Tesla’s (TSLA) D1 Chip Is a Brilliant Checkmate to the Company’s NHTSA Investigation Disaster

[TruckElectric]

Tesla’s (TSLA) D1 Chip Is a Brilliant Checkmate to the Company’s NHTSA Investigation Disaster

By Rohail Saleem

Image Source: Getty

Tesla (NASDAQ:TSLA 673.47 -2.25%), the world’s largest EV manufacturer, just pulled off a brilliant gambit to neutralize the creeping threat from the ongoing NHTSA investigation while also solidly establishing itself as a tech company, thereby justifying the stock’s hefty tech premium.

As we noted in a previous post, the past couple of weeks have not been kind to Tesla. Back in June, the company canceled its Model S Plaid+ version, which reportedly featured a range of around 520 miles. Then in early August, Tesla delayed the launch of its much-anticipated Cybertruck from the end of 2021 to 2022, while citing bottlenecks such as the steel exoskeleton body requiring a revamp of the production process.

The biggest threat to the company’s bullish narrative, however, came this week when the US National Highway Traffic Safety Administration (NHTSA) dropped a proverbial bombshell. To wit, NHTSA’s Office of Defects Investigation (ODI) has now initiated an evaluation of Tesla’s Autopilot system following a series of troubling accidents. Readers should note that the investigation may plod along for months, with this added uncertainty acting as a conduit for further stock price weakness.

So why does NHTSA’s investigation matter so much? Well, depending on the severity of the findings, Tesla may be forced to fundamentally alter its Autopilot system or disable it altogether. Bear in mind that Tesla’s semi-autonomous driving system, dubbed the Autopilot, has recently gone through a major overhaul, with the company abandoning the radar sensor in favor of eight cameras. While the change does bring cost benefits, some analysts argue that it would render the system more vulnerable to adverse weather conditions. Readers should note that the company continues to test LiDAR sensors for its Autopilot system, setting the stage for possible incorporation at a later stage.

This brings us to the crux of the matter. Tesla held its AI Day on the 19th of August. During the event, the EV manufacturer unveiled a humanoid robot prototype, known as the Tesla Bot, while also detailing the company’s AI-focused ambitions. Nonetheless, Tesla’s D1 chip stole the proverbial thunder. The chip, built completely in-house using a 7nm architecture, will power the company’s Dojo supercomputer – the engine behind Tesla’s vast neural network training ambitions. Crucially, according to Elon Musk, the chip would empower Dojo to process images around four times faster than other computing systems. Tesla aims to regularly refine its Dojo-trained AI software and then push those incremental improvements to its EVs using OTA updates.

Image Credit: Tesla

During the presentation, Tesla conceded that the Full Self-Driving (FSD) capability of the Autopilot was being hampered by two factors: temporary blockage of the cameras’ field of view and faulty memory of road signs. To counter this, Tesla engineers are now utilizing a spatial recurring network video module to train the AI to keep track of spatial as well as chronal queues. In theory, this should help the Autopilot system overcome its two shortcomings.

So why do we believe that the D1 chip is a brilliant checkmate? Well, Elon Musk has now essentially neutralized the growing concern around the Autopilot system’s current capabilities. By conjoining the FSD with the D1 chip, the company has neutered the argument that the Autopilot is a dud product. More importantly, this gambit buys crucial time for Tesla to navigate stormy waters ahead. Musk says that the chip will launch in 2022.

However, never has any product announced by Musk launched on time. So more likely, the D1 chip will debut in 2023. Until then, Tesla can hold its detractors at bay while also continuing to improve the underlying capabilities of the Autopilot system. Quintessentially, the D1 chip means time for Tesla. More time to refine its product. More time to delay the reckoning that Musk’s detractors claim is just around the corner. I guess time will tell.


SOURCE: wccftech

 

[TruckElectric]

 

[TruckElectric]

 

[firsttruck]

Tesla’s (TSLA) D1 Chip Is a Brilliant Checkmate to the Company’s NHTSA Investigation Disaster
By Rohail Saleem
.....
Tesla (NASDAQ:TSLA 673.47 -2.25%), the world’s largest EV manufacturer, just pulled off a brilliant gambit to neutralize the creeping threat from the ongoing NHTSA investigation while also solidly establishing itself as a tech company, thereby justifying the stock’s hefty tech premium.

D1 Chip is not Tesla's first chip. There is current production FSD HW3 and completed design FSD HW 4. Also world leading electronics in power controller that drive motors of Tesla Model 3, Model Y, 2022 Model S. Heat pump Octovalve. Only FSD software (still in beta) that is targeted to work on most any street with out needing previous hi-def LIDAR mapping (ie Waymo).

Tesla already had solid reputation as tech company. This article adds no new info but implies Tesla was not a tech company before.

