NVIDIA DRIVE

The NVIDIA DRIVE® platform consists of both the AI infrastructure and in-vehicle hardware and software to deliver everything needed to develop autonomy at scale.

https://www.nvidia.com/en-us/data-center/dgx-platform/

NVDIA self driving cars

Self driving cars require a fuck ton (that’s a scientifically accurate measurement) of image processing, deep learning, and other things that use lots of math.

With your standard 4 core CPU, it can take several minutes just to process a single image to identify pedestrians and bikers and other objects on the road. Not quite as real time as we’d like. Similarly, if we wanted to train the deep learning model on a CPU, it’d take years, and we’d all be long dead by the time they were on the road.

So you can see, there’s quite a problem here. CPUs simply can’t do the math fast enough for self driving cars to be viable. This is where NVIDIA comes in. GPUs are essentially CPUs with most of the pipelining and optimization stripped out, so they’re not very good at handling logic. What they are really good at, though, is math. Take out the features of a core that you don’t need, and each core becomes considerably smaller. And so then you can fit 1, 2, 3853 more cores, meaning that if you have lots of calculations that don’t depend on each other, you can do them all at once. This is called parallelization. And parallelization is exactly what self driving car math requires. It reduces the time to process an image from minutes to a fraction of a second. It trains deep learning in hours, not years. And that’s why NVIDIA is so important in regards to self driving cars. NVIDIA’s hardware is what makes self driving cars possible in the first place.