-
Notifications
You must be signed in to change notification settings - Fork 14
Home
The Altis benchmark suite consists of a series of benchmark applications used to test performance and other aspects of systems with Graphics Processing Units (GPUs). Its primary focus is on NVIDIA-based devices, and on the Compute Unified Device Architecture (CUDA) computing platform. It is developed in SCEA lab at University of Texas at Austin.
The Altis benchmark suite is divided into three levels. Each level represents benchmark applications whose behaviors of interest range from low-level characteristics such as bus bandwidth to end-to-end performance of real world applications. This categorization is adopted from the Scalable Heterogeneous Computing (SHOC) Benchmark Suite. The level structure is:
- Level 0: Measures low level characteristics of the targeted hardware. This level includes simple benchmarks such as maxflop and bus bandwidth.
- Level 1: Includes basic parallel algorithms which are commonly seen in parallel computing and used in kernels of real applications.
- Level 2: more complicated real-world applications kernels, often found in industry.
Altis is publicly available at https://github.com/utcs-scea/altis. Simply execute git clone https://github.com/utcs-scea/altis.git to access the source.
For any questions regarding this project, please send an email to [email protected] or [email protected]
Bibtex is shown below:
@INPROCEEDINGS{9238617,
author={B. {Hu} and C. J. {Rossbach}},
booktitle={2020 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS)},
title={Altis: Modernizing GPGPU Benchmarks},
year={2020},
volume={},
number={},
pages={1-11},
doi={10.1109/ISPASS48437.2020.00011}}
B. Hu and C. J. Rossbach, "Altis: Modernizing GPGPU Benchmarks," 2020 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), Boston, MA, USA, 2020, pp. 1-11, doi: 10.1109/ISPASS48437.2020.00011.
This research was supported by NSF grants CNS-1618563 and CNS-1846169.