Tesla’s (TSLA) D1 Chip Is a Brilliant Checkmate to the Company’s NHTSA Investigation Disaster
By Rohail Saleem
.....
Then in early August, Tesla delayed the launch of its much-anticipated Cybertruck from the end of 2021 to 2022, while citing bottlenecks such as the steel exoskeleton body requiring a revamp of the production process.

Elon did not say Cybertruck steel exoskeleton was one of the primary causes of delay. The exoskeleton was purposely designed to use a different manufacturing process. This was not a surprise it was by design. How is it a revamp when it will be the first design for making the world's first exoskeleton 1/2 ton crew cab pickup truck with 6.5ft bed?

A change from late 2021 to early 2022 is minor. In today's environment it is to be expected. Notice that the article writer says nothing about legacy OEMS are experiencing severe delays on models already designed and in large scale production due to semiconductor chip shortage. Toyota just announce production will be reduced by 40%. There is still a world wide COVID pandemic going on. So Tesla has to spend a lot of attention & resources to keep production of Model 3 & Model Y up while opening two new gigantic factories (Austin, Berlin). I expect Tesla will take much less production hit than the legacy OEMs.

Tesla’s (TSLA) D1 Chip Is a Brilliant Checkmate to the Company’s NHTSA Investigation Disaster
By Rohail Saleem
....
So why does NHTSA’s investigation matter so much? Well, depending on the severity of the findings, Tesla may be forced to fundamentally alter its Autopilot system or disable it altogether. Bear in mind that Tesla’s semi-autonomous driving system, dubbed the Autopilot, has recently gone through a major overhaul, with the company abandoning the radar sensor in favor of eight cameras. While the change does bring cost benefits, some analysts argue that it would render the system more vulnerable to adverse weather conditions. Readers should note that the company continues to test LiDAR sensors for its Autopilot system, setting the stage for possible incorporation at a later stage.

Radar was causing more problems in important areas than it was helping in some fringe weather situations. Tesla has already shown it does not need radar.

Tesla said it does not need LIDAR and proven it does not need LIDAR. There is no later.

Tesla’s (TSLA) D1 Chip Is a Brilliant Checkmate to the Company’s NHTSA Investigation Disaster
By Rohail Saleem
.....
However, never has any product announced by Musk launched on time. So more likely, the D1 chip will debut in 2023.

Bullshit. One of the major products in Tesla history, Model Y, was not delayed in fact shipped several months early.

Tesla’s (TSLA) D1 Chip Is a Brilliant Checkmate to the Company’s NHTSA Investigation Disaster
By Rohail Saleem
.....
During the presentation, Tesla conceded that the Full Self-Driving (FSD) capability of the Autopilot was being hampered by two factors: temporary blockage of the cameras’ field of view and faulty memory of road signs. To counter this, Tesla engineers are now utilizing a spatial recurring network video module to train the AI to keep track of spatial as well as chronal queues. In theory, this should help the Autopilot system overcome its two shortcomings.

So why do we believe that the D1 chip is a brilliant checkmate? Well, Elon Musk has now essentially neutralized the growing concern around the Autopilot system’s current capabilities. By conjoining the FSD with the D1 chip, the company has neutered the argument that the Autopilot is a dud product. More importantly, this gambit buys crucial time for Tesla to navigate stormy waters ahead. Musk says that the chip will launch in 2022.

The new D1 chip has very little to do with Tesla Autopilot.

Tesla Autopilot is a adaptive cruise control for use on highways (not city & suburban streets).
Adaptive cruise control is primarily for limited access divided highways that do not have stoplights/signs, NO intersections (no threat of being t-boned by cross traffic), almost no pedestrian or cyclists, etc.

Ford, GM, Toyota, Honda, Mercedes all have adaptive cruise control for use on highways.
None use LIDAR for adaptive cruise control for use on highways.
Tesla does not need D1 chip and DOJO for Tesla Autopilot.

Tesla needs D1 chip and DOJO to make FSD better. FSD is used city & suburban streets where there is non-divided streets that have stoplights/signs, intersections, pedestrians, cyclists, ball from kids accidentally kicked in street with maybe kid following, etc.

D1 chip and DOJO has no bearing on NHTSA and Tesla Autopilot.

 

[CyberGus]

The terms “Autopilot” and “FSD” are often (and incorrectly) used interchangeably, but they are obviously different in their capabilities.

 

[Ogre]

I've worked in the technology industry for a long time and one thing I learned a long long time ago is shipping is the most important thing. Lots of companies have demonstrated products that never made it to production or that came to production so late their competitors were already shipping competitive products.

Tesla has made a *giant pile* of promises about FSD and so far what is widely available is only marginally better than Autopilot. Autopilot is easily best in class, but FSD as shipping to normal consumers is more or less a toy and it's extremely hard to justify the price. They've blown through so many Musk style "End of year" "Next Month" "In 2 weeks" timelines it's become a bit of an industry joke.

