{"690611":{"#nid":"690611","#data":{"type":"news","title":"A Common Language to Understand AI Systems","body":[{"value":"\u003Cp\u003EThere\u2019s a simple idea that shows up in just about every engineering discipline: you can\u2019t improve what you can\u2019t measure.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThat principle is especially relevant today across the artificial intelligence (AI) landscape. As systems scale, they increasingly become harder to measure, compare, and fix, particularly within proprietary environments.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EA team led by Georgia Tech, working with collaborators across industry, has developed a new approached called \u003Ca href=\u0022https:\/\/mlcommons.org\/working-groups\/research\/chakra\/\u0022 rel=\u0022noreferrer\u0022 title=\u0022(opens in a new window)\u0022\u003E\u003Cstrong\u003EChakra\u003C\/strong\u003E\u003C\/a\u003E to bring greater clarity to complex AI systems.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cImagine a room where everyone is trying to collaborate, but each person speaks a different language,\u201d said \u003Ca href=\u0022https:\/\/ece.gatech.edu\/directory\/tushar-krishna\u0022\u003E\u003Cstrong\u003ETushar Krishna\u003C\/strong\u003E\u003C\/a\u003E, an associate professor in the \u003Ca href=\u0022https:\/\/ece.gatech.edu\/\u0022\u003E\u003Cstrong\u003ESchool of Electrical and Computer Engineering\u003C\/strong\u003E\u003C\/a\u003E, who is leading the effort. \u201cThat\u2019s a bit like today\u2019s AI ecosystem. The internet worked because it was built on shared practices and protocols. In AI, we\u2019re still building that kind of common foundation.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThe work, which Krishna leads through the nonprofit \u003Ca href=\u0022https:\/\/mlcommons.org\/\u0022 rel=\u0022noreferrer\u0022 title=\u0022(opens in a new window)\u0022\u003E\u003Cstrong\u003EMLCommons\u003C\/strong\u003E\u003C\/a\u003E, was released \u003Ca href=\u0022https:\/\/arxiv.org\/abs\/2605.11333\u0022 rel=\u0022noreferrer\u0022 title=\u0022(opens in a new window)\u0022\u003E\u003Cstrong\u003Ealongside a paper\u003C\/strong\u003E\u003C\/a\u003E at the \u003Ca href=\u0022https:\/\/mlsys.org\/\u0022 rel=\u0022noreferrer\u0022 title=\u0022(opens in a new window)\u0022\u003E\u003Cstrong\u003E2026 Conference on Machine Learning and Systems\u003C\/strong\u003E\u003C\/a\u003E(MLSys) in Bellevue, Wash.\u003C\/p\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cp\u003E\u003Cstrong\u003EUnderstanding Systems Without Exposing Them\u003C\/strong\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003ECloud companies, chip designers, software developers, and infrastructure providers all describe their systems differently, relying largely on internal, proprietary approaches that are not publicly shared.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThis slows innovation, reduces efficiency, and increases the cost of running AI at scale.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EChakra, named after the Sanskrit word for \u201cwheel\u201d to reflect a continuous cycle of improvement, is designed around that reality. Its release is not a single finished system, but a set of shared tools and building blocks.\u0026nbsp;\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EResearchers are making available a standardized format for representing AI workloads, along with tools for collecting and analyzing data from what\u2019s known as an execution trace.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cAn execution trace is essentially a recording of how an AI system behaves,\u201d Krishna said. \u201cRather than focusing only on outcomes like speed or accuracy, it captures what computations happened, when machines needed to communicate, and where delays or bottlenecks occurred.\u201d\u0026nbsp;\u003C\/p\u003E\u003C\/div\u003E\u003C\/div\u003E\u003C\/div\u003E\u003C\/div\u003E\u003C\/div\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cdiv\u003E\u003Cp\u003EThose traces don\u2019t expose the underlying code or data. Instead, they reflect patterns of behavior.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThose traces don\u2019t expose the underlying code or data. Instead, they reflect patterns of behavior.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cIt\u2019s a bit like sharing a map of traffic patterns in a city, instead of handing over the blueprints for every building,\u201d Krishna said.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThe approach can also be used to explore how future systems might behave, giving researchers a way to test ideas and identify potential bottlenecks before those systems are built.