But, no local hardware means constant SSH juggling. Upload code, compile remotely, run rocprof, download results, repeat. We were spending more time managing infrastructure than optimizing kernels.

We're currently competing in the GPU mode kernel optimization competition. Our goal is to build the fastest AMD kernels in the world. We spend a good chunk of our time setting up the infra to profile these kernels.

So we built Chisel internally to make GPU development feel local to us. One command spins up a droplet, syncs your code, runs profiling, and pulls results back. It handles the SSH, rsync, and teardown automatically. The profiling integration was the killer feature for us.

Since building Chisel, I have Claude constantly running experiments seamlessly and in parallel. I feel so much more productive when exploring different optimization strategies for these kernels.

Available on PyPI: pip install chisel-cli

Source: https://github.com/Herdora/chisel

Would love feedback and thoughts from the community, especially from people doing GPU work and exploring AMD alternatives to NVIDIA. If you're interested in contributing, we'd welcome any help in making chisel better.