HW3

[Troy]

Here is how long it takes on average from the Fremont factory to delivery:

Each quarter, production starts with furthest away destinations and finishes with California. Tesla prefers to deliver Q1 production in Q1 and Q2 production in Q2 because it makes the financials look good in shareholder letters. Therefore, they will now start making HW3 cars for Australia, UK, Japan so they can be delivered within Q2. Then they will continue with HW3 cars for China, Europe, and Canada. Normally HW3 production for the US should start around 10-15 May 2019. Even if we don't see a single HW3 delivery before 20 May 2019, that would be perfectly normal and it would not change the fact that all production switched to HW3 by the end of Q1. However, there is a chance a small number of HW3 cars could be delivered in the US in April.

 

[EVDRVN]

I, respectfully disagree with that. I was at Monster.com in 2010. They had a meeting on using social media to build audience and it was going to be the next thing in customer engagement. My question, "what is an 'RT'"? There used to be a company Constant Contact that allowed companies to stay engaged with customers. Yeah, it's really a thing. How do you learn about the next gen Honda mower or Maytag washer? Remember this too in your thinking, Tesla has definitively shunned typical media advertising, Radio/TV/Billboards. They came out of the gate on Social Media. And I am a new-comer to social media.

I think you are missing my point so I will clarify. If you have something to sell an existing customer you contact them via the most direct method possible. You don't hope that they see something in social media. Tesla never directly contacted customers about the $2K upgrade, they left incredible numbers of customers uninformed and upset, This is the first and primary step anyone would take and would not exclude other less efficient methods that have little intrinsic value to an existing customer. This is not about advertising, marketing or anything else. I can tell you that as a former exec for a fortune 50 company, no person in our marketing/sales organization would have a job if they made such a rudimentary mistake. Even as an intentional matter it is a terrible strategy at multiple levels. Little shocks me with Tesla anymore.

Lastly anyone with a correspondence degree in marketing lists the key benefits and deliverables of a product to promote the sale and value of the offering. One example is listing say HW3 as included with the $2K upgrade. What many miss is that there is a very specific reason that was not listed and if it was an oversight then Tesla needs to fire a bunch of people. I'm going with intentional omission.

 

[Knightshade]

I think you are missing my point so I will clarify. If you have something to sell an existing customer you contact them via the most direct method possible. You don't hope that they see something in social media. Tesla never directly contacted customers about the $2K upgrade, they left incredible numbers of customers uninformed and upset, This is the first and primary step anyone would take and would not exclude other less efficient methods that have little intrinsic value to an existing customer. This is not about advertising, marketing or anything else. I can tell you that as a former exec for a fortune 50 company, no person in our marketing/sales organization would have a job if they made such a rudimentary mistake. Even as an intentional matter it is a terrible strategy at multiple levels. Little shocks me with Tesla anymore.

Lastly anyone with a correspondence degree in marketing lists the key benefits and deliverables of a product to promote the sale and value of the offering. One example is listing say HW3 as included with the $2K upgrade. What many miss is that there is a very specific reason that was not listed and if it was an oversight then Tesla needs to fire a bunch of people. I'm going with intentional omission.

I think your primary mistake is assuming Tesla has a bunch of marketing people to fire.

Also you might want to apply Hanlon's Razor to the HW3 thing.

 

[wcorey]

as a former exec for a fortune 50 company

I don't need to be a former Fortune 50 exec to know they are nowhere near having to worry about lack of demand. Their problem is still production meeting demand. I haven't seen numbers of Model Y orders but I do know of people who wanted one before they even knew what they looked like. So, yeah, it doesn't appear demand is an issue. And that is reflected in their sensitivity to customer perception.

 

[CCIE]

That is completely opposite of what Musk is saying. By end of 2019 FSD will be feature complete..OK, you say, that doesn't mean runnable. But Musk goes on to say it will work but has the same caveats as NoA and AutoSteer, the driver is still pilot in command and must prove (s)he's paying attention by providing some degree of torque on the steering wheel. He goes on to say by end of 2020 it should be completely autonomous (capable of having driver in back seat reading a book) notwithstanding regulatory approval. That all implies at least Model 3/Model Y hardware.
If, on the other hand, you meant older S and X models will never see FSD...I have to agree with that. This forum now has a different constituency than it had 18 months ago. But you did say snowball's chance in of of Model 3... That is incorrect but both sides can shout at the top of their lungs but we won't know until Dec 31st 2019. You comment implies Elon Musk is FOS.