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cAll of this dramatically lowers the barrier to participating in AI systems innovation,\u201d Krishna said.\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003EBuilding a Shared Standard\u003C\/strong\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThe Chakra project began in 2023 as a collaboration between Georgia Tech and Meta, building on parallel efforts to better understand how AI workloads behave across production systems and simulation environments.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EPart of that work built on \u003Ca href=\u0022https:\/\/astra-sim.github.io\/\u0022 rel=\u0022noreferrer\u0022 title=\u0022(opens in a new window)\u0022\u003E\u003Cstrong\u003EASTRA-sim\u003C\/strong\u003E\u003C\/a\u003E, an open-source distributed AI system simulator developed and maintained by Krishna\u2019s group, which models how large-scale AI systems perform across hardware and software.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cWe knew that for AI to scale responsibly, we needed better ways to understand what\u2019s happening under the hood,\u201d Krishna said. \u201cCompanies struggle to compare systems fairly or reproduce why something worked well\u2014or failed\u2014because everyone uses different tools and proprietary setups.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThe early collaboration expanded into a broader effort called the Chakra Working Group (CWG) within MLCommons, a consortium that brings together companies and researchers to develop shared benchmarks and standards for AI systems, including widely used efforts like MLPerf.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EDavid Kanter, co-founder of MLCommons and head of MLPerf, has praised the group for \u201cdefining an industry roadmap for AI workload tracing support and benchmarking.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EToday, CWG includes industry partners such as NVIDIA, AMD, Meta, HPE, and Keysight, along with contributions from multiple Georgia Tech faculty, students, and alumni (seven of whom are now working across partner organizations).\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cChakra is a fantastic showcase of the role ECE and Georgia Tech play in connecting academic research with real-world systems,\u201d said \u003Ca href=\u0022https:\/\/ece.gatech.edu\/directory\/arijit-raychowdhury\u0022\u003E\u003Cstrong\u003EArijit Raychowdhury\u003C\/strong\u003E\u003C\/a\u003E, Steve W. Chaddick School Chair of ECE. \u201cWe can bring together expertise spanning the full AI stack in really the only way that makes complex work like this possible.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThat level of collaboration is essential to developing something that can be used across the broader AI ecosystem, according to Winston Liu, a chief architect at Keysight Technologies and a member of CWG.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cWhat the Chakra community has built is meaningful, but the collaboration model that produced it is worth recognizing just as much,\u201d he said. \u201cThat combination\u2014early enough to shape the spec together and open enough that the output belongs to everyone\u2014is genuinely rare.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003EA Real-world Testbed at Georgia Tech\u003C\/strong\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EMuch of the team\u2019s work has depended on access to infrastructure capable of running AI systems at a realistic scale. Georgia Tech has built that capability through its \u003Ca href=\u0022https:\/\/coe.gatech.edu\/academics\/ai-for-engineering\/ai-makerspace\u0022\u003E\u003Cstrong\u003EAI Makerspace\u003C\/strong\u003E\u003C\/a\u003E, one of the largest computing clusters in the world dedicated to supporting student-driven AI workloads while also serving as a real-world testbed for large-scale systems research.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EIn collaboration with the \u003Ca href=\u0022https:\/\/pace.gatech.edu\/\u0022\u003E\u003Cstrong\u003EPartnership for an Advanced Computing Environment\u003C\/strong\u003E\u003C\/a\u003E (PACE), CWG researchers utilized the AI Makerspace to run workloads across 128 advanced GPUs and collect execution traces from systems operating under real conditions.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cThe AI Makerspace was built on a simple belief: AI should be accessible to as many as possible,\u201d said \u003Ca href=\u0022https:\/\/coe.gatech.edu\/directory\/person\/matthieu-bloch-phd\u0022\u003E\u003Cstrong\u003EMatthieu Bloch\u003C\/strong\u003E\u003C\/a\u003E, associate dean in the College of Engineering. \u201cIt\u2019s exciting to see our colleagues using it to amplify impact and give back to the broader community.