Keep in mind that other companies (with shipping chips) keep their R&D secret have invested massively in this industry and by the time Tesla actually ships the D1, it may not be this huge advantage.

 

[tidmutt]

Radar was causing more problems in important areas than it was helping in some fringe weather situations. Tesla has already shown it does not need radar.

Tesla said it does not need LIDAR and proven it does not need LIDAR. There is no later.

Indeed.

I've seen this LIDAR claim a few times. People don't realize they were using some LIDAR systems to validate the LIDAR like capabilities they developed from the vision system. It was NOT a sign that Tesla is going to use LIDAR.

 

[carpedatum]

Judging by the posts above, I'm not the only one who felt grumpy when reading this article. Sensationalizing things in order to sell clicks is starting to get on my nerves. The author wrote the headline and then shaped the narrative in a sad attempt to support it. Happens all the time.

I am particularly puzzled, though, as to why language like "product" and "ship" gets applied to the D1. Anybody care to enlighten me? AFAICT, Dojo is a system element and the D1 is an element of Dojo. These bits are on the back end, at some Tesla-operated facility. The consumer can only buy access to the results of their existence, by buying FSD with their Tesla vehicle.

Nobody buys the D1 (except Tesla, of course, as they have to pay someone to fabricate it). Unless I'm missing something. Is there some public evidence that HW4 exploits the D1 chip internally, or something?

 

[Ogre]

Hardware 4 is going to be physically shipped.

FSD will be shipped when it's out of beta regardless of whether pieces of it are in Tesla's offices somewhere or in my car.

D1 will be "Shipped" when they've installed it in a production machine.

 

[tidmutt]

Hardware 4 is going to be physically shipped.

FSD will be shipped when it's out of beta regardless of whether pieces of it are in Tesla's offices somewhere or in my car.

D1 will be "Shipped" when they've installed it in a production machine.

Yep. D1 will be shipped internally.

 

[firsttruck]

Judging by the posts above, I'm not the only one who felt grumpy when reading this article. Sensationalizing things in order to sell clicks is starting to get on my nerves. The author wrote the headline and then shaped the narrative in a sad attempt to support it. Happens all the time.

I am particularly puzzled, though, as to why language like "product" and "ship" gets applied to the D1. Anybody care to enlighten me? AFAICT, Dojo is a system element and the D1 is an element of Dojo. These bits are on the back end, at some Tesla-operated facility. The consumer can only buy access to the results of their existence, by buying FSD with their Tesla vehicle.

Nobody buys the D1 (except Tesla, of course, as they have to pay someone to fabricate it). Unless I'm missing something. Is there some public evidence that HW4 exploits the D1 chip internally, or something?

D1 chip is the hardware. DOJO is the complete system (D1 hardware, operating system, training software applications).

D1 Chip & DOJO will be used internally by Tesla. Probably never in cars.

Tesla might sell access to DOJO just like Amazon sells AWS cloud computing.
Other companies might want to use Tesla DOJO for neural network training for their products.

 

[Crissa]

They had a D1 in hand, so what is 'shipped' when they already have some prototypes for something which is largely a one-off to begin with?

But... Not so many others have their own chips. When Intel has a new chip, they most certainly tout it like Tesla did here. I have been to enough developer presentations and recruitment presentations to know there's nothing ut of the ordinary for what Tesla did at AI day (aside from the technology). You have what they have done, what they're planning to do, what they're hiring to do, and then you have something aspirational and funny at the end. If you skip that part, people forget the meeting altogether. Even Intel had the shiny dancing bunny suits.

-Crissa

 

[Throwcomputer]

Judging by the posts above, I'm not the only one who felt grumpy when reading this article. Sensationalizing things in order to sell clicks is starting to get on my nerves. The author wrote the headline and then shaped the narrative in a sad attempt to support it. Happens all the time.

I am particularly puzzled, though, as to why language like "product" and "ship" gets applied to the D1. Anybody care to enlighten me? AFAICT, Dojo is a system element and the D1 is an element of Dojo. These bits are on the back end, at some Tesla-operated facility. The consumer can only buy access to the results of their existence, by buying FSD with their Tesla vehicle.

Nobody buys the D1 (except Tesla, of course, as they have to pay someone to fabricate it). Unless I'm missing something. Is there some public evidence that HW4 exploits the D1 chip internally, or something?

Someone in the qa actually asked this specific question regarding if the dojo chip would be used for the actual fsd hardware in the car. They either didn't understand the question to mean that, or they ignored it and indirectly answered by saying as much about the dojo chip being the neural network number cruncher separate from the in car hardware, and the product which the cars benefit from is the fsd software only.