Intel vs. Nvidia: The self-driving car AI chip war is about to begin

Information source: eettaiwan

Please indicate the source and origin when reprinting

As Nvidia continues to promote its Drive PX platform and strives to position itself as a leader in AI-based autonomous driving technology, Intel is also gearing up to launch a multi-chip platform for autonomous driving soon, fully competing in this brewing technical specification war...

In a recent keynote address at the 2017 Los Angeles Auto Show and AutoMobility LA, Intel CEO Brian Krzanich detailed his predictions on how autonomous technology will change every aspect of future vehicles, from cabin design to entertainment and life-saving safety systems.

As a major self-driving car chip company, Intel also took the opportunity to tout the efficiency record of the EyeQ 5 chip developed by Mobileye (now an Intel company) and compare it with Nvidia's Drive PX Xavier SoC designed specifically for autonomous driving.

Krzanich mentioned the recently completed acquisition of Mobileye in his speech, and emphasized that Intel "can provide deep learning efficiency that is two times higher than that of competitors (in this case, Nvidia). "

As Nvidia continues to position itself as a leader in artificial intelligence (AI)-based autonomous driving technology by promoting its Drive PX platform, Intel seems to have put its full efforts into this brewing technical specifications war.

Misinterpreted?

Jack Weast, Intel's chief engineer and chief architect of autonomous driving solutions, said in an interview with EE Times that Intel has always been conservative in promoting the performance of its own chips. However, Weast said that as the war of words intensified, "We were tired of seeing us being quoted and misinterpreted all the time."

Weast complained that competitors and the media often mistakenly compare Nvidia's Drive PX to Intel's desktop PC chips. He said that if there is a true 'apples-to-apples' comparison, Mobileye's fifth-generation vision sensor fusion chip must be compared to Nvidia's Xavier SoC.

EyeQ5 can achieve 24 trillion operations per second (TOPS) of performance and consumes 10 watts of power. In contrast, DRIVE PX Xavier can provide 30TOPS of performance but consumes 30W of power. "Our chip is 2.4 times more energy efficient," said Weast.

Of course, Nvidia is now promoting its latest Pegasus SoC, an AI computing chip scheduled to be launched in 2018, designed to perform 320TOPS of performance (10 times the performance of its predecessor) and consume 500W of power, which is said to be used for L5 fully autonomous driving applications. "Although Pegasus is a new product, its energy efficiency performance has not been significantly improved," said Weast.

Nvidia's Pegasus is powered by the company's two latest Xavier SoCs, which feature a new generation of discrete GPUs with hardware acceleration.

But one unsolved mystery is how Intel plans to combine Mobileye's "eyes" (EyeQ vision chips) with Intel's microprocessor "brains" in a highly automated vehicle, something that even some experts may not understand.

In fact, Intel itself may have to bear some responsibility for this misunderstanding and confusion in the market. However, apart from the Mobileye EyeQ5, the CPU giant has remained tight-lipped about which autonomous driving SoC it is developing.

Autonomous driving multi-chip platform is coming soon

According to Weast, Intel plans to launch a "multi-chip platform for autonomous driving" at the International Consumer Electronics Show (CES) in early 2018. He explained that the solution will combine the EyeQ 5 SoC, Intel's low-power Atom SoC, and hardware including I/O and Ethernet connectivity.

When Intel launched its GO autonomous driving development platform earlier this year, it described the Atom processor C3000 as a chip that "provides a high performance-to-power ratio and can handle large amounts of computing in a low-power design."

So how exactly does the Atom SoC share processing tasks with the EyeQ 5? Weast said: "We first look at the overall workload required by the autonomous vehicle, and then distribute and divide the corresponding computing load among multiple chips."

As for whether FPGAs are also used in this multi-chip solution? Weast's answer is no. Weast said: "Some customers may seek FPGAs to meet certain specific applications, such as custom I/O or security, but this will not appear in the new multi-chip platform we are about to release."

How to divide the work?

Mobileye released EyeQ 5 before it was acquired by Intel. The Israeli company touted the new SoC as the "brain" of autonomous vehicles, capable of performing the tasks of "sensor fusion visual central computer" for fully autonomous vehicles (Level 5).

If it is really as Mobileye plans, then where does Intel's Atom SoC fit into this multi-chip solution?

Weast explained that autonomous driving requires different levels of sensor fusion. Some of the sensor fusion in deep learning acceleration applications requires the chip to be able to handle a lot of highly parallel multi-threaded program code for it. "So, EyeQ 5 is ideal." At the same time, higher-level environmental sensor fusion is required to focus on motion trajectory and data verification. "At this time, the CPU is better suited to handle such tasks," Weast explained.

In Intel’s view, achieving highly automated driving does not “require cramming everything into a single SoC (such as the EyeQ 5) to handle it all.” Weast explained that Intel has “the opportunity and leeway to fully understand where the idle cycles of system operation are and how the computing workload should be divided.”

Weast said that once the Mobileye acquisition was completed last August, every member of the team was "immediately involved in the development of the multi-chip platform program."

Weast promised that Intel will soon release details of its multi-chip platform designed for autonomous driving, and he said, "At that time, we should be able to do a platform-level comparison between Nvidia Drive PX and Intel's solution."

Related Reading:

Today's Core News: Qualcomm Snapdragon 845 is released, Lei Jun is as happy as a child...

What kind of business is wafer foundry?

A summary of smart home application solutions, a one-time scientific advancement!

Advertisement | MLCC hotline: 18926059102