INFRA Signal 264
Waymo publishes its own compute architecture with custom 5nm ASIC and seven suppliers
↗
Waymo published the architecture of its onboard computer, introducing a custom 5nm ASIC for front-end sensor processing and naming seven suppliers.
For engineers, this is a rare look at Waymo's compute stack, showing that the custom chip is a front-end processor, not the decision-maker. The heterogeneous system still relies on Nvidia and AMD parts, so the ASIC adds capability rather than replacing existing hardware. The redundancy design, with two independent engines, is a notable safety approach for a system with no human backup.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
The custom 5nm ASIC handles raw sensor data from lidar, radar, and cameras, including temporal denoising, before feeding into the ML inference engine.
Waymo pairs its own silicon with CPUs, GPUs, and accelerators from Nvidia, AMD, and others, so the custom chip is additive, not a replacement.
The compute is designed as two independent engines that run parallel workloads and take over if one fails, with no human backup.
THE CLUSTER