nCompass Performance Optimization IDE
Optimize performance on GPUs - 10x faster
168 followers
Optimize performance on GPUs - 10x faster
168 followers
nCompass helps you debug and improve the performance of your GPU system - 10x faster. Our VSCode extension allows you to run novel tools like trace diffs and our agent can analyze the torch.profile / nsys / ncu trace files to surface performance bottlenecks and optimization strategies that are guided by runtime data. We integrate with Claude Code and Cursor to save you weeks of work normally spent battling a fragmented set of tools and large dumps of raw trace data.
nCompass Performance Optimization IDE
Hey PH! Aditya here, co-founder of nCompass.
My team and I have spent years accelerating GPU kernels. It’s complex and time consuming, primarily because we spent most our time identifying which kernel to optimize and then identifying what the bottlenecks were in our new kernel. Using the dev tool we’re launching today, we implemented a Hopper GEMM kernel that outperformed NVIDIA's CUTLASS GEMMs by 3%, within a day - this took us months before.
Here's the problem today: if you profile a system like vLLM, you have to copy a giant trace file to your local machine just to view it. Then you spend hours identifying which GPU kernels to target. Then you profile the kernel, spend hours or days digging through .ncu traces that are massive data dumps. Then you identify your bottlenecks and formulate a plan. Running this loop till you have a performant kernel can take weeks, even months before you have a performant kernel.
All of this is going to change. Coding agents have gotten great at writing GPU kernels once they know what to target. What's missing is the tooling to identify these targets.
We built the tool we wish we had. With our VSCode extension, you can view traces natively — no more copying files around. Once our MCP is integrated, just "@" an execution trace into Claude Code / Cursor and our agent takes over: our agent analyzes the trace and gives you a prioritized list of GPU kernels to target. Using NCU, profile your kernel source code, "@" the .ncu-rep in your favorite coding agent, and the nCompass agent works together with your coding agent to develop the high performing kernel you were looking to build!
If you're an experienced GPU engineer — this removes the busywork. You stay in control, vet the agent's outputs, and only act on what makes sense.
If you're learning GPU optimization — you now have an expert pair-programmer you can always consult. It's the best way to learn how to optimize GPU code.
We'd love for you to try it and let us know how it works for you. We're building fast and your feedback directly shapes what ships next.
Try it → https://docs.ncompass.tech/quick...
Full feature list → https://docs.ncompass.tech