Quix about:
Tesla Fsd Computer: How 144 Tops Enables Real-time Neural Network InferenceTechnical analysis of Tesla's Full Self-Driving computer architecture, examining neural processing unit design, memory bandwidth, and real-time inference optimization.
OPEN
405 views
FSD computer targets 10^9 hour MTBF, but robotaxis need 10^12+ hours for true autonomy. Is the dual-chip redundancy sufficient?
Additional Context
Current automotive reliability is 10^6-10^7 hours MTBF. Robotaxis operating 24/7 need aircraft-level reliability (10^12 hours). Triple redundancy with voting might be required. Tesla's betting on software validation rather than hardware redundancy.
Asked by:
Pete Bannon•Director of Autopilot Hardware, Tesla
Pete Bannon•Director of Autopilot Hardware, TeslaResponses (1)
Please sign in to respond to this quix
PB
Pete Bannon
•Director of Autopilot Hardware, TeslaFrom what I've seen, dual redundancy probably isn't enough for robotaxis. Aviation uses triple redundancy with voting logic for flight-critical systems. Tesla's approach relies heavily on software validation - proving the system is correct before deployment rather than hardware fault tolerance. This works for aircraft but cars operate in more unpredictable environments. They'll likely need more hardware redundancy.