{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T06:49:25Z","timestamp":1791442165193,"version":"4.3.3"},"reference-count":36,"publisher":"IEEE","funder":[{"DOI":"10.13039\/100004316","name":"IBM","doi-asserted-by":"publisher","id":[{"id":"10.13039\/100004316","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,6]]},"DOI":"10.23919\/isc.2026.11520488","type":"proceedings-article","created":{"date-parts":[[2026,6,8]],"date-time":"2026-06-08T19:48:54Z","timestamp":1780948134000},"page":"1-18","source":"Crossref","is-referenced-by-count":0,"title":["Training Foundation Models on a Full-Stack AMD Platform: Compute, Networking, and System Design"],"prefix":"10.23919","author":[{"given":"Quentin","family":"Anthony","sequence":"first","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yury","family":"Tokpanov","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Skyler","family":"Szot","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Srivatsan","family":"Rajagopal","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Praneeth","family":"Medepalli","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Anna","family":"Golubeva","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Vasu","family":"Shyam","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Robert","family":"Washbourne","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Rishi","family":"Iyer","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Ansh","family":"Chaurasia","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Tomas","family":"Figliolia","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xiao","family":"Yang","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Abhinav","family":"Sarje","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Drew","family":"Thorstensen","sequence":"additional","affiliation":[{"name":"IBM,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Amartey","family":"Pearson","sequence":"additional","affiliation":[{"name":"IBM,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zack","family":"Grossbart","sequence":"additional","affiliation":[{"name":"IBM,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Jason","family":"van Patten","sequence":"additional","affiliation":[{"name":"IBM,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Emad","family":"Barsoum","sequence":"additional","affiliation":[{"name":"AMD,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Zhenyu","family":"Gu","sequence":"additional","affiliation":[{"name":"AMD,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Yao","family":"Fu","sequence":"additional","affiliation":[{"name":"AMD,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Beren","family":"Millidge","sequence":"additional","affiliation":[{"name":"Zyphra,USA"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","article-title":"The AMD CDNA\u2122 3 architecture","volume-title":"AMD, White Paper","year":"2024"},{"key":"ref2","volume-title":"AMD Pollara 400 Card","year":"2024"},{"key":"ref3","volume-title":"Rail-only: A low-cost high-performance network for training llms with trillion parameters","author":"Wang","year":"2024"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1145\/1402958.1402967"},{"key":"ref5","article-title":"7060x6 series 800g data center switches: Data sheet","volume-title":"Arista Networks, Santa Clara, California, Data Sheet","year":"2025"},{"key":"ref6","volume-title":"Primus: A flexible and high-performance training framework for large-scale foundation model training and inference","year":"2025"},{"key":"ref7","volume-title":"Flashattention-2: Faster attention with better parallelism and work partitioning","author":"Dao","year":"2023"},{"key":"ref8","volume-title":"Babel\u2019Strearn","author":"Deakin","year":"2019"},{"key":"ref9","doi-asserted-by":"crossref","DOI":"10.1145\/3673038.3673136","volume-title":"The case for co-designing model architectures with hardware","author":"Anthony","year":"2024"},{"key":"ref10","volume-title":"Transformer Engine: A library for accelerating Transformer models on NVIDIA GPUs","year":"2023"},{"key":"ref11","volume-title":"NVLINK High-Speed Interconnect","year":"2014"},{"key":"ref12","volume-title":"Gshard: Scaling giant models with conditional computation and automatic sharding","author":"Lepikhin","year":"2020"},{"key":"ref13","volume-title":"Megatron-lm: Training multi-billion parameter language models using model parallelism","author":"Shoeybi","year":"2020"},{"key":"ref14","doi-asserted-by":"crossref","DOI":"10.1109\/HOTI63208.2024.00020","volume-title":"Demystifying the communication characteristics for distributed transformer models","author":"Anthony","year":"2024"},{"key":"ref15","volume-title":"Zero: Memory optimizations toward training trillion parameter models","author":"Rajbhandari","year":"2020"},{"key":"ref16","volume-title":"Gpipe: Efficient training of giant neural networks using pipeline parallelism","author":"Huang","year":"2019"},{"key":"ref17","volume-title":"Understanding RCCL bandwidth and xGMI performance on AMD Instinct\u2122 MI300X","author":"Kolla","year":"2025"},{"key":"ref18","article-title":"Tree attention: Topology-aware decoding for long-context attention on gpu clusters","author":"Shyam","year":"2024","journal-title":"arXiv preprint arXiv"},{"key":"ref19","article-title":"Ring attention with blockwise transformers for near-infinite context","author":"Liu","year":"2023","journal-title":"arXiv preprint arXiv"},{"key":"ref20","volume-title":"Compressed convolutional attention: Efficient attention in a compressed latent space","author":"Figliolia","year":"2025"},{"key":"ref21","volume-title":"Beyond chinchilla-optimal: Accounting for inference in language model scaling laws","author":"Sardana","year":"2025"},{"key":"ref22","volume-title":"A large batch optimizer reality check: Traditional, generic optimizers suffice across batch sizes","author":"Nado","year":"2021"},{"key":"ref23","doi-asserted-by":"crossref","DOI":"10.1145\/3458817.3476209","volume-title":"Efficient large-scale language model training on gpu clusters using megatron-lm","author":"Narayanan","year":"2021"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.1109\/MM.2025.3531323"},{"key":"ref25","article-title":"Apex: A pytorch extension: Tools for easy mixed precision and distributed training in pytorch","volume-title":"gitHub repository","year":"2025"},{"key":"ref26","volume-title":"Transformer Engine: A library for accelerating transformer models on NVIDIA GPUs","year":"2025"},{"key":"ref27","article-title":"Megatron-lm: Distributed muon implementation","volume-title":"fork of NVIDIA\/Megatron-LM, branch: moonshot\/distributedmuon-impl","year":"2025"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.1109\/sc.2010.18"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.1109\/SCWS55283.2021.00018"},{"key":"ref30","article-title":"AITER: AI tensor engine for ROCm","volume-title":"high performance AI operators for AI workloads acceleration","year":"2025"},{"key":"ref31","volume-title":"nvbandwidth: A tool for bandwidth measurements on nvidia gpus","year":"2024"},{"key":"ref32","volume-title":"rocm-systems: Super repository for rocm systems projects","year":"2024"},{"key":"ref33","article-title":"Qwen3 technical report","author":"Yang","year":"2025","journal-title":"arXiv preprint arXiv"},{"key":"ref34","volume-title":"Introducing Meta Llama 3: The most capable openly available LLM to date","year":"2024"},{"key":"ref35","article-title":"Gemma 3 technical report","author":"Team","year":"2025","journal-title":"arXiv preprint arXiv"},{"key":"ref36","article-title":"Olmoe: Open mixture-of-experts language models","author":"Muennighoff","year":"2024","journal-title":"arXiv preprint arXiv"}],"event":{"name":"ISC High Performance 2026 Research Paper Proceedings (41st International Conference)","location":"Hamburg, Germany","start":{"date-parts":[[2026,6,22]]},"end":{"date-parts":[[2026,6,26]]}},"container-title":["ISC High Performance 2026 Research Paper Proceedings (41st International Conference)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/11520448\/11520449\/11520488.pdf?arnumber=11520488","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,10,8]],"date-time":"2026-10-08T05:50:55Z","timestamp":1791438655000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11520488\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,6]]},"references-count":36,"URL":"https:\/\/doi.org\/10.23919\/isc.2026.11520488","relation":{},"subject":[],"published":{"date-parts":[[2026,6]]}}}