Asynchronous Data Parallelism
Asynchronous Data Parallelism is a data parallelism strategy in which model replicas independently perform computations on local data shards and update shared parameters without requiring synchronization across all replicas at each iteration, thereby allowing for varying update frequencies and reducing idle time in distributed training environments.
16
Publications
1.1K
Citations
76
Authors
27
Institutions
Publications per year
2017–2026
7
Parents
16
NVIDIA A100 Tensor Core GPU: Performance and Innovation
Jack Choquette, Wishwesh Gandhi, Olivier Giroux et al. · IEEE Micro · 2021 · 380 citations
1–5 of 16
76
1
h-index
Rutgers, The State University of New Jersey, United States
1 publications · 137 citations
1
h-index
1
h-index
1
h-index
1
h-index
1–5 of 76
27
Argonne National Laboratory
Lemont, United States
2 h-index
5 publications
268 citations
University of California, Berkeley
Berkeley, United States
1 h-index
4 publications
56 citations
Rutgers, The State University of New Jersey
New Brunswick, United States
1 h-index
3 publications
411 citations
Microsoft (United States)
Redmond, United States
1 h-index
3 publications
33 citations

Universidad de Málaga
Málaga, Spain
1 h-index
2 publications
82 citations
NEC (Japan)
Tokyo, Japan
1 h-index
6 publications
180 citations
AT&T (United States)
Bedminster, United States
1 h-index
2 publications
220 citations

Barcelona Supercomputing Center
Barcelona, Spain
1 h-index
9 publications
189 citations
1–8 of 27
1 publications · 14 citations
1
h-index
1 publications · 41 citations
1
h-index
1 publications · 40 citations
1
h-index
TSE · 1 publications · 110 citations
1
h-index
1 publications · 11 citations
1
h-index
1–5