I agree that we won't know for a couple more years. But, if you believe that the Model 3 will be able to drive you around while you sit in the back seat in 2021, then I've got a bridge to sell you! They can't even get autosteer and TACC to operate reliably, nevermind autonomous driving.

 

[wcorey]

I agree that we won't know for a couple more years.

We'll know in 9 months if Musk is FOS on FSD. In the ARK interview he was unequivocal in his end of 2019 statement. A tad equivocal for end of 2020 being able to read a book. To my knowledge Tesla has no plans for a no steering wheel, no accelerator/brake version. That implies they'll always be a driver.
As for sitting in the back seat.. That's an easy layup. I challenge any adult to get from the front seat to the back seat, much less sit comfortably in the back seat, independent of how he or she gets there.

Why am I confident? Because I use Autosteer the vast majority of the time on secondary roads and NoA almost exclusively on highways. I am confident in where the edge cases are, for the use cases I have. I know how to hint to the car what I want it to do, for those cases I've identified as edge.

 

[Troy]

Independent of the redundancy of the CPUs they would need heaters in the camera wells as well as radar well. I had a 2012 Prius Plugin Advanced that featured Adaptive cruise control (TACC). Whenever there was a sticky snow the radar would go offline.

That's why Tesla moved the radar. It used to be visible when they had the nosecone Model S. They moved it to behind the bumper. As for heaters in cameras, that already exists. See the video here. In addition, the Model 3 has a second power steering motor. Also, at some point, they added secondary wiring to the front radar. Also, there was something related to backup power. Generally speaking, the trend has been towards adding more and more redundancy to support driverless-FSD.

I have heard Elon use the term "driverless" which makes me think there will be two different versions of FSD: driver-present-FSD and driverless-FSD. I wouldn't be surprised if only native-HW3 supports driverless-FSD but retrofit-HW3 supports only driver-present-FSD.

I don't think I've seen Tesla refer to AP3, just the NVIDIA chip to Tesla plug compatible chip (HW3)

The following are the same thing:

• HW1.0 = AP1.0
• HW2.0 = AP2.0
• HW2.5 = AP2.5
• HW3.0 = AP3.0

Everybody uses these terms interchangeably including Tesla. See the example here from Tesla's website here.

 

[EVDRVN]

I don't need to be a former Fortune 50 exec to know they are nowhere near having to worry about lack of demand. Their problem is still production meeting demand. I haven't seen numbers of Model Y orders but I do know of people who wanted one before they even knew what they looked like. So, yeah, it doesn't appear demand is an issue. And that is reflected in their sensitivity to customer perception.

Where in my post was I talking about demand of any Tesla product or vehicle? Your argument is short-sighted and assumes that present demand is an indicator of competency of a company or long tern success. This is a key reason most companies fail and it does not address the issue mentioned in any way. You don't need to be a exec of a company to see how Tesla fails at customer contact and retention, this is already being seen with many long-term owners disillusioned with the brand. Speak to some front line Tesla people that are connected to the business side and you can learn quite a bit about how the majority of customers feel. However if present demand is your only metric for business success then that sounds like a solid business plan, just add a few tweets to round it out.

 

[Knightshade]

I have heard Elon use the term "driverless" which makes me think there will be two different versions of FSD: driver-present-FSD and driverless-FSD. I wouldn't be surprised if only native-HW3 supports driverless-FSD but retrofit-HW3 supports only driver-present-FSD.

That's going to be very awkward for him when all those who materially relied on his public statements that there was no reason to wait for HW3, because the only difference was the HW3 computer that 2.5 FSD buyers would get swapped in free, file a class action.

 

[EVDRVN]

...I have heard Elon use the term "driverless" which makes me think there will be two different versions of FSD: driver-present-FSD and driverless-FSD. I wouldn't be surprised if only native-HW3 supports driverless-FSD but retrofit-HW3 supports only driver-present-FSD.

.

Hint: Someone may be on to something that just may be related to the non-commitments of HW3 and FSD not to mention the fact that much terminology may change and be recategorized. Anyone consider that many things are being changed repeatedly and on the fly, not that Tesla works that way...

Ultimately I it will end up likely being very costly for Tesla to make this whole with all the loaded costs.

