FlashMem: Supporting Modern DNN Workloads on Mobile with GPU Memory Hierarchy Optimizations 2179 Conference: Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2; March 22–26, 2026; Pittsburgh, PA, USA Proceedings of
Zhihao Shu, Md Musfiqur Rahman Sanim, Hangyu Zheng, Kunxiong Zhu, Miao Yin, Gagan Agrawal, Wei Niu · 2026
Read FlashMem: Supporting Modern DNN Workloads on Mobile with GPU Memory Hierarchy Optimizations 2179 Conference: Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2; March 22–26, 2026; Pittsburgh, PA, USA Proceedings of by Zhihao Shu, Md Musfiqur Rahman Sanim, Hangyu Zheng, Kunxiong Zhu, Miao Yin, Gagan Agrawal, Wei Niu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-15379.txt.
Description
Shu and colleagues offer FlashMem, a flowing setup for running large modern deep brain nets and multi-DNN jobs on mobile GPUs while cutting memory use and cutting run lag. Current setups mostly use an early loading plan, loading all model weights into memory before runs. The team argues this method is weak for modern jobs holding very large models and maybe some other models in a row.
Instead of fully preloading weights, FlashMem sets model loading plans ahead and flows them live on need. It uses two-and-a-half-dimensional surface memory to cut reshapes and boost speed. Test findings on 11 models show FlashMem hits 2 to about 8 times memory cut and about 2 to 75 times boost versus current setups. That allows lean runs of large-scale models and multi-DNN aid on tight mobile GPUs.
Catalog record
| CREATOR | Zhihao Shu, Md Musfiqur Rahman Sanim, Hangyu Zheng, Kunxiong Zhu, Miao Yin, Gagan Agrawal, Wei Niu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-15379