CES 2018: Mobileye claims autonomous driving solution more efficient than Nvidia

Mobileye has said its acquisition by Intel means it is able to bring both hardware and software together as a differentiation to rival Nvidia's hardware-focused autonomous driving solution.
Written by Corinne Reichert, Contributor

Mobileye's acquisition by Intel provided it with the ability to produce a more complete autonomous driving solution than Nvidia, according to Mobileye senior VP and chief communications officer Dan Galves, because it has been able to bring both hardware and software together.

Speaking to media during CES 2018, Galves said software is the most important piece of the puzzle. Once that it written, it then needs to run on a powerful compute architecture -- and Mobileye now has both thanks to its work with the biggest chip maker in the world.

As soon as the $15.3 billion acquisition closed in August, Galves said the company brought a number of Intel engineers onto its team.

By comparison, Nvidia is largely "a hardware play", he said, and while it is a strong competitor in terms of having "very sophisticated processors", it still requires someone else to write the software. Producing the hardware without knowing what software will go on top of it therefore means it is less efficient, he claimed.

"If you only have hardware ... it's more difficult to create an efficient piece of hardware because you don't know what software is going to be on it," Galves argued.

"We bring the hardware and software together."

Mobileye is also differentiating its offering through the crowd-sourced high-resolution maps being created this year in partnership with BMW, VW, and Nissan from vehicles already on the road at "essentially zero cost" to the company.

By using cameras -- an inexpensive, mature sensor technology -- as its primary source of information, Intel is also ensuring that it keeps the pricing of its system down, as does its use of just two Mobileye EyeQ5 system on a chip (SoC) and one Intel Atom chip for the entire autonomous driving platform, resulting in low electricity consumption.

Comparatively, Galves said Nvidia announced that its L5 system will require two Pegasus chips at 400 watts each, which will involve high costs across both electricity and the chips themselves.

"We're in a good position to compete with them," he concluded.

Nvidia announced its L5 AI autonomous driving stack, named Xavier, during CES 2018 this week, calling it the "world's most powerful SoC".

Xavier uses data from a vehicles radar, lidar, cameras, and ultrasonic systems with what it said is more energy efficiency than any other solutions on the market, with more than 9 billion transistors and a custom 8-core CPU, 512-core Volta GPU, 8K HDR video processor, deep-learning accelerator, computer-vision accelerators, and the capability to perform 30 trillion operations each second on 30 watts of power.

See also: CES 2018 special coverage (CNET)

It now forms the foundation of two software platforms, Drive IX and Drive AR, which are part of Nvidia's Pegasus AI computing platform.

"We developed the entire stack of autonomous vehicle software," Nvidia chief executive Jensen Huang said during his CES 2018 keynote.

"Today we have over 300 customers developing on Nvidia Drive."

Mobileye this week unveiled its product roadmap out to 2020, announcing that it would be launching 15 autonomous driving projects during 2018 across 14 car manufacturers.

Mobileye said its "master plan" for 2018 involves working on its L4 development partnerships; extending its L2+ program to increase the level of autonomy of in-production vehicles; working on road experience management (REM) as a data strategy; and forming safety standards with industry and regulatory bodies.

New features being launched during 2018 by Mobileye for its autonomous vehicle technology include REM, 3D vehicle detection (3DVD), traffic light detection and recognition, and advanced road features including free space and path prediction.

The building blocks of REM involve harvesting data, aggregating it, and localising it, with Mobileye cooperating with NavInfo and SAIC to bring REM into China, with ongoing harvesting deals with global OEMs set for 2019 and beyond.

Mobileye will this month launch its new aftermarket device Mobileye 8 Connect, which Shashua said is additionally REM supported.

The company's 28nm EyeQ4H SoC, which will feature 6x VMP, 2x PMC, and 4x CPU, is going into series production from March 2018, and launching with four OEMs later in 2018 and 12 OEMs in 2019.

Shashua said Mobileye is currently utilising Intel's expertise and resources while working on the EyeQ5 SoC by opening it up to third parties as an open compute platform with SDKs and libraries for the first time ever. Intel is further helping create a fleet of 100 vehicles for testing, data collection, validation, and customer support; and providing 250Pb in datacentres to support fleet, validation, and customer support.

The first silicon of the 7nm EyeQ5H SoC is set to be produced in August 2018, with series production from March 2020 and four OEMs launching with it from 2020 onwards.

It is also partnering with Audi, BMW, Fiat-Chrysler, Honda, NIO, Nissan, and SAIC on L3 production from 2019; with BMW, Fiat-Chrysler, SAIC, NIO, and three OEMs on ongoing L4 sourcing decisions from 2020; and with Delphi Automotive on an L4 turnkey solution and central sensing localisation and planning (CSLP) platform.

Disclosure: Corinne Reichert travelled to CES 2018 in Las Vegas as a guest of Intel

CES Coverage

Editorial standards