Performance Modeling of the NVIDIA Deep-Learning Accelerator (NVDLA) using Performance Representatives
Bachelor’s Thesis / Master’s Thesis / Student Research Project
Abstract
This student project’s goal is to perform benchmarks on the Nvidia NVDLA Machine Learning Accelerator and use those measurements to create a statistical performance estimator using the Performance Representatives (PR) approach. This model should then be compared to existing analytical models, like AMAIX, and other performance estimation approaches, like ACADL/AIDG.
References
- A. L.-F. Jung, J. Steinmetz, J. Gietz, K. Lübeck, und O. Bringmann, „It’s all about PR – Smart Benchmarking AI Accelerators using Performance Representatives“. arXiv, 12. Juni 2024.
- Mika Markus Müller, Alexander Richard Manfred Borst, Konstantin Lübeck, Alexander Louis-Ferdinand Jung, Felix Wedlich, Oliver Bringmann - Using the Abstract Computer Architecture Description Language to Model AI Hardware Accelerators
- Konstantin Lübeck, Alexander Louis-Ferdinand Jung, Felix Wedlich, Oliver Bringmann - Work-in-Progress: Ultra-fast yet Accurate Performance Prediction for Deep Neural Network Accelerators
- N. Zurstraßen, L. Jünger, T. Kogel, H. Keding, und R. Leupers, „AMAIX In-Depth: A Generic Analytical Model for Deep Learning Accelerators“, Int J Parallel Prog, Bd. 50, Nr. 2, S. 295–318, Apr. 2022.
Requirements
- Python
- Linux
- Successfully atteded the lecture “Grundlagen der Rechnerarchitektur” and/or “Parallele Rechnerarchitekturen” and/or “Modellierung und Analyse Eingebetteter Systeme” (optional)