four main components:
    1. Accel-Sim Tracer: An NVBit tool for generating SASS traces from CUDA applications.
    2. Accel-Sim SASS Frontend: A simulator frontend that consumes SASS traces and feeds them into a performance model. The intial release of Accel-Sim coincides with the release of GPGPU-Sim 4.0, which acts as the detailed performance model.
    3. Accel-Sim Correlator: A tool that matches, plots and correlates statistics from the performance model with real hardware statistics generated by profiling tools.
    4. Accel-Sim Tuner: An automated tuner that automates configuration file generation from a detailed microbenchmark suite.


    Our ISCA 2020 video details the framework in more detail. We also plan to hold a tutorial in an upcoming architecture confernece, which will be made available online. The Accel-Sim effort includes an extensive update to GPGPU-Sim to make the performance model more closely model contemporary GPU hardware. Output from the Accel-Sim Correlator details the cycle-level accuracy improvement.


    Manual

    If you have any questions, please feel free to join our Google group.

    Accel-Sim Roadmap

    Accel-Sim is an Open-Source, community-driven framework. If you are interested in contributing, or influencing the roadmap contact us!