Tesla has been equipping all its vehicles produced since October 2016 with a “supercomputer” as part of its Autopilot 2.0 hardware to run its deep learning neural net computer vision technology, which the company believes will eventually enable fully self-driving capability through future software updates.
While it is likely the most powerful computer in a production car today, some had doubts about it being powerful enough or have enough redundancy to truly enable level 5 autonomous driving. Tesla CEO Elon Musk even said that they could eventually upgrade the system if it can’t achieve full autonomy.
That was less than a year ago and we now learn that the automaker is already shipping a new Autopilot hardware suite with a more powerful version of its onboard computer.
We already knew that Tesla updated its Autopilot hardware suite for the Model 3 when we realized that the vehicle is equipped with a driver-facing camera – something not currently available in Model S or X.
But now several sources told Electrek that it is not the only change that coincided with the start of Model 3 production. We are told that a new onboard computer updated to include a secondary GPU for more computing power was also added. The new Autopilot hardware suite is internally called “HW 2.5”.
In a statement to Electrek, a spokesperson confirmed the existence of the new hardware suite, but they downplayed its importance:
“The internal name HW 2.5 is an overstatement, and instead it should be called something more like HW 2.1. This hardware set has some added computing and wiring redundancy, which very slightly improves reliability, but it does not have an additional Pascal GPU.”
The spokesperson also said that all Model S, Model X, and Model 3 vehicles being ordered today come with the new hardware.
Tesla’s current onboard computer in HW 2.0 Model S and Model X vehicles is based on Nvidia’s Drive PX2 platform for autonomous driving, but Nvidia offers several iterations of the same platform with different computing power in order to enable different levels of autonomous driving.
The companies working on self-driving technology disagree on the computing capacity needed to achieve autonomous driving, which results in as many computing solutions as autonomous vehicle programs. Tesla achieved level 2 with 0.256 trillion of operations per seconds (TOPS) in Autopilot 1.0, but level 3 to 5 are expected to need significantly greater capacity – anything from 2 to 20 TOPS, depending on who you ask.
It was never clear exactly what version of the platform Tesla has been using in its vehicles with Autopilot 2.0 hardware until Model S owner Kyle Day dismantled the entire thing to reveal the actual board.
The teardown showed that it is actually a custom Drive PX2 board with a 1 SoC and 1 discrete GPU.
Nvidia believes that 2 SoCs and 2 GPUs, like its ‘Autochauffeur’ Drive PX2 platform, would be required on its platform to achieve self-driving capability.
Our sources say that Tesla’s new Autopilot 2.5 suite is using a secondary node to enable more computing power, which brings the capacity closer to chipmaker’s own idea of the power needed to enable level 5 full autonomy – Tesla’s ultimate goal for the Autopilot program.
After announcing Autopilot 2.0, Musk said that he believes Tesla could achieve self-driving capability on the computer, but he left the door open for an upgrade and highlighted that they build the mount under the glove box to be easily accessible if they need to swap the system for a more powerful computer.
Tesla still believes that it can achieve the promised full autonomy on the 2.0 suite, but they now say that they could upgrade HW 2.0 cars with the new 2.5 hardware at no cost in the “highly unlikely” possibility that it is actually needed:
“However, we still expect to achieve full self-driving capability with safety more than twice as good as the average human driver without making any hardware changes to HW 2.0. If this does not turn out to be the case, which we think is highly unlikely, we will upgrade customers to the 2.5 computer at no cost.”
Furthermore, the company also reiterated their goal of “driving from a parking lot in downtown LA to a parking lot in downtown NY without touching the controls by the end of the year.”
The move is the latest example of Tesla consistently adding new technologies and improving on its current vehicle programs without going through the usual annual model change, which is most common in the auto industry.
It can complicate the timing to purchase a Tesla vehicle for someone looking to get the latest technology, but you would have to wait forever if you keep waiting for the latest technology. At least, this particular change can be retrofitted.
Now whether or not Tesla can achieve full autonomy on those vehicles remain to be seen, but if they are right, I don’t think the magnitude of the claim is well appreciated.
They are saying that they could virtually flip a switch and through an over-the-air software update, an entire fleet of currently over 70,000 cars would potentially become fully autonomous overnight, if local regulations allow it.
The impact on the overall transportation industry is hard to imagine.
The timeline for that to happen is still unclear at this point. As Tesla reiterated, they still plan to do a demonstration drive by the end of the year, but that doesn’t mean that the software build on which they plan to do the demo will be ready to be pushed to the fleet or that regulatory approval will be in place for consumer vehicles.
It’s still incredibly exciting! What do you think? Let us know in the comment section below.
Subscribe to Electrek on YouTube for exclusive videos and subscribe the podcast.