{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,5]],"date-time":"2026-03-05T07:21:31Z","timestamp":1772695291704,"version":"3.50.1"},"reference-count":50,"publisher":"IEEE","license":[{"start":{"date-parts":[[2026,1,31]],"date-time":"2026-01-31T00:00:00Z","timestamp":1769817600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2026,1,31]],"date-time":"2026-01-31T00:00:00Z","timestamp":1769817600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2026,1,31]]},"DOI":"10.1109\/hpca68181.2026.11408582","type":"proceedings-article","created":{"date-parts":[[2026,3,4]],"date-time":"2026-03-04T20:47:22Z","timestamp":1772657242000},"page":"1-14","source":"Crossref","is-referenced-by-count":0,"title":["SCALE: Tackling Communication Bottlenecks in Confidential Distributed Machine Learning"],"prefix":"10.1109","author":[{"given":"Joongun","family":"Park","sequence":"first","affiliation":[{"name":"Georgia Institute of Technology"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Yongqin","family":"Wang","sequence":"additional","affiliation":[{"name":"University of Southern California"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Huan","family":"Xu","sequence":"additional","affiliation":[{"name":"Georgia Institute of Technology"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Hanjiang","family":"Wu","sequence":"additional","affiliation":[{"name":"Georgia Institute of Technology"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Mengyuan","family":"Li","sequence":"additional","affiliation":[{"name":"University of Southern California"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Tushar","family":"Krishna","sequence":"additional","affiliation":[{"name":"Georgia Institute of Technology"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"263","reference":[{"key":"ref1","article-title":"NVLinkTM high-speed gpu interconnect: Application performance brief","journal-title":"White paper, 2016, basic overview of NVLink interconnect architecture; does not explicitly describe the retry protocol"},{"key":"ref2","article-title":"Nvidia hopper architecture technical overview","year":"2022","journal-title":"NVIDIA Corporation, Tech. Rep."},{"key":"ref3","article-title":"Nvidia hopper architecture technical overview","year":"2022","journal-title":"NVIDIA Corporation, Tech. Rep."},{"key":"ref4","article-title":"Optimizing multi-gpu communication with nccl and gpudirect async","volume-title":"GTC Conference Proceedings","year":"2024"},{"key":"ref5","year":"2025","journal-title":"NCCL Documentation: Environment Variables"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.70"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.1145\/3623393.3623391"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.6028\/nist.sp.800-38d"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1109\/HPCA56546.2023.10071130"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.1007\/s11023-020-09548-1"},{"key":"ref11","volume-title":"Confidential computing: Secure your workloads with confidential vms","year":"2024"},{"key":"ref12","article-title":"The llama 3 herd of models","author":"Grattafiori","year":"2024","journal-title":"arXiv preprint"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1007\/s10836-014-5494-0"},{"key":"ref14","author":"Haakegaard","year":"2015","journal-title":"The elliptic curve diffie-hellman (ecdh)"},{"key":"ref15","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-030-57077-4_10"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.1145\/3297858.3304021"},{"key":"ref17","article-title":"Mixtral of experts","author":"Jiang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1007\/978-1-4842-6168-2_6"},{"key":"ref19","article-title":"Nemo: a toolkit for building ai applications using neural modules","author":"Kuchaiev","year":"2019","journal-title":"arXiv preprint"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.1109\/ISSCC42614.2022.9731739"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1145\/3719027.3765121"},{"key":"ref22","first-page":"1257","article-title":"Exploiting unprotected I\/O operations in AMD\u2019s secure encrypted virtualization","volume-title":"28 th USENIX Security Symposium (USENIX Security 19)","author":"Li","year":"2019"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.14778\/3415478.3415530"},{"key":"ref24","article-title":"The Galois\/Counter mode of operation (GCM)","author":"McGrew","journal-title":"NIST Modes of Operation submission, 2005, sec. 6, \u201cIn hardware, GCM adds a negligible amount of overhead compared to a pipelined AES implementation.\u201d"},{"key":"ref25","year":"2024","journal-title":"Gpu options for azure confidential computing"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.1109\/HPCA57654.2024.00025"},{"key":"ref27","year":"2022","journal-title":"NVIDIA Hopper Architecture In-Depth"},{"key":"ref28","year":"2022","journal-title":"Nvidia linux open gpu kernel module source"},{"key":"ref29","year":"2023","journal-title":"Confidential computing on h100 gpus for secure and trustworthy ai"},{"key":"ref30","year":"2024","journal-title":"Announcing nvidia secure ai general availability"},{"key":"ref31","year":"2024","journal-title":"Megatron distributed checkpoint"},{"key":"ref32","year":"2024","journal-title":"NCCL (NVIDIA Collective Communications Library)"},{"key":"ref33","year":"2024","journal-title":"Nvidia blackwell architecture: Advancing the era of generative ai"},{"key":"ref34","year":"2025","journal-title":"Nvidia trusted computing solutions release notes"},{"key":"ref35","author":"Rajbhandari","year":"2020","journal-title":"Zero: Memory optimizations toward training trillion parameter models"},{"key":"ref36","author":"Sakata","year":"2025","journal-title":"Announcing NVIDIA Secure AI General Availability"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1109\/PACT.2004.1342547"},{"key":"ref38","doi-asserted-by":"publisher","DOI":"10.1109\/ISCA.2005.30"},{"key":"ref39","article-title":"Megatron-lm: Training multi-billion parameter language models using model parallelism","author":"Shoeybi","year":"2019","journal-title":"arXiv preprint"},{"key":"ref40","article-title":"Chakra: Advancing performance benchmarking and co-design using standardized execution traces","volume":"arXiv:2305.14516","author":"Sridharan","year":"2023","journal-title":"arXiv preprint"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1109\/MDT.2007.179"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1145\/3669940.3707224"},{"key":"ref43","author":"Team","year":"2021","journal-title":"Issue #689: Nccl channel allocation discussion"},{"key":"ref44","author":"Toon","year":"2020","journal-title":"Introducing 2nd generation ipu systems for ai at scale"},{"key":"ref45","article-title":"Confidential computing within an AI accelerator","volume-title":"2023 USENIX Annual Technical Conference (USENIX ATC 23)","author":"Vaswani","year":"2023"},{"key":"ref46","doi-asserted-by":"publisher","DOI":"10.5555\/3291168.3291219"},{"key":"ref47","author":"Wan","year":"2025","journal-title":"Bytecheckpoint: A unified checkpointing system for large foundation model development"},{"key":"ref48","author":"Wang","year":"2024","journal-title":"Fastrack: Fast io for secure ml using gpu tees"},{"key":"ref49","doi-asserted-by":"publisher","DOI":"10.1109\/ISPASS57527.2023.00035"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.1109\/ISCA.2006.22"}],"event":{"name":"2026 IEEE International Symposium on High Performance Computer Architecture (HPCA)","location":"Sydney, Australia","start":{"date-parts":[[2026,1,31]]},"end":{"date-parts":[[2026,2,4]]}},"container-title":["2026 IEEE International Symposium on High Performance Computer Architecture (HPCA)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/11408404\/11408433\/11408582.pdf?arnumber=11408582","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,5]],"date-time":"2026-03-05T06:49:57Z","timestamp":1772693397000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11408582\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,1,31]]},"references-count":50,"URL":"https:\/\/doi.org\/10.1109\/hpca68181.2026.11408582","relation":{},"subject":[],"published":{"date-parts":[[2026,1,31]]}}}