NVIDIA: In gigantic AI clusters, chips are already failing every minute - now Huang invests in the solution

In AI clusters with hundreds of thousands of processors, hardware failures are now occurring somewhere in the system every minute. This makes fault tolerance the new infrastructure problem. NVIDIA is participating in a $125 million financing round for iPronics. The startup is developing optical switches that can automatically reroute data streams in the event of failures. The larger the GPU clusters become, the more important Optical Circuit Switching, Silicon Photonics, and a dynamically reconfigurable AI infrastructure will be.  ...