OpenAI’s chip chief tells More Than Moore he doesn’t fully trust AI design

The Next Web
Read the full article
OpenAI's hardware lead Richard Ho revealed that their AI models autonomously optimized parts of the Jalapeño chip design, saving over 13% die area, though engineers did not always understand the changes. The chip, co-developed with Broadcom, runs AI models and unifies inference stages to adapt to varying workloads, with all modifications fully validated despite some unexplained model-driven changes.

More in Chips & Compute

Microsoft Prepares to Put Gigawatts of Its Chips in Data Centers

Covered by 5 sources
Chips & Compute5 min read

Celestica Unveils DS6005 Co-Packaged Optics Switch for AI Data Centers

Unite.AI

Amazon in Talks to Buy AI Startup Decart for Around $7 Billion

Covered by 2 sources