Intel (INTC 4.50%) has been trying to upstage NVIDIA (NVDA 1.40%) in driverless cars. So, it isn't surprising to see Intel executives touting that their EyeQ5 autonomous driving chip is more efficient than NVIDIA's Xavier platform. CEO Brian Krzanich recently said that Intel's product "can deliver more than twice the deep-learning performance efficiency than the competition," which is a direct jab at NVIDIA, but does this claim pass muster?
A battle of specs
Jack Weast, principal engineer and chief architect of autonomous driving solutions at Intel, told EE Times that the EyeQ5 chip is 2.4 times more efficient than the current-generation Xavier chip. More specifically, EyeQ5 can deliver 24 trillion operations per second (TOPS) of computing power while consuming 10 watts of electricity. Intel claims that rival NVIDIA's DRIVE PX chip, based on the Xavier platform, carries 30 trillion operations per second (TOPS) of computing power but consumes 30 watts of electricity.
Image Source: Getty Images.
But this isn't an apples-to-apples comparison, according to NVIDIA. NVIDIA executive Danny Shapiro countered in another EE Times article that Chipzilla is comparing the power consumption of just one chip to NVIDIA's entire ecosystem that powers autonomous cars, which includes the CPU, the GPU, and the memory.
Moreover, Intel's comparison won't hold much water when considering that NVIDIA's next-generation autonomous car platform, the Pegasus, is capable of powering level 5 autonomy (the highest level of autonomy in self-driving cars where no human intervention is needed). NVIDIA's new system is 10 times more powerful than the current DRIVE PX2 platform, delivering 320 TOPS of computing power while consuming 500 watts of electricity.
A platform's computing power is very important in autonomous cars, and Pegasus delivers since it can support more advanced levels of autonomous driving despite higher power consumption.
Therefore, NVIDIA's upcoming artificial intelligence (AI) engine for self-driving cars can deliver a stronger performance when compared to Intel. More importantly, NVIDIA's chip is already being tested by at least 25 of its partners, and it will be commercially available in 2018. By comparison, Intel's EyeQ5 chip is expected to go into production only in 2020, so there's a good chance that NVIDIA's self-driving car platform will advance by another generation by that time.
Intel wants to steal the limelight
Intel has been a late entrant into the self-driving car race. The acquisition of Mobileye earlier this year gave the chipmaker a ticket to this ride, but NVIDIA has been involved in self-driving car development ever since it revealed its first-generation DRIVE PX platform almost three years ago.
Intel is trying to steal NVIDIA's thunder by trying to remain in the news, and it has done a good job so far. It has gradually built a formidable alliance of partners that includes carmakers BMW and Fiat Chrysler, as well as a marquee client in the form of Alphabet subsidiary Waymo. So its latest barb at NVIDIA isn't entirely unexpected.
More importantly, Intel is trying to close the technology lead that NVIDIA enjoys in the self-driving car space. Recently, the company revealed that it is working on a multichip platform to power autonomous driving, combining Mobileye's EyeQ5 computer vision chip with its own Atom processors and connectivity chips.
Intel plans to reveal this platform soon, which means that it will have an entire AI engine that will be comparable to NVIDIA in terms of computing power and energy consumption. More importantly, Intel's deal with entertainment company Warner Bros. indicates that the chipmaker has its eyes set on the future of self-driving cars.
The two companies are collaborating to develop "in-cabin, immersive experiences in autonomous vehicle (AV) settings." They will roll out a proof-of-concept car as a part of Intel's 100-car autonomous test fleet, aiming to take advantage of the additional time that an autonomous car user will have at his/her disposal.
Such an idea could help Intel differentiate itself from NVIDIA, but it is still far from turning into a reality. Moreover, it is still early days in the self-driving car race, so it is difficult to point out a winner at this stage.
NVIDIA currently has technology on its side and has built a massive ecosystem of 225 partners that Intel will go up against. But Chipzilla has started gaining traction, though it remains to be seen if it can actually walk the talk by landing more automotive clients.
