Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[PANN]

BMW Q1 Delivery by Region:

 

[Reciprocity]

But TSLA is for sale. I picked up some recently for 240. Google should have done the same— and if they haven’t yet, they should start now. I’d be happy if Tesla issued new shares to Google (because, as an investor, I prefer $6.5B in cash instead of $4.5B in cash), but if not, Google should pick some up on the open market.

Tesla doesn’t need Google, and Google doesn’t need Tesla, but they could benefit each other, and help de-risk each other’s futures. And I believe cooperation between the two could speed up the transition to sustainable transport and sustainable energy. So, I’m strongly in favor.

It's too bad that the real world doesn't work that way. Elon wants to maintain control so if Google started buying up huge amounts of stock it would actually be seen as hostile. Elon isn't going to allow dilution of his position.

 

[SO16]

I personally write the release notes for all of our code. I include improvements AND bug fixes (although we, ahem, never write any bugs.) Gives me the warm fuzzies.

If I had a cool "get rear-ended and avoid frontal damage by swerving" functionality, I'd be bragging about it in blog posts from here to kingdom-come. I think Elon would too (and rightly so).

I WANT TO BELIEVE

Maybe Tesla wants to make sure it has a pretty high success rate before bragging about it.

 

[Tslynk67]

Was "fate loves irony" an anagram? It's easy to get "fiat" out of that, and "elon" could be in there too.

looney fiats rev
fiat levy sooner
very elon so fiat

These don't make as much sense as I'd like, though. Maybe there's a better joke in there, or maybe it's just a coincidence.

Related: FCA's Maserati brand announced that it will never go all-electric. “This is a brand that needs combustion engines," said Al Gardner, head of Maserati North America. Sales were down 28% in 2018 and down about 40% in 2019Q1 — I'm not sure if that's for USA or global.

So very Elon, Fiat

 

[Bladerskb]

Mobileye's latest presentation from January (which i'm sure you have seen) clearly says mass production is due in march 2021 "The EyeQ® 5 chip was sampled in December ▪ Successfully passed all functional tests ▪ Design wins by 4 OEMs from 2021, volume above 8M ▪ Designed to support 3rd party programmability ▪ Series prod from 3/2021".

Wrong. You need to do better research. EyeQ5 will be production ready Q1 2020 and series production for traditional automakers who have agreed to put it in specific models that will launch starting in 2021.

Also, I'm sure you know Mobileye's chip is just the accelerators and needs to be built into a full computer system with a CPU.

Wrong again. EyeQ5 is NOT just an NN accelerator and doesn't need to be built into a board with a cpu.

EyeQ5 accelerator chip is 24 Tops & 10 Watts. Tesla's NN accelerator chip is 36 TOPs and 15 Watts. So both 2.4 Tops/Watt. Though I would expect EyeQ5 is unlikely to reach its maximum 24 Tops for Tesla's code.

Wrong again, Tesla's FSD chip has 2 NN accelerators, with a total of 72 TOPs and TDP of around 100 watts.

For Intel's full solution, (8-core Intel Atom® C3XX4 SoC and two Mobileye EyeQ®5 SoCs) Intel says this is 1x Xavier in Power and 1.6x Xavier in DL Tops. Presumably this means 48 Tops and 30 Watts or 1.6 Tops/Watt (below Tesla's full system at 144 Tops, 72 Watts & 2.0 Tops/Watt). But again, Tesla's code will likely run more efficiently on its own system, the Mobileye chip isn't built for redundancy in the same way as Tesla's, Mobileye's full system is just 48 Tops vs Tesla's 144 Tops, & Mobileye's chip isn't being mass produced until 2021.

Wrong again, stop doing shallow google searches. You are literally just making stuff up. The picture you saw of Intel Atom SOC, is one of the config of Mobileye's AV KIT. The Atom was supposed to run the RSS validation of the driving policy. Mobileye later settled with 3x EyeQ5 config with a separate backup board with another EyeQ5. A total of 4x EyeQ5, 96 TOPs at 40 watts

Later on..

