Dutchy Ron schreef op 14 oktober 2017 18:02:
It's Nvidia inside GM/Cruise self-driving Bolts
For some time, I've suspected that Nvidia's (NVDA) Drive PX 2, or some component of it, was at the heart of the General Motors (GM) self-driving Chevy Bolts being prepared by its Cruise Automation acquisition in San Francisco. Recently, I came across this job posting by Cruise, which I think is virtually a smoking gun:
static.seekingalpha.com/uploads/2017/...CUDA is the Nvidia proprietary programming environment for general computation on the GPU. Here, the requirements are much more focused on GPU programming via CUDA than in previous listings I have seen.
Most importantly, the Responsibilities section points to two key areas in which CUDA processing on the GPU is used. One is the vehicle's “perception system,” which I take to mean object recognition and classification, and which could also include local 3D mapping of objects via LIDAR data. The other important function is deep learning, which is also “to be deployed on GPU."
Deep Learning is the classic AI function that Nvidia has advocated for its GPUs, and Nvidia has developed many Deep Learning APIs that are part of DriveWorks, its software to support autonomous vehicle operation on Drive PX 2.
The job listing also implies strongly that GPU processing has become central to the Cruise approach. The Senior GPU Engineer is to “identify bottlenecks and optimize existing CUDA kernels and libraries.” This indicates that Nvidia's GPU processing has become the limiting factor in the performance of the system.
It's still not clear whether Cruise is using Drive PX 2, a direct variant, or just standalone Nvidia GPU accelerators, but the adoption of Nvidia GPU technology by GM/Cruise represents a big win for the company. This was facilitated by Nvidia's investment in CUDA software development and research in deep learning and autonomous vehicles.
As Cruise and other companies continue to develop their software, they will find it increasingly difficult to eliminate their dependence on Nvidia's underlying APIs. This will make the processing hardware more “sticky” than it otherwise would be.
And, at least at this juncture, there really is nothing better available to automakers than Nvidia's GPUs and Drive PX. Nvidia offers the “path of least resistance” to get to fully autonomous vehicles.
This Tech Brief is a summary of a longer analysis available exclusively to Rethink Technology subscribers. Nvidia is part of the Rethink Technology Portfolio and is a recommended buy.