NVIDIA puts open platform and world’s 22nd most powerful supercomputer in a car
Around this time last year, Connected Car talked with NVIDIA, the company that pioneered a supercharged form of computing loved by the most demanding computer users in the world — scientists, designers, artists, and gamers – and installed it in a car.
Fuelled by the insatiable demand for better 3D graphics, on-board computing power and the transition of the automotive industry from a car manufacturing business to a technology/mobility/ride-sharing/transitional financing/who knows what hotbed of devlopment, NVIDIA has evolved the GPU into a computer brain at the exciting intersection of virtual reality, high performance computing, and artificial intelligence.
There has not been any doubt that NVIDIA is a company we need to talk to on a regular basis. When the company sent us a recent press release extolling the virtues of DRIVE AGX, a scalable, open autonomous vehicle computing platform for autonomous vehicles, this prompted us to pick up the phone and arrange to talk again. And so it was that Vince Holton chatted again with Danny Shapiro, Senior Director of Automotive at NVIDIA.
VH: It’s great to be talking again, Danny. We will look specifically at DRIVE AGX in a moment, but before we do that, bring us up to speed with what is going on in the industry and NVIDIA’s main developments since we talked one year ago.
DS: Well, today, all types of autonomous vehicles are using our technology from end-to-end – that includes in their data centres, for training on detailed networks and using our DRIVE Constellation simulator to test and validate all of the hardware and software that is in the vehicles. Then we have recently made an announcement with Volvo trucks concerning the use of our AI in their commercial vehicles. It’s a big, big deal.
VH: Isn’t that scary? Autonomy and AI in large trucks on the road?
DS: There’s no rush to get vehicles on the road. Volvo’s reputation is of course based on safety and, in that, there is shared DNA with NVIDIA. The focus will always be on safety, and this has been the case in the trials that have been going on. Many of these have been taking place in mines and on construction sites, but the goal is for the next step to be trials on public roads. The difference is that these vehicles are big and have huge payloads. There’s a lot of weight and the stopping distances are much greater than in passenger cars, so there has been a lot of work going on to make sure that we have the right robustness of systems. There could be new types of sensor, different to the ones in passenger cars, plus AI that has been specifically developed for trucks and commercial vehicles.
In another development, the International Supercomputing Conference recently took place in Frankfurt. Every year this event publishes a list of what it calls the Top 500. These are the fastest supercomputers in the world. These are at datacentres typically used by federal agencies and then there are supercomputers in Japan and China that are doing incredible work. A couple of years ago at NVIDIA we introduced something called the DGX which is a server that is full of graphics processors (GPs) for doing high-performance AI computing. We’ve recently unveiled DGX SuperPOD. This is basically a mini data centre kit or reference design. We use it in-house for our autonomous vehicle testing. It is 96 of our DGX’s that are configured in a unified way to be a massive supercomputer. That product will be available to our automotive customers and those in other industries. It takes about 3 weeks to build out, compared to the average datacentre that takes 6-9 months. We submitted results from SuperPOD to the Supercomputing Conference and it ranked number 22 in the list of the world’s top 500 supercomputers. This, then, is an incredibly high-performance, energy efficient system that car companies can use in their testing, training and development of autonomous vehicles.
VH: Well, with connected vehicles generating one terabyte of data per hour, per vehicle, the industry is going to need this sort of computing power, isn’t it?
DS: This is absolutely true. I think that the challenge is not only in the vehicle and that the whole industry – NVIDIA included – has underestimated the challenges created by the complexity, and so the timelines are slipping out a little bit ……..