Ex-NVIDIA engineer tells the story of testing a chip with a broken memory controller
blelbach · x · 2026-10-07
An engineer shares a hardware war story: a chip came back from the fab with a completely busted memory controller. NVIDIA hates respins, so before deciding on one, the team needed to verify whether anything else was wrong.
Their solution: they ran the full bring-up test suite directly out of L2 cache with no main memory attached, isolating the faulty memory controller while validating the rest of the silicon. A rare glimpse into extreme measures chip teams take to avoid a costly respin.
More from Fun
- Runway unveils 'What We Forgot,' an AI film set in 2030 looking back at pre-AI life — c_valenzuelab · 2026-10-07
- Three types of people vibe coding SaaS clones: nerds, broke devs, and paid shills — kylegawley · 2026-10-07
- 'Below n log n' Claim Exposed: It's Actually n(log n)^0.54999... — BLUECOW009 · 2026-10-07
- Your test suite costs $10 per run: the hidden CI bill of PR maxers — zeeg · 2026-10-07
- Opus 5.5 updates a classic meme after integer multiplication beats n log n — ctjlewis · 2026-10-07
- Beff Jezos quips: human civilization is a hierarchy of meat proxies for the Kardashev scale — beffjezos · 2026-10-07