For Intel's full solution, (8-core Intel Atom® C3XX4 SoC and two Mobileye EyeQ®5 SoCs) Intel says this is 1x Xavier in Power and 1.6x Xavier in DL Tops. Presumably this means 48 Tops and 30 Watts or 1.6 Tops/Watt (below Tesla's full system at 144 Tops, 72 Watts & 2.0 Tops/Watt).

Wow, it seems like you are literally just making stuff up.
The comparison Intel made with Nvidia is of a single EyeQ5 versus a single Xavier chip.
EyeQ5 is 24 TOPS on 10 watts, Xavier is 30 TOPS on 30 watts.

24/10 = 2.4
30/30 = 1
2.4/1 = 2.4

2.4x increase, its literally basic math.

the Mobileye chip isn't built for redundancy in the same way as Tesla's,

Why do you keep making stuff up? Its a theme in this thread i can see.
also Tesla's chip isn't even ASIL-D

Finally, the core of Tesla's strategy is Data First. All their software and hardware choices are built around that. Until somebody else starts rolling out infrastructure to filter, process and upload data from hundreds of thousands of cars, they do not have the same strategy as Tesla and Tesla is not copying their strategy.

They didn't need that to have a 2+ years lead in NN deployment. Which i will post a comparison which i made in Feb 2019 here later today. Also Mobileye is the only one crowdsourcing HD Map and drive-able trajectory data from millions of cars today.

 

[Reciprocity]

2-3 years lead in what exactly? Neural networks? Chips? software? mapping? sensor redundancy? Because they trail in all of that.

LoL, bleederscab is back. Let's just say you're right. And let's say mobileye which started 15 years before Tesla does have some slight lead in all those areas. Which I doubt. The reason being that AP1 is still the best implementation of anything mobileye has ever been part of and Tesla still improves it with every release, but setting that aside. Let's assume for a brief second you are right.

Mobileye is beholding to their clients for the final implementation. Any robotaxi that is not an ev is pointless because once you remove the driver, fuel and maintenance are the next biggest cost and EV has a tremendous lead there. Does Mobileye make EVs? Does anyone?

The data mobileye collects is only map data and nothing like what Tesla can collect, which has everything. Every movement of the wheel, a ever change in speed, every path, every thing the car sees.. everything. Even the thoughts of the driver in some cases. If they saw something that spooked them are turned to avoid something.

Tesla fired mobileye il at about the same time they started their chip. At that same time, mobileye was deep into eyeq4 and it's still not out and won't be before Tesla is feature complete. Even if mobileye was able to finish it before, they won't have then in a million cars, which Tesla will. They won't have when in 1000 EVs much less a million cars. Assuming you're correct again, which you aren't. Tesla went from nothing, to a million cars with eyeq4 level chip in less time then it took mobileye to iterate one step. Tesla had 0 when they fired mobileye and now are set to eclipse them, if not already, in 3 treats with fsd2.0 chip that's already in development.

Oh and BTW, don't bother responding, I don't read any of your responses as I find them lacking facts.

 

[SO16]

Haven't you seen all the mobileye cars signalling on their own and changing Lanes, etc.?

And geez, who would ever think that we should put 3 cameras looking forward!?! Those mobileye guys are genius. Let's file patent for that!!

If amazon can get the photography on white background patent....

United States Patent: 8676045

 

[Bladerskb]

Five comments:

1)

I find it probable/possible that Tesla cross-licensed most of MobilEye's patents that resulted from their early partnership, in which case Intel couldn't sue Tesla based on those patents.

Incorrect, there's is no possibility of a cross license, Tesla has literally 0 autonomous related patents. ZERO. While mobileye has hundreds.

Tesla was MobilEye's main source of income and it's routine to not build a sole IP dependency on such partners.

Hugely incorrect, Tesla represented like 0.1% of Mobileye's income. Mobileye chips are in over 27 million cars.

The assumption that MobilEye was allowed to undermine Tesla via submarine patents financed by Tesla income is IMHO naive. (Anyway, it's pure speculation: I don't know whether they did so (they might not have), and the agreements would generally be confidential.)

