Archivio Istituzionale della RicercaAs AI workloads scale to serve billions of users, reliability of custom AI silicon becomes a first-order design constraint. This work describes resilience mechanisms and operational learnings from production deployment of MTIA (Meta Training and Inference Accelerator) across multiple datacenter regions. It highlights distinct reliability requirements in large-scale training versus inference.
Innovative Practices Session: Recent Approaches in Dealing with Silent Data Corruption / Dixit, H.D., Sinha, A., Jagannathan, N., Lerner, D.P., Angione, F.. - (2026), pp. 1-1. (44th VLSI Test Symposium (VTS) Napa, CA (USA) 27-29 April 2026) [10.1109/vts69484.2026.11563339].
Innovative Practices Session: Recent Approaches in Dealing with Silent Data Corruption
Angione, Francesco
2026
Abstract
As AI workloads scale to serve billions of users, reliability of custom AI silicon becomes a first-order design constraint. This work describes resilience mechanisms and operational learnings from production deployment of MTIA (Meta Training and Inference Accelerator) across multiple datacenter regions. It highlights distinct reliability requirements in large-scale training versus inference.
| File | Dimensione | Formato | |
|---|---|---|---|
|
Innovative_Practices_Session_Recent_Approaches_in_Dealing_with_Silent_Data_Corruption.pdf
accesso riservato
Tipologia:
2a Post-print versione editoriale / Version of Record
Licenza:
Non Pubblico - Accesso privato/ristretto
Dimensione
139.46 kB
Formato
Adobe PDF
|
139.46 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/11583/3012445