Innovative CUDA to ROCm Migration System—Achieve Seamless TransitionInnovative CUDA to ROCm Migration System—Achieve Seamless Transition
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Built an autonomous CUDA → ROCm migration platform in 5 days. Kernel Olympics came out of the AMD Developer Hackathon 2026 (Jul 8–13). It's a multi-agent system that analyzes entire CUDA repositories, plans migration strategies, ports kernels to HIP, compiles with hipcc, and runs on real AMD MI300X hardware.
Not a wrapper around hipify. The platform reasons about architecture, dependencies, and unsupported APIs before touching a single line. 665 tests passing, ~$0.09 per LLM run, pattern cache hitting at 0.2ms.
Stack: Python, FastAPI, OpenAI, CUDA, ROCm. Full case study with architecture breakdown and live demo: Kernel Olympics on Contra
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started