This conclusion is derived from a very flawed premise.

Are you claiming that mobileye has developed their own ASIC? If so, where is the proof? if not, how can they even begin to compete on processing power at the same cost and power draw? or is this a six foot square box that draws 2 kilowatts?
Its no good having the most amazing 3d vision system in the world if it costs ten thousands of dollars and adds weight and power-draw to the car.

Yes its called the EyeQ4 the first ever NN processor for self driving which came out Q4 2017 and also the EyeQ5 coming out Q1 2020

 

[Todd Burch]

Maybe Tesla wants to make sure it has a pretty high success rate before bragging about it.

But that's kinda my point. If they weren't confident of its success rate, they wouldn't enable code that has the potential to kill multiple people (swerving into an active lane of traffic).

 

[Boomer19]

bloomberg headline:
bmw car unit posts first loss in decade:

Bloomberg - Are you a robot?

 

[kbM3]

Tesla’s Deep Learning at Scale: Using Billions of Miles to Train Neural Networks

 

[TheTalkingMule]

Can these OT FSD postings move to the Autonomous Vehicles forum?

 

[Pras]

So, my wife just showed me her Facebook feed with a positive Tesla article from CNN. She said, “finally some positive Tesla news”.

Of course she missed the point that the SP being so depressed and the capital raise putting a floor to the SP have removed any incentive to smear Tesla for now. Or maybe the gang of Chanos and Einhorn are taking a breather after the relentless onslaught over the last few months. Even the valiant (pun intended) Einhorn was quoting Napoleon!

Well time will tell, I am still expecting the SP to go to $500 in the next 24 month.

Another thought. I am so looking forward to the day Tesla flicks the switch for FSD..it will be like with one show of hands the Night King raised the army of dead (not a positive analogy but equally powerful for Uber and auto OEMs).

 

[mongo]

But that's kinda my point. If they weren't confident of its success rate, they wouldn't enable code that has the potential to kill multiple people (swerving into an active lane of traffic).

You say "swerve", yet there is no reason to think that, were the maneuver done by the NN, it had not already determined the lane was clear. The system is continually monitoring all cameras, it's not like it has to think 'I want to change lanes, I should now check my blind spot'. It always knows if the lane is open.

NoA already does lane changes without driver intervention. AEB/ FCW does anti-rear ending. If a car has both, it is possible for the car to detect an impending crash and change lanes. Heck, you don't even need to combine them, at a slightly different time scale, this event is no different that what NoA does anyway: if the car is closing on the one in front, change lanes, if possible.

They do list NoA, AEB, and FCW as features. There is no reason to expect them to be testing every combination of events and then enumerating the grouped abilities. Especially since that opens the path for people to complain that their car did not perform maneuver X.

 

[Boomer19]

what is the current production status of batteries? all this other stuff is interesting, but distant.

i kinda lost touch with where they are at gf1 ...how far theyve come with redoing the battery lines. we had 2 done by end of 2018 and the 3rd in q1 2019?
is this correct? does anyone have latest status of this...and given the panasonic news over the last month or so

 

[Reciprocity]

I've said it before, and I'll say it again: access to in-car data about everything about how drivers drive is an insurer's dream come true.

They'll be able to sculpt precise risk profiles for every driver, rather than just broad demographic and accident / ticket history info. Potentially down to the level of "how often do they drive through X intersection in Y direction at Z speed, versus how likely is an accident at X intersection in Y direction at Z speed?" Their pricing will far more accurately reflect actual driver risk profiles vs. its competitors, and adapt in real-time as driving patterns change. So safe drivers will get lower bids and go with them (since competitors won't realize how safe said drivers are), and risky drivers will get higher bids and choose a competitor (since the competitors won't realize how risky they are).

Complete dream come true for them. They should always be able to underbid for safe drivers, however much is necessary to ensure that the vast majority of safe drivers choose them. Any difference between "what they feel they need to bid in order to get safe drivers to choose them" and "what they actually feel they need the premium to be in order to profit", they pocket.

They'll basically be doing to insurance what Google did to advertising.

