Intel is pissed. That’s the only way to summarize an extraordinary blog post published by Professor Amnon Shashua, senior VP of Intel and CEO of Mobileye.
Corporations, as a rule, don’t get too nasty with each other in public messaging. While it’s not uncommon to see dueling companies blast each other in legal filings, most blog posts and PR statements will emphasize putting a positive spin on one’s own products rather than attacking a rival company for its. In many cases, companies don’t even like to mention rivals by name. This is how we end up with Intel and AMD both referring to each other as “the competition,” as if VIA somehow grabbed ten percent of the x86 market when no one was looking.
Today, Intel threw that rule out the window. Professor Shashua’s blog post argues at great length that Nvidia is nothing but a slavish imitator of Intel’s technology. That’s not a paraphrase. Shashua writes:
It is said that imitation is the sincerest form of flattery, and our innovations have not gone unnoticed with many embracing the same concepts that we pioneered. One industry player in particular habitually follows our lead and today I would like to set the record straight on its latest imitation.
What follows is a discussion of Mobileye’s technological innovations and what Shashua believes are Nvidia’s imitations of those same breakthroughs. In 2017, Mobileye published a proposal for a standard it calls RSS, or Responsibility-Sensitive Safety. One noteworthy thing about the paper is that it anticipates some of the safety concerns we’ve seen raised about autonomous driving in the past 12 months. When it was published, in 2017, companies like Waymo and Tesla were still marketing self-driving as a safer replacement for human decision-making based on references to the number of miles traveled without an accident.
The RSS paper points out in its opening paragraphs that this purely data-driven approach to safety management is “naive at best.” The entire point of RSS is to move beyond such a simplistic model and create something better. The paper proposes an extensive formal mathematical model for building safe self-driving vehicles. Intel says it “openly shared all the technical details and mathematics behind RSS because we believe that the safety of automated vehicles should not be proprietary, and that the industry should collaborate with governments on what it means for an AV to drive safely.”
According to Shashua, Nvidia reached out to Intel in 2018 about collaborating on its development of RSS, only to later back out of the project without explanation. Shashua states:
Imagine our surprise last week when Jensen Huang, CEO of Nvidia, announced a “first-of-its-kind” safety model for AVs. Curious to see what “first-of-its-kind” innovation Nvidia had created, we eagerly read the publicly released white paper about Safety Force Field (SFF), only to have the eerie feeling that we were looking in the mirror.
Intel then breaks down exactly how Nvidia’s SFF mirrors Intel’s RSS. There’s an eight-page PDF comparing the two, one section of which is shown below:
You can read Nvidia’s own documentation for its SFF in an introductory article or consult the full whitepaper. Intel’s characterization of the document as a close copy of its RSS concept with fewer details provided appears to be accurate to our eyes.
Of Copying and Credit
Given that Shashua acknowledges that Intel released its RSS concept openly, there’s not much the company can do to prevent Nvidia from developing its own take on the concept. But that’s also not really the problem here. The issue isn’t that Nvidia chose to build on the RSS concept, but that Intel feels Nvidia took its work, refused to be part of the coalition it wants to build, and then claimed to have developed a first-of-its-kind initiative that apes something Intel released publicly two years ago.
Calling the resulting product ‘Open and Transparent’ may have been poorly received by Intel, under the circumstances.
Giving appropriate credit to others by properly attributing work is a critical principle in both the closed-source and open source world. Even when the final product or standard is given away for free, being able to identify and claim work as one’s own or being part of the body that develops a critical safety standard both have value.
Over the long term, we would expect the self-driving industry to converge on a common safety framework. The existence of this type of optimized solution is a key factor in whether workloads can be accelerated at all. But this isn’t the first time we’ve seen Nvidia’s CEO make eyebrow-raising claims — earlier this year, Jen-Hsun told journalists there was no proof that AMD’s FreeSync solution worked, despite having literally announced days earlier that Nvidia would be enabling FreeSync support for its GPUs.
“We never competed,” Huang told PCWorld concerning FreeSync support. “(FreeSync) was never proven to work. As you know, we invented the area of adaptive sync. The truth is most of the FreeSync monitors do not work. They do not even work with AMD’s graphics cards.”
Nvidia did build the first commercially available adaptive sync solution, known as G-Sync, but neither Jen-Hsun nor Nvidia have ever released a shred of evidence to back up the claim that FreeSync was never proven to work or that “most” FreeSync displays fail to operate even when equipped with an AMD GPU. We await, with great interest, the explanation for how its new SFF standard is superior to what Intel released two years ago.