 

[wcorey]

which makes me think there will be two different versions of FSD

That would be incredibly stupid...from a software design perspective. The more versions of software you have the more bugs you have and inject. A steering wheel is unique as it is an observer as well as producer. In other words, there is a controller for it. In the case of accelerator and brake pedals, they are input only. If the car does a programmed brake action the pedal doesn't depress. So the difference between a true driverless car and a std car is the presence or absence of the physical wheel and pedals. The software is the same. Upon startup the steering wheel registers for steering actions and the non-existing wheel and pedals obviously don't register as action producers.
That may be more software centric than you were referring to. There are a lot of players in the true driverless arena as evidenced by CES. I envision something somewhat different If Tesla has an entry in that arena, it'll likely be something like an X with no driver seat, n number of captain chairs or bench seats, large central table, built in wet bar. <- maybe not, as difficult to control. But it'll drive a given geographic route with minor variations picking up passengers at point x, delivering them at point Y. Running a traveling salesman (Transportation) Lp.

Thanks for the references Troy!

 

[Troy]

They already have different versions of hardware for FSD.

Example 1:
HW2.0 has different cameras than HW2.5. That's why Sentry Mode, Dog Mode, Dashcam and Blind Spot Chime don't work with HW2.0. Do you think they will replace the cameras of HW2.0 too when retrofitting the HW3.0 computer? I'm not sure about that.

Example 2:
Model 3 has redundancy for power steering. See the article here. Tesla started selling FSD in October 2016. Obviously, the AP2.0 cars they were selling at the time don't have this feature. In addition, because only the Model 3 has an internal camera, the expectation was that only Model 3 would support the Tesla Network, (Tesla's version of Uber but driverless).

Tesla promised FSD to AP2.0 buyers but they said it was subject to regulatory approval. Regulators are likely to require redundancy for FSD. Therefore AP2.0 is unlikely to support driverless-FSD. I'm not sure about AP2.5 or retrofit AP3.0 either.

In Germany, the Federal Motor Transport Authority [Kraftfahrt-Bundesamt (KBA)] was testing the capability of Tesla's Autopilot hardware. I don't think they will approve driverless-FSD without a rear radar which is available in Mercedes' version. Without a rear-radar, I don't see how Teslas can perform Auto Lane Change without confirmation on the Autobahn. It's possible that native HW3 could have the wiring in place for a rear radar. This way when they release HW3.5 with a rear radar, it can be retrofitted to native-HW3.0 produced after 29 March 2019.

 

[voip-ninja]

Hint: Someone may be on to something that just may be related to the non-commitments of HW3 and FSD not to mention the fact that much terminology may change and be recategorized. Anyone consider that many things are being changed repeatedly and on the fly, not that Tesla works that way...

Ultimately I it will end up likely being very costly for Tesla to make this whole with all the loaded costs.

I imagine that Tesla really, REALLY wants to realize FSD revenue that they have taken on AP 2.5 systems over the last 1.5+ years. All they have to do to realize that is deliver a new feature and call it FSD and book all that revenue.

One tactic is to book the revenue, install the HW 3.0 on new builds only for the first quarter or two and let the hardware cost of the new card drop a bit before they start manufacturing the ones that will be retrofit into existing vehicles that bought the new feature.

Whether that feature requires HW 3.0 to be installed at a later date to be feature complete is something they can figure out over the next six months.

I'm with @CCIE I just don't see the current cars, even with a spiffy AP GPU upgrade being capable of being fully autonomous in most situations but I guess we'll see.

Anyone want to hazard a guess on how much the AP 3.0 guts cost Tesla? My guess is the board costs <$1,000.

 

[mobilesuit818]

I love how alot of people are getting over excited. (once again)

I am going to throw another variable that hasnt even been mentioned here..

Remember Elon giving away FSD and EAP for free to employees? Which required AP 3.0.

My money is on that the employees will get these first...i could be wrong.

 

[CCIE]

We'll know in 9 months if Musk is FOS on FSD. In the ARK interview he was unequivocal in his end of 2019 statement. A tad equivocal for end of 2020 being able to read a book. To my knowledge Tesla has no plans for a no steering wheel, no accelerator/brake version. That implies they'll always be a driver.
As for sitting in the back seat.. That's an easy layup. I challenge any adult to get from the front seat to the back seat, much less sit comfortably in the back seat, independent of how he or she gets there.

Why am I confident? Because I use Autosteer the vast majority of the time on secondary roads and NoA almost exclusively on highways. I am confident in where the edge cases are, for the use cases I have. I know how to hint to the car what I want it to do, for those cases I've identified as edge.

I just noticed you're from CT. Can you get the car to drive on I-95 from New Haven to NJ, over the GWB, with autosteer? Given the awful lane markings and ridiculous interchanges, I don't blame it for failing. But, my experiences with Autosteer on I-95 in the northeast are why I don't believe FSD is coming anytime soon.

