See that thing there? Nvidia’S Drive PX2, it’s a beast, a true end-to-end supercomputer. And it’s coming to a car near you!
Nvidia have over the last few years been inventing new markets for themselves, typically entering with barnstorming hardware that blanks opportunity for competitors. You want supercomputer processing? Nvidia does that & in a far more accessible & configurable manner than Cray et al.
Drive PX2 is an AI supercomputer platform for self-driving cars, & it’s just been selected to power the fully autonomous Tesla electric vehicle range. “Fully autonomous” means just that – robotic cars that drop you at work then go out to join the taxi fleet, earn an income, then pick you up after work.
As the AI for Tesla, Drive PX2 will “see” through 8 cameras, 360 radar & 12 ultrasonic sensors to create vision that understands both hard (lamp-post) & soft (children) objects fully at 250 metres. The data is processed for local use – driving – & shared into the Tesla Neural Net, increasing the smarts of every Tesla vehicle on the fly. The new processing unit is rated at 40x more powerful than the already impressive on board computer in Tesla vehicles.
Water cooling dissipates heat from the 250w consumed, Drive PX2 has computation power to burn. Equivalent to 150 Macbook Pros, Drive PX2 has been provided to Nvidia development partners at $US15,000 each. I’m sure Elon Musk negotiated a bulk discount rate for this end-to-end AI platform.
Even with a discounted price, Drive PX2 will probably be the single most expensive component in Tesla vehicles. Considering it’s also the most important single component, I don’t have a problem with this.
Musk says that the decision to purchase Nvidia product was a close run thing, with AMD subsidiary ATI & Intel also putting in strong bids. But really, how could Tesla not choose the company who also uses Tesla as the moniker for their high-end graphics processors? Just had to be that way.
“NVIDIA DRIVE PX 2 is an end-to-end AI computing system that uses groundbreaking approaches in deep learning to perceive and understand the car’s surroundings.”
12 individual processing cores thumping through an incredible 11 TFLOPS/sec & 24,000,000,000,000 Deep Learning operations/sec mean that no, you shouldn’t fancy your chances in a head to head PerformanceMark benchtest vs your Haswell i7 Extreme gaming rig. Even if you do use RAID-0 SSD’s to boot from.
Comparison – in 2010, the Japanese government replaced their Earth Simulator supercomputer with Earth Simulator 2. (I know, the naming convention is a little predictable.) Designed to predict weather patterns by simulating Earth’s planetary development, ES2 is a massive project requiring enormous resources. In November 2010, ES2 topped the HPC Challenge awards – the Supercomputer bragging list – with a verified 11.876 TFLOPS. 6 years later, NVIDIA deliver Drive PX2 with matching TFLOP performance.
Tech titans are backing Elon Musk’s vision, throwing their resource into the most exciting disruptive movement happening today. Just 10 years ago ICE (Internal Combustion Engine) had no competitor, theoretical only. As for piloting options, Google’s self-driving car was perceived as a madcap experiment of the billionaire boy founders, nothing more. Manual gearboxes were still a preferred option because “it gives me more control” – yeah right.
Today, the entire industry has turned. Yesterday (25/10/16) Budweiser beer was delivered inter-city by a driverless big-rig, and March sales figures had Tesla’s new Model 3 fully electric car pre-sales orders outselling the entire Ford range by 150,000 vehicles. The world’s fastest accelerating production car is the luxury 4-door Tesla Model S P100D, offering 0-100km/h in 2.5sec, embarrassing all the traditional performance manufacturers – Ferrari, Porsche & Lamborghini included.
Remember the motoring conditions of today – they’re fast coming to an end. These are the Wild West days, the evolution is a revolution.