Speeding up GPU kernels by 38% with a multi-agent system · Cursor | Cursor | Blogs | TechiSeek

AddAdd your Listings

Blogs

Speeding up GPU kernels by 38% with a multi-agent system · Cursor

Our multi-agent system autonomously optimized 235 CUDA kernels for NVIDIA Blackwell 200 GPUs, achieving a 38% geomean speedup over baselines in just 3 weeks.

Blogs Cursor

BlogsInteract with agent-created visualizations in canvases · CursorCursor BlogsContinually improving our agent harness · CursorCursor BlogsHow we set up our cloud agent environment · CursorCursor BlogsGoverning agent autonomy with Auto-review · CursorCursor BlogsCursor earns AIUC-1 certification for agent security and reliability · CursorCursor BlogsAgent swarms and the new model economics · CursorCursor BlogsHow Layers Slashed Analytics Costs and Gave Every Customer a Private Data WarehouseMotherduck EventsQuacking the Code to Multi-Tenant Embedded Analytics with GoodData & MotherDuckMotherduck Newsnference Accelerates Self-Supervised Language Model Training with Cerebras CS-2 System - CerebrasCerebras NewsCerebras Systems Sets Record for Largest AI Models Ever Trained on A Single Device - CerebrasCerebras NewsAleph Alpha Selects Cerebras to Build Next-Gen Sovereign AI Models - CerebrasCerebras BlogsAnnouncing the Cerebras Architecture for Extreme-Scale AI - CerebrasCerebras BlogsVary This, Vary That - Introducing New Custom Query Parameter Vary Cache SettingsBunny ResourcesVercel QueuesVercel BlogsCollaborating with Anthropic on Claude Sonnet 4.5 to power intelligent coding agentsVercel

BlogsIntroducing Vercel AI SDK 3.2Vercel ResourcesHTTP instrumentation configurationOpentelemetry ResourcesOpenTelemetry Tracing ShimOpentelemetry LearnSignal Handling in RubyBetterstack NewsBiome Announces v2.0 Beta with Plugin System and Type Information SupportSocket