Meta's Second Generation AI Chip: Model-Chip Co-Design and Productionization Experiences
Authors:
- Joel CoburnJCJoel Coburn
- Chunqiang TangCTChunqiang Tang
- Sameer Abu AsalSASameer Abu Asal
- Neeraj AgrawalNANeeraj Agrawal
- Raviteja ChintaRCRaviteja Chinta
- Harish DixitHDHarish Dixit
- Brian DoddsBDBrian Dodds
- Saritha DwarakapuramSDSaritha Dwarakapuram
- Amin FiroozshahianAFAmin Firoozshahian
- Cao GaoCGCao Gao
- Kaustubh GondkarKGKaustubh Gondkar
- Tyler GrafTGTyler Graf
Open Access
- 20 June 2025
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 1689-1702
Abstract
No abstract availableThis publication has 17 references indexed in Scilit:
- Distributed Training of Large Language Models on AWS Trainium2024
- Next Gen MTIA -Recommendation Inference Accelerator2024
- PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation2024
- MTIA: First Generation Silicon Targeting Meta's Recommendation Systems2023
- Cerebras Architecture Deep Dive: First Look Inside the Hardware/Software Co-Design for Deep LearningIEEE Micro, 2023
- AI accelerator on IBM Telum processor2022
- NVIDIA A100 Tensor Core GPU: Performance and InnovationIEEE Micro, 2021
- Ascend: a Scalable and Unified Architecture for Ubiquitous Deep Neural Network Computing : Industry Track Paper2021
- Triton: an intermediate language and compiler for tiled neural network computations2019
- Practical Lessons from Predicting Clicks on Ads at Facebook2014