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThat level of access allowed the work behind Chakra to move beyond theory and into environments where performance challenges actually emerge.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EIn one case study, Chakra helped identify a hidden communication bottleneck that only appeared under realistic conditions when different types of workloads were running at the same time. More simplified tests failed to surface the issue.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003EWhat Comes Next\u003C\/strong\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EAs the Chakra tools and standards are released, the focus now turns to how they will be adopted and extended.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EKrishna sees the current moment less as a finish line and more as a starting point for broader participation across the field.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cFive years from now, Chakra will help make AI systems development dramatically more reproducible and accessible,\u201d he said. \u201cResearchers could test ideas against realistic workloads without needing access to massive datacenters, and companies could identify problems much earlier in the design process.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EAs AI infrastructure grows more costly, the ability to model new system designs allows researchers and companies to make informed decisions before committing to large-scale investments.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cLonger term, it could move us toward a \u2018digital twin\u2019 of AI infrastructure,\u201d Krishna said. \u201cA way to model and optimize systems before they\u2019re ever built.\u201d\u003C\/p\u003E\u003C\/div\u003E\u003C\/div\u003E\u003C\/div\u003E\u003C\/div\u003E\u003C\/div\u003E","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003E\u003Cem\u003ELike the internet before it, AI systems need shared standards to work together. Tushar Krishna and industry collaborators have released Chakra, a new set of tools designed to help make that possible.\u0026nbsp;\u003C\/em\u003E\u003C\/p\u003E","format":"limited_html"}],"field_summary_sentence":[{"value":"Like the internet before it, AI systems need shared standards to work together. Tushar Krishna and industry collaborators have released Chakra, a new set of tools designed to help make that possible. "}],"uid":"36172","created_gmt":"2026-06-03 14:33:16","changed_gmt":"2026-06-03 14:37:18","author":"dwatson71","boilerplate_text":"","field_publication":"","field_article_url":"","location":"Atlanta, GA","dateline":{"date":"2026-06-03T00:00:00-04:00","iso_date":"2026-06-03T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"680409":{"id":"680409","type":"image","title":"DSC05583.jpg","body":"\u003Cp\u003EAssociate Professor Tushar Krishna (center) and members of his research team \u2014 William Won (recently graduated, now at AMD), Changhai Man, Hanjiang Wu, and Jinsun Yoo \u2014 have announced Chakra, a new shared platform for understanding and improving complex AI systems.\u003C\/p\u003E","created":"1780497224","gmt_created":"2026-06-03 14:33:44","changed":"1780497224","gmt_changed":"2026-06-03 14:33:44","alt":"Associate Professor Tushar Krishna (center) and members of his research team \u2014 William Won (recently graduated, now at AMD), Changhai Man, Hanjiang Wu, and Jinsun Yoo \u2014 have announced Chakra, a new shared platform for understanding and improving complex AI systems.","file":{"fid":"264665","name":"DSC05583.jpg","image_path":"\/sites\/default\/files\/2026\/06\/03\/DSC05583.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2026\/06\/03\/DSC05583.jpg","mime":"image\/jpeg","size":1141078,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2026\/06\/03\/DSC05583.jpg?itok=Y0JL6Acd"}}},"media_ids":["680409"],"groups":[{"id":"1188","name":"Research Horizons"}],"categories":[{"id":"145","name":"Engineering"},{"id":"194609","name":"Industry"},{"id":"135","name":"Research"}],"keywords":[{"id":"195159","name":"Chakra"},{"id":"173453","name":"Tushar Krishna"},{"id":"193101","name":"MLCommons"},{"id":"195160","name":"AI systems"},{"id":"11444","name":"benchmarking"},{"id":"195161","name":"execution trace"},{"id":"170894","name":"standards"},{"id":"185447","name":"ASTRA-sim"},{"id":"195162","name":"AI Makerspace"},{"id":"139771","name":"Arijit Raychowdhury"}],"core_research_areas":[{"id":"39431","name":"Data Engineering and Science"},{"id":"39541","name":"Systems"}],"news_room_topics":[],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003EDan Watson\u003C\/p\u003E","format":"limited_html"}],"email":["dwatson71@gatech.edu"],"slides":[],"orientation":[],"userdata":""}}}