{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/gpu-acceleration-for-large-scale-tree","title":"GPU-acceleration for Large-scale Tree Boosting","arxiv_id":"1706.08359","date":"2017-06-26","proceeding":null,"authors":["Huan Zhang","Si Si","Cho-Jui Hsieh"],"abstract":"In this paper, we present a novel massively parallel algorithm for\naccelerating the decision tree building procedure on GPUs (Graphics Processing\nUnits), which is a crucial step in Gradient Boosted Decision Tree (GBDT) and\nrandom forests training. Previous GPU based tree building algorithms are based\non parallel multi-scan or radix sort to find the exact tree split, and thus\nsuffer from scalability and performance issues. We show that using a histogram\nbased algorithm to approximately find the best split is more efficient and\nscalable on GPU. By identifying the difference between classical GPU-based\nimage histogram construction and the feature histogram construction in decision\ntree training, we develop a fast feature histogram building kernel on GPU with\ncarefully designed computational and memory access sequence to reduce atomic\nupdate conflict and maximize GPU utilization. Our algorithm can be used as a\ndrop-in replacement for histogram construction in popular tree boosting systems\nto improve their scalability. As an example, to train GBDT on epsilon dataset,\nour method using a main-stream GPU is 7-8 times faster than histogram based\nalgorithm on CPU in LightGBM and 25 times faster than the exact-split finding\nalgorithm in XGBoost on a dual-socket 28-core Xeon server, while achieving\nsimilar prediction accuracy.","url_abs":"http://arxiv.org/abs/1706.08359v1","url_pdf":"http://arxiv.org/pdf/1706.08359v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"gpu-acceleration-for-large-scale-tree","repo_url":"https://github.com/cdg1921/LightGBM","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"gpu-acceleration-for-large-scale-tree","repo_url":"https://github.com/ibr11/LightGBM","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}},{"paper_slug":"gpu-acceleration-for-large-scale-tree","repo_url":"https://github.com/perfgao/lua-ffi-lightGBM","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"unanswered"}}],"tasks":[{"task_slug":null,"task_name":"CPU"},{"task_slug":null,"task_name":"GPU"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1706.08359","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}