{"id":98338,"title":"nCompass - Optimize performance on GPUs, 10x faster","tagline":"Automate performance optimization at all levels of the AI infrastructure stack","body":"Hey everyone,\n\n**tldr;**\n\nI’m Aditya, co-founder of [nCompass Technologies](https://ncompass.tech). We’re building a developer tool that unifies profiling, trace collaboration and trace analysis of AI systems. We automate performance optimization of AI systems across all levels of the infrastructure stack.\n\nUsing this tool, we implemented a Hopper GEMM kernel that outperformed NVIDIA's CUTLASS GEMMs by 3%, within a day - this took us months before.\n\nCheckout our product demo below. It’s free to use and you can get started today in **VS Code, Cursor or Claude Code** - [Quick Start](https://docs.ncompass.tech/quick-start)\n\n\u003chttps://www.youtube.com/watch?v=Q3Pq-BPU2Ec\u003e\n\n**THE PROBLEM**\n\nIdentifying the root cause of performance bottlenecks is 4-8x slower than writing the code to fix them.\n\nIf you are optimizing a system like vLLM, you have to:\n\n* Run a profile and then copy a giant trace file to your local machine just to view it.\n* Spend hours identifying opportunities for performance improvement.\n* If this involves writing a kernel, you profile the kernel, spend hours or days digging through ncu traces that are massive data dumps.\n* Then you identify your bottlenecks and formulate a plan.\n\nRunning this loop till you have a performant system can take weeks, even months.\n\n**OUR SOLUTION**\n\nBy building an AI agent that can analyze profiling data as well as interact with a bank of deep technical knowledge and expertise, we’re automating the process of identifying performance bottlenecks.\n\nNow in a single VSCode interface, you can:\n\n* Open and view trace files\n* Use our novel tools like trace diffs to analyze them\n* Generate share links to easily share traces with team members\n* Feed source + profiling data to our AI agent and get back actionable analysis on how you can optimize the performance of your system.\n\nThis applies to both systems and GPU kernel level analysis and our agent integrates directly into Cursor / Claude Code, so you never have to leave your normal workflow!\n\n**Anyone** can now write both correct and performant code with AI!\n\n**ASKS**\n\n**Install our VSCode extension** and start optimizing your systems performance!\n\n* [Quick Start](https://docs.ncompass.tech/quick-start)\n* [Detailed Docs](https://docs.ncompass.tech)\n\n**We also offer FDE services** - if you would like us to step in and analyze your system’s performance and provide you with an analysis of how much we could improve it by - reach out at [hello@ncompass.tech](https://mailto:hello@ncompass.tech)\n\n![uploaded image](/media/?type=post\u0026id=98338\u0026key=user_uploads/1641909/fc628a90-5565-43db-8aff-7c61e476e976)\n\n","slug":"Pa6-ncompass-optimize-performance-on-gpus-10x-faster","created_at":"2026-03-02T17:12:19.098Z","updated_at":"2026-05-25T05:51:17.453Z","total_vote_count":19,"url":"https://www.ycombinator.com/launches/Pa6-ncompass-optimize-performance-on-gpus-10x-faster","share_image_url":"https://www.ycombinator.com/media/?type=post\u0026id=98338\u0026key=user_uploads/1641909/fc628a90-5565-43db-8aff-7c61e476e976","company":{"id":29266,"name":"nCompass Technologies","slug":"ncompass-technologies","url":"https://www.ncompass.tech","logo":"https://bookface-images.s3.amazonaws.com/small_logos/ec22cfec2eb1602e4ae48a85862a05e4be149771.png","batch":"Winter 2024","industry":"B2B","tags":["Artificial Intelligence","Developer Tools","Hardware","Open Source"],"search_path":"https://bookface.ycombinator.com/company/29266"}}