NVIDIA engineer reflects on 9 years: rebuilding Thrust and CUB into the CCCL team
blelbach · x · 2026-09-11
NVIDIA engineer blelbach marks nine years at the company, a move made against advice that its compensation was below market. Convinced by people he met on the C++ committee that SIMT and Volta's independently scheduled threads were the future of GPUs, he joined after an 8-round interview.
Highlights:
- As a modern C++ person, he landed in the CUDA driver org, which was written almost entirely in C.
- On day one he was handed Thrust and CUB, NVIDIA's first two open source libraries, with no PM and no active maintainers. He built a roadmap, revived the GitHub repos, and started sending updates under the name of a team that didn't yet exist — CCCL (CUDA Core Compute Libraries) — until he got headcount for three people.
- The team built libcu++ and abstractions for new chip features serving rapidly growing AI workloads.
- He later moved to guiding HPC C++ compiler roadmap, spanning stdpar, CUDA-Q, CUDA Python, CUDA Tile, and autoresearch.
Related event: NVIDIA Engineer Marks Nine Years from CUDA Driver Team to CCCL(2 posts)→
More from Companies & People
- Anne Griffin hosts 20-person SF salon on the product decisions behind AI agents — annetgriffin · 2026-10-28
- dhh: Windows Copilot key was so bad it may have seeded mainstream AI backlash — mhdfaran · 2026-10-28
- Thermo-backed CTO says AI is in its dial-up phase, touts disaggregated compute for an ultrafast 'broadband moment' — beffjezos · 2026-10-28
- Design team: AI now produces the work overnight, humans just make the calls — thesephist · 2026-10-28
- Nonprofit hiring Director of AI to build foundation models on 11M+ genomes — anshulkundaje · 2026-10-28
- India Inc hunts AI talent to transform manufacturing — nordicinst · 2026-10-28