skip to main content


This content will become publicly available on April 27, 2025

Title: RAP: Resource-aware Automated GPU Sharing for Multi-GPU Recommendation Model Training and Input Preprocessing
Award ID(s):
2124039
PAR ID:
10538949
Author(s) / Creator(s):
; ; ; ; ;
Publisher / Repository:
ACM
Date Published:
ISBN:
9798400703850
Page Range / eLocation ID:
964 to 979
Format(s):
Medium: X
Location:
La Jolla CA USA
Sponsoring Org:
National Science Foundation
More Like this
  1. Genomic analysis is the study of genes which includes the identification, measurement, or comparison of genomic features. Genomics research is of great importance to our society because it can be used to detect diseases, create vaccines, and develop drugs and treatments. As a type of general-purpose accelerators with massive parallel processing capability, GPUs have been recently used for genomics analysis. Developing GPU-based hardware and software frameworks for genome analysis is becoming a promising research area. To support this type of research, benchmarks are needed that can feature representative, concurrent, and diverse applications running on GPUs. In this work, we created a benchmark suite called Genomics-GPU, which contains 10 widely-used genomic analysis applications. It covers genome comparison, matching, and clustering for DNAs and RNAs. We also adapted these applications to exploit the CUDA Dynamic Parallelism (CDP), a recent advanced feature supporting dynamic GPU programming, to further improve the performance. Our benchmark suite can serve as a basis for algorithm optimization and also facilitate GPU architecture development for genomics analysis. 
    more » « less
  2. We develop a dynamic dictionary data structure for the GPU, supporting fast insertions and deletions, based on the Log Structured Merge tree (LSM). Our implementation on an NVIDIA K40c GPU has an average update (insertion or deletion) rate of 225 M elements/s, 13.5x faster than merging items into a sorted array. The GPU LSM supports the retrieval operations of lookup, count, and range query operations with an average rate of 75 M, 32 M and 23 M queries/s respectively. The trade-off for the dynamic updates is that the sorted array is almost twice as fast on retrievals. We believe that our GPU LSM is the first dynamic general-purpose dictionary data structure for the GPU. 
    more » « less