 

[S4WRXTTCS]

My money is on that the employees will get these first...i could be wrong.

I'm of the opinion that some employees already have HW3.

My reason for thinking this is at least 1 or two popped up on TeslaFI at least a week or two before vehicles were even shipping with HW3.

They were quickly purged from TeslaFI, but someone managed to capture a screenshot before the purge.

 

[wcorey]

Can you get the car to drive on I-95 from New Haven to NJ, over the GWB, with autosteer? Given the awful lane markings and ridiculous interchanges, I don't blame it for failing. But, my experiences with Autosteer on I-95 in the northeast are why I don't believe FSD is coming anytime soon.

I haven't tried. I am, quite literally, at the other corner of the state. I do, however, use it all the time on I-395 and I-270 and on I-95 in PVD. I picked up the car in Mt Kisco. I am guessing you did as well. I was on I-684 when calibration completed. I immediately pulled over to restart the car for EAP/NoA and from that point on it drove to Hartford thru to where I-84 exited to, I think it was 72 to 101 (PVD-HFD). Ya know a good person to talk to? Ian Pavelko of M3OC fame, he drove his M3 from Quebec to Ft Lauderdale. I've been trying to cajole him into a treatise on the good, bad, and ugly of driving that. So, while not in the executive suite of a fortune 50 company, I am a pilot and have flown IFR transitioning from waypoint to waypoint on autopilot. I do understand what you're eluding to and I get it. I am not in the worlds worst driving environment which Greenwich to Fort Lee may well be. My point here is navigating a 'fan', driving or flying into thru the handle and exiting off one of the spokes is not an unsolved software problem. I suspect the answer to your scenario is removing the requirement for lane markers. Through the way highways, esp interstates, are geocoded it's simply which gate do you want to traverse, much like going through a toll both. Remember to, in order to get to level 3 you need to be able to do it when lane markings are obscured, you need to know where in space you are and where in space the lane is. Going back to my flying days, I had pc software to do flight plans and you could watch it traverse the IFR routes. It was a series of eliminations or, perhaps more specifically, optimizations. I don't have to design that software, I know it's already been designed and field tested. When applying it to a Tesla, you add the prime rule, don't hit anything. If I had gone through years of someone promising this, yes, I'd be jaundiced too.

 

[Knightshade]

I imagine that Tesla really, REALLY wants to realize FSD revenue that they have taken on AP 2.5 systems over the last 1.5+ years. All they have to do to realize that is deliver a new feature and call it FSD and book all that revenue.

I don't think that's how that works.

You can't deliver "part" of a product and recognize all the revenue for it with the bulk of the product still not delivered.

It'd be like if Tesla shipped just the license plate holder to Model Y pre-orders and recognized the $2500 per car revenue immediately. That'd get laughed out of tax court.

One tactic is to book the revenue, install the HW 3.0 on new builds only for the first quarter or two and let the hardware cost of the new card drop a bit before they start manufacturing the ones that will be retrofit into existing vehicles that bought the new feature.

As pointed out in another thread, Teslas quanity is tiny by custom processor standards... and it's not like anybody else is making the part to help drive costs down further so they're not gonna see such reductions.

On the contrary it's probably cheaper to bulk order them from whatever fab is manufacturing them than to wait and do a separate run for retrofits.

Anyone want to hazard a guess on how much the AP 3.0 guts cost Tesla? My guess is the board costs <$1,000.

Same as AP2.5, according to the investor call and elsewhere

They already have different versions of hardware for FSD.

Example 1:
HW2.0 has different cameras than HW2.5. That's why Sentry Mode, Dog Mode, Dashcam and Blind Spot Chime don't work with HW2.0. Do you think they will replace the cameras of HW2.0 too when retrofitting the HW3.0 computer? I'm not sure about that.

At the time of the 2.5 announcement Tesla explicitly said they expect they can still do FSD with 2.0, but if they can't they'd upgrade the 2.0 HW for free.

So if it turns out the cameras don't work for FSD yes I'd expect they'd replace them- but I don't know that we have any evidence that's the case.

Dashcam isn't an FSD feature so the fact they can't get "good looking" video for that use isn't relevant.

 

[wcorey]

I'm of the opinion that some employees already have HW3.

Oh, they absolutely do! And they have FSD as well. It's already in round two of employee testing, call that alpha testing or internal beta. Another reason I am confident in 12/31/19.

 

[RedBall]

For what it's worth, and if it helps discussion on timelines for HW3, my order (UK vehicle with APH4) went into production 16th March. I was notified yesterday 26th it was built and is now being transported.