Elon Musk's Chipmaking Claims Don't Match the Reality
Elon Musk asks, “How could it be that Tesla, who has never designed a chip before, would design the best chip in the world?”
(Bloomberg) -- Tesla Inc. co-founder Elon Musk has already shaken up the automotive industry. On Monday he took a swipe at semiconductors, arguing that his own engineers have created technology that puts chipmakers’ efforts in the shade.
Speaking at an investor day focused on autonomous driving, Musk rhetorically asked the question: “How could it be that Tesla, who has never designed a chip before, would design the best chip in the world?”
He quickly got an answer from former supplier Nvidia Corp., which said comparisons he gave to back up his claims were wrong. Musk also knocked other approaches to building systems that will make human drivers redundant and made claims about the way Tesla is bucking conventional wisdom to tackle one of the most complex engineering problems today. Here is a fact-check of the technology behind driverless cars.
The Numbers
Tesla’s computer system is capable of 144 TOPS (trillion operations per second) while Nvidia’s was ticking by at just 21 TOPs, Musk told his audience. Not true, according to Nvidia, whose chips were replaced by Tesla’s own in all of its models recently. First, Nvdia says Musk was comparing an entire autonomous system containing multiple chips to just one chip designed to provide only limited driver assistance. A fair comparison with an equivalent Nvidia computer would have Tesla’s 144 TOPS stacked up against 320 TOPS from a machine based on semiconductors from the graphics chip maker.
Technology
Musk also took a shot at Lidar (a term referring to “light detection and ranging,” or a portmanteau of laser and radar) calling it a “fool’s errand’’ and saying anyone using the technology is doomed. Tesla’s system will focus on image recognition. While others in the industry are also using image recognition-based systems, many are talking about “sensor fusion.” They argue that it’s better to use a combination of cameras, radar, ultrasonic and Lidar sensors to build a complete computer picture of the world around a vehicle that will enable it to make better decisions. Each type of sensor has strengths and weaknesses. Some only work over short distances, some perform poorly in snow, and some can be thrown off by bright sunlight or don’t work well after dark.
Software
Musk made the bold claim that with the new computer, Tesla vehicles are ready to pilot themselves, once the software is fully developed. Again, he took a shot at the process that others are using to develop the software to be as effective at driving as humans. Simulation -- using computer-generated images to train systems -- is an approach several companies, including Nvidia, are taking. Musk dismissed this, saying it doesn’t capture the “long tail of weird things that happen in the real world.” Musk’s opponents argue that software can be taught to react to dangerous situations far more quickly with simulation because most driving on the road is incident free. By bombarding a computer with tricky scenarios over and over again in a simulated environment, it learns faster without putting anyone at risk. Musk preferred a science-fiction analogy. “If the simulation fully captured the real world, well, I mean, that will be proof that we’re living in a simulation, I think. It doesn’t. I wish,” he said.
Economics
Musk claimed leadership in the race to create a computer that will make cars independent, saying Tesla’s chips were ahead. That’s bound to raise hackles at not just Nvidia but other large chipmakers working on the problem, including Intel Corp. and Qualcomm Inc. Tesla’s CEO brushed off the economics of chip-building by telling analysts that, like most chips, his are being made by someone else in an outsourcing arrangement, in this case, Samsung Electronics Co. But developing and designing powerful processors requires thousands of engineers and hundreds of millions of dollars. Companies like Intel and Nvidia are able to make that work because their products go into more than 200 million personal computers sold every year. Qualcomm sells chips for smartphones, a market of more than 1.4 billion units. Apple Inc., which designs its own chips, sells tens of millions of iPhones every year. Meanwhile, Tesla delivered 63,000 cars globally in the first quarter.
Reaction
Nvidia thanked Musk for shining a light on the difficulty of the engineering task ahead of the industry and said Tesla has set a “high bar” for other automakers who will need Nvidia’s help to keep up. Nvidia’s stock rose Monday while Tesla’s fell.
“After Tesla’s boasting about results of its three-year initiative, we saw nothing tangible that would suggest Tesla could have a lead over well-established competitors,” said Roth Capital Partners analyst Craig Irwin. “The company’s misrepresentation of Nvidia chips, which Tesla previously used, undermined Tesla’s message.”
“Tesla’s rejection of the technology (Lidar) as a ‘fool’s errand’ due to the currently high additional expense is likely penny-wise but pound-foolish, especially since Lidar prices continue to drop,” according to Cowen & Co. analyst Jeffrey Osborne.
“It will be tough for them to make it on their own -- simply put, Tesla is not a semi company so they may need to partner with someone to do it,” according to Lynx Equity Strategies analysts K.C. Rajkumar and Jahanara Nissar. “But then again, Musk has created a car by himself so a chip is certainly possible.”
--With assistance from Jeran Wittenstein and Dana Hull.
To contact the reporter on this story: Ian King in San Francisco at ianking@bloomberg.net
To contact the editors responsible for this story: Jillian Ward at jward56@bloomberg.net, Molly Schuetz
©2019 Bloomberg L.P.