In the new era of mobility, some rivalries have nothing to do with 0-to-60 mph times. Boasts are about more obscure figures such as teraoperations per watt.
Chipmakers are becoming increasingly prominent players in the auto industry, as connectivity and autonomous driving require greater magnitudes of computing power. With just a few companies vying for what could become a lucrative role in the car of the future, competitors are beginning to call each other out as they claim to have the fastest and most efficient chips.
At AutomobilityLA last month, Intel CEO Brian Krzanich showed a slide claiming that Intel's Mobileye EyeQ5 chip was almost two and a half times more efficient than competitor Nvidia's Xavier platform.
Nvidia shot back the following week, calling the comparison "fake news."
"They're comparing what we have today to what they're thinking about doing in two years," said Danny Shapiro, head of automotive at Nvidia.
Intel's claims are based on its own calculations for the EyeQ5, a chip intended for Level 4 autonomous driving that will be available for sampling next year. It is slated to be in full-scale production by 2020 and power the BMW iNext coming in 2021. The company measures the efficiency of the chip's deep-learning operations — which allow the car to learn on its own — at 24 teraoperations per 10 watts of energy used.
Intel compared these numbers with those posted on Nvidia's site for Xavier — a platform for automated driving that will be available to automakers in early 2018 — that measure efficiency for all of the platform's functions at 30 teraoperations per 30 watts of energy used. Intel says that comparison shows its chip is 2.4 times more efficient than Nvidia's.
But that comparison is too surface-level, said Shapiro. A more accurate comparison, he said, would be to isolate Xavier's deep-learning processes and measure efficiency for 24 teraoperations on a process similar to the EyeQ5 chip's. Using those calculations, Nvidia came up with an efficiency rate of 24 teraoperations per 7 watts used, which makes it more efficient than Intel's chip.
Mike Ramsey, a technology analyst at Gartner Research, said Intel and Nvidia have been touting two different capabilities — performance per watt and speed of processing — that may not be directly comparable.
"I think it's fair to say the performance specs and the way they have been framed don't fully encompass capabilities," he said.
Intel stood by its claims, saying it was an "apples-to-apples" comparison.
"The point we're trying to make is the EyeQ5 ... chip is significantly more efficient than Xavier," said Dan Galves, senior vice president at Mobileye, an Intel company.
