Bandana: Using Non-volatile Memory for Storing Deep Learning Models

doi:doi:10.57702/wuu05chh

Bandana: Using Non-volatile Memory for Storing Deep Learning Models

Typical large-scale recommender systems use deep learning models that are stored on a large amount of DRAM. These models often rely on embeddings, which consume most of the required memory. We present Bandana, a storage system that reduces the DRAM footprint of embeddings, by using Non-volatile Memory (NVM) as the primary storage medium, with a small amount of DRAM as cache.

BibTex:

@dataset{Assaf_Eisenman_and_Maxim_Naumov_and_Darryl_Gardner_and_Misha_Smelyanskiy_and_Sergey_Pupyrev_and_Kim_Hazelwood_and_Asaf_Cidon_and_Sachin_Katti_2025,
    abstract = {Typical large-scale recommender systems use deep learning models that are stored on a large amount of DRAM. These models often rely on embeddings, which consume most of the required memory. We present Bandana, a storage system that reduces the DRAM footprint of embeddings, by using Non-volatile Memory (NVM) as the primary storage medium, with a small amount of DRAM as cache.},
    author = {Assaf Eisenman and Maxim Naumov and Darryl Gardner and Misha Smelyanskiy and Sergey Pupyrev and Kim Hazelwood and Asaf Cidon and Sachin Katti},
    doi = {10.57702/wuu05chh},
    institution = {No Organization},
    keyword = {'Bandana', 'Cache', 'DRAM', 'Deep Learning', 'NVM', 'Non-Volatile Memory', 'Recommender Systems'},
    month = {jan},
    publisher = {TIB},
    title = {Bandana: Using Non-volatile Memory for Storing Deep Learning Models},
    url = {https://service.tib.eu/ldmservice/dataset/bandana--using-non-volatile-memory-for-storing-deep-learning-models},
    year = {2025}
}