They could even offer plans that change monthly, so if for example you go on an overseas vacation for several weeks and don't drive your car during that timeperiod, your insurance bill for that month would be very low. And for rental cars, they could generate instantaneous profiles on renters' driving habits, alongside how frequently each car is being rented. If the rental company also shared profile data for the renters, the insurance company could then have its driver profiles follow the users forward when they're not driving Teslas. Or make a distinction between how they drive when they're in their own car vs. when they're in a rental. And they could share data back to partnered rental agencies, which could lead to discounts or surcharges on car rental due to the higher or lower insurance costs. The possibilities are limitless. Insurance companies would kill for this sort of data. Meanwhile, their competitors will be running blind.

Of course, anyone who doesn't want their data shared should always be able to opt out... knowing that that's automatically going to mean no possibility of getting reduced rates.

I think you might be a bit over optimistic on terms of how much data they can leverage, but certainly they have a data advantage much like Google ads, which are the best in terms of the ability to accurate target potential customers.

One low key thing people are missing. Today when a Tesla is damaged in an accident, yes expensive to fix and the insurance pays a body shop much more than it actually costs to fix. They need profits and so does Tesla on the parts. If you all recall, Tesla is also getting into body shops as part of it's servicing. I believe like Tesla Kim had her car fixed at one of these shops. If Tesla is the body shop and parts supplier and insurer, they will vertically cut all the costs in that stack and pass much of that along to customers, the rest will be used to enhance the speed and quality of the services. I met a girl that is working on this specific thing within Tesla. She couldn't talk much about it but she is a big data scientist and specially working on body shops. My guess is that it's about logistics and costs. Where to put them, how much to stock them with parts and how much costs can be squeezed out of the entire system. Tesla knows where and when they sent parts and for which VIN and what the accident was and so on and so on. This data advantage is as important as the driving set of data is for risks in terms is saving costs for the repairs.

 

[bhtooefr]

I do think Tesla finally ran out of buyers with more money than sense. I never could figure out who was buying the superultraloaded Teslas, and was always astonished that people would pay an extra $10,000 for a tenth of a second faster acceleration. Now they're starting to bring it down to more plausible prices where people with a little common sense might spring for it.

Another consideration is that the Roadster could be cannibalizing the S (although probably not X) Performance Ludicrous sales...

OK! So the plan is Albany + Rochester. I still think Syracuse makes more sense, but I can live with Albany + Rochester.

That probably covers more people, though - Albany does a better job than Syracuse would of providing coverage to western MA and VT, and Rochester does a better job than Syracuse would of providing coverage to western NY (which is currently only served by Cleveland, Pittsburgh, and Toronto-Oakville (which is on the other side of an international border). (And, Scranton and Wilkes-Barre wouldn't really be helped by a Syracuse service center, as Devon, PA is already closer.)

...and now I decided to skip forward 100 pages because this thread is ridiculous.

 

[Zaxxon]

Huh, the official Tesla twitter is actually entertaining... when did this happen?

When Elon switched accounts...

 

[Toumal]

Incorrect, there's is no possibility of a cross license, Tesla has literally 0 autonomous related patents. ZERO.

Incorrect, for example WO2018222274, as well as, via Tesla patents new neural AI chip to improve self driving ability:

The automotive firm filed a handful of patents for:

• An accelerated mathematical engine
• A Computational Array Microprocessor system with a variable latency memory access
• A Computational array microprocessor system using non-consecutive data formatting
• A Vector Computational Unit.

We can debate whether MobilEye's higher number of self-driving related patents is of importance. We can't not debate whether or not Tesla has no patents at all in this area, because they do.

Also patents != application. I can apply to, and receive, patents without them having relevance or practical applications.
Furthermore, MobilEye doesn't do fleet learning. Like, at all. I mean sure, if I was working at MobilEye I'd probably write about them in the same enthusiastic tone as you do.

Do you work for MobilEye?

Also, is it now the norm to use user titles to project competence?

 

[Tslynk67]

Hmmm, pre-market just popped 1.5%, everything else in the red. Something, Anything?