alt Hacker NewsA look under our trunk: what's in our compute
Comments
The more I think about it the more autonomous cars look like extremely demanding edge computers. You have limited power and cooling, unreliable connectivity and very strict latency requirements. It's probably one of the few places where designing the hardware around the workload makes more sense than trying to make a general purpose platform do everything
They really are the impressive tech, it's really a shame that they aren't the one capturing all the media spotlight.
I don't know if we'll get there eventually, but if we do, I won't be surprised if it's because of Waymo.
Very cool. To be honest, I'd assumed they were using commodity hardware, but no real reason for that. Cool chip. It is quite interesting that so many large companies have their own chips for these things. I almost managed to get my hands on a couple of accidentally liquidated MTIA chips but I was too slow. I wonder what the chip is like and what the dev experience is writing for it.
It would just be cool to see what these guys have cooked up dev tools wise since Google has always been famed for good developer experience (often to such a degree that Xooglers are useless outside the exosuit).
It really doesn't say much. Tesla had more detailed talks on their HW and AI stack. This reads like a glorified marketing blurb.
Interesting that the article uses "ML" everywhere instead of "AI". I'm sure some marketing person chose that term deliberately, and I wonder what went into that decision.
>high-fidelity data from 13 high-resolution cameras
is it me or their example photo doesn't look even like 1 high-resolution camera (and very noisy at that, looks like a pretty high ISO setting). Their example looks similar and i'd say a bit worse than what my Fuji camera did 25 years ago in the dark after i'd adjust dynamic range of the image in GIMP. (and the cameras on the Waymo cars i see around have about the same lens size as my Fuji)
If it is really a showcase of their visual pipeline than it makes sense that they are still so reliant on lidar - which is pretty expensive because of the number of rays you have to support to maintain good situation-awareness and which is still hard and expensive to scale to get good resolution. That also means that their tech can't really be put onto other platforms like drones for example.
Custom silicon while it still looks like they've got multiple gaming PC's worth of compute gear mass isn't a flex when you remember what Tesla has with their FSD computer comparably just sipping power, while processing signals that are less straightforward as unscalable LiDAR
they now are doing a 5nm asic, what did they have before this?
Compute is under the trunk ??? Noted. At the price of RAM nowadays it's like the new copper !
Apologies for repeating myself, but it's hard to convey how far ahead Waymo is in every area: sensors, vehicles, training data, markets served, simulation, infrastructure, operations, regulatory.
They're growing fast, but I do wonder what they see as the bottlenecks: municipal approval, training, manufacturing? It's really unfortunate that lobbying is slowing self-driving rollout in some of the biggest and most important cities: NYC, Chicago, DC.