Editorial illustration for Rivian Develops Custom AI Chips for Safer, More Efficient Electric Vehicle Driving
Rivian's Custom AI Chips Boost EV Autonomous Driving
Rivian builds AI chips for driving with efficiency/performance, ASIL compliance
Rivian is now building its own brain. The electric truck company, best known for its adventure-ready pickups, has decided the most important part of a self-driving car shouldn't come from a catalog. It is designing its own AI chips, a move that plunges it into one of the most capital-intensive and technically fraught arenas in tech.
The goal is a processor that meets the automotive industry's highest safety standard while crunching data at staggering rates. This isn't an accessory. It's the core of Rivian's attempt to control its own destiny, a direct challenge to the Nvidia-powered future the rest of the auto world is quietly accepting.
Rivian says the chip's architecture will deliver "advanced levels of efficiency, performance, and Automotive Safety Integrity Level compliance," referencing a risk classification system for safety-critical automotive electronics. Rivian estimates its neural engine can perform 800 trillion operations a second (TOPS) while its third generation computer can do 1,600 trillion 8-bit integer operations per second (INT8 TOPS) while utilizing data sparsity. For comparison, Nvidia's H100 class GPUs are quoted at 3,000-3,900 INT8 TOPS on datasheets with sparsity, while Google's TPU v5e per-chip INT8 number is estimated 393 INT8 TOPS.
(Google recently announced its seventh-generation TPUs capable of over 40 exaflops in clustered pods.) Rivian estimates its AI chip can perform 1,600 trillion 8-bit integer operations per second (INT8 TOPS) while utilizing data sparsity. It has a processing power of 5 billion pixels per second. And it features RivLink, a low latency interconnect technology that allows chips to be connected to multiply processing power.
The processor is also enabled by an in-house developed AI compiler and platform software. Most notably, the announcement of the proprietary silicon aligns Rivian with Tesla, the other major automaker that has been trying to brute-force its way to self-driving cars by making its own chips, while the rest of the auto industry increasingly lines up behind Nvidia. Rivian is an EV-only manufacturer, just like Tesla, and has said that vertical integration is a key element to its future growth.
Rivian will use a variety of sensors to power its autonomous driving, including lidar.
The raw specs are a flex. 1,600 trillion operations. Five billion pixels per second.
But the benchmark that matters is ASIL-D compliance, the automotive world's guarantee of fail-safe operation. This turns a fast chip into a certified component you can bet a life on. The decision to go in-house, to develop the compiler and the RivLink interconnect, is a massive bet on vertical integration.
It is expensive. It is difficult. It is exactly what Tesla did.
Rivian seems to believe that in the long run, owning the stack is cheaper than paying a premium for someone else's black box and waiting in line for their updates. This is a declaration of technological independence. Whether it's a masterstroke or a misstep depends entirely on execution.
The road to autonomy is littered with good silicon that arrived too late, or cost too much, or simply didn't work as promised outside a lab. Rivian is now building that road itself.
Related Reading
Industry Applications Westinghouse teams with Google Cloud to build AI platform for nuclear power
Industry Applications NVIDIA NeMo powers telco reasoning model for autonomous network workflows
Industry Applications Month-1 Agent Adds Holistic Observability with Trace IDs and Token Tracking
From Archives Two Men Tied to China's Salt Typhoon Hackers Likely Trained at Cisco Academy
From Archives FastAPI tutorial installs fastapi, uvicorn, scikit-learn, pydantic, joblib
Common Questions Answered
How many operations per second can Rivian's neural engine perform?
Rivian's neural engine is capable of performing 800 trillion operations per second (TOPS). This impressive computational power is designed to enhance the processing of complex driving scenarios in electric vehicles.
What is Rivian's approach to automotive safety with their custom AI chips?
Rivian is targeting Automotive Safety Integrity Level (ASIL) compliance with their custom AI chips, demonstrating a methodical approach to integrating advanced computational capabilities with rigorous automotive safety standards. The chip's architecture is specifically designed to deliver advanced levels of efficiency, performance, and safety.
How do Rivian's custom AI chips compare to existing automotive computing technologies?
Rivian's third-generation computer can perform 1,600 trillion 8-bit integer operations per second (INT8 TOPS) while utilizing data sparsity. This represents a significant advancement in automotive computing power, potentially setting a new standard for how electric vehicles process complex driving scenarios.
Further Reading
- Rivian’s big bet on custom chips and ‘universal’ hands-free driving — TechCrunch
- Rivian goes all in on ‘universal hands-free’ driving at its first Autonomy and AI Day — Engadget
- Rivian unveils custom silicon and next-gen autonomy platform with deep AI integration — Rivian Newsroom
- Inside Rivian’s autonomy roadmap: in-house AI silicon, safety, and scalable driver assistance — The Verge
- Why EV makers like Tesla and Rivian are building their own AI chips for self-driving — Reuters
