Waymoが5nmの自社設計チップを公開、処理性能は1000TOPS超

"To handle the massive influx of raw data before it reaches our core ML brain, we are excited to introduce our purpose-built 5nm ASIC. While this chip is just one of several exciting custom components we’re developing, it is a specialized ML powerhouse engineered exclusively to process, fuse, and run advanced neural networks on raw sensor data in real time."（Waymo's purpose-built 5nm ASIC 節）— Waymo 公式ブログより

"The ASIC’s specialized accelerators instantly extract critical information from raw lidar, radar, and camera streams, including temporal denoising for superior low-light perception."／"Our latest system (right) processes high-fidelity data from 13 high-resolution cameras simultaneously and in real time, delivering exceptional low-light perception and revealing critical environmental details that traditional cameras (left) miss."（Waymo's purpose-built 5nm ASIC 節）— Waymo 公式ブログより

"Developing silicon, sensors, and algorithms side-by-side allows us to push the boundaries of sensor fidelity, bandwidth efficiency, and quantization to execute a heterogeneity of models from sparse convolutions to dense transformers."／"This data feeds into our purpose-built inference engine to run sensor fusion ML models, enabling greater efficiency without sacrificing fidelity."（Waymo's purpose-built 5nm ASIC 節）— Waymo 公式ブログより

"While these ASICs alone deliver over 1,000 TOPS of ML performance dedicated to front-end processing and ML models, we optimize across the full stack to maximize achieved performance, especially in the low-batch regimes we often operate."（Waymo's purpose-built 5nm ASIC 節）／"We built an ML-primary architecture to run advanced neural networks at minimal latency. To manage critical non-ML tasks like orchestration, data movement, and logging while maximizing time for ML computation, we pair our ML technologies with the best CPUs, GPUs, and accelerators. The result is a balanced, heterogeneous system."（custom systems 節）— Waymo 公式ブログより

"Drawing from more than 200 million miles of fully autonomous experience, we engineer our compute around three non-negotiable requirements:"／"We have engineered our stack for ultra-low latency, minimizing the delay from first pixel to action."／"Unlocking this requires impressive raw compute power—which we’ve scaled 20x in just eight years—paired with a deeply optimized software stack to harness it efficiently."／"By integrating directly with the vehicle's liquid cooling system, we sustain peak performance whether navigating freezing Midwest winters or the blistering heat of Phoenix."／"Our compute is designed like two independent engines. While they normally operate as one unit running full parallel workloads, if one experiences a fault, the other seamlessly takes over."（three non-negotiable requirements 節）／"Evolving from off-the-shelf components to custom systems required rethinking our physical and architectural design. Our highly integrated system delivers immense processing power without compromising the rider experience, maximizing battery efficiency while preserving ample trunk space and running silently."（custom systems 節）— Waymo 公式ブログより

"Along with our custom silicon efforts, we partner closely with industry leaders whose world-class computing solutions provide the powerful foundation needed to scale our technology efficiently. We are proud to work alongside a number of partners like AMD, Micron, NVIDIA, Samsung, Sandisk, Socionext, and TSMC to deliver the most capable autonomous computing system."（partners 節）— Waymo 公式ブログより

"If you're interested in learning more about Waymo's approach to compute, join us at our talks at Hot Chips"（末尾の案内文）— Waymo 公式ブログより