Hacker News new | past | comments | ask | show | jobs | submit login
Tesla's Simultaneous Brilliance and Incompetence Revealed in Teardown of Model 3 (forbes.com/sites/joannmuller)
214 points by Maro on April 27, 2018 | hide | past | favorite | 178 comments



Is the intellectual asymmetry between creating a design and analyzing it so large that a small teardown team can reliably analyze trade-offs in a design created by a large engineering group?

For example, when they say "this flange is larger than necessary", how can they know all the possible reasons it might be necessary? Perhaps it adds stiffness or crash energy absorption, or is used to grasp the part during assembly, or avoids a resonant frequency.


Monroe & Associates has been tearing down cars and other things as their sole business for quite a while. https://jalopnik.com/the-fascinating-company-that-tears-cars... is a pretty good overview of their business and the insane depths of knowledge that their engineers have.

They’ve accumulated a ton of knowledge about How To Cars. Of course it is possible that Tesla is making a bunch of design decisions that M&A thinks are weird because they are optimizing for very different things than most other carmakers; it would be pretty interesting to be a fly on the wall in a discussion between Tesla’s engineers and M&A’s.


It makes sense. Munroe knows their stuff and it could be wise for Tesla to employ their services at some point for a little analysis. Such a cool and obscure business. There’s a little doc about the company. The extent to which they breakdown and analyze a product is anxiety-inducing.


The only reason Tesla is making bad decisions is because of schedule, everything else stems from that. That said I am pretty positive on Tesla because even thought they release half baked products they still improve and innovate at a breakneck pace. Turns out shit happens when you have a charismatic leader attracting a bunch brilliant and driven people and telling them to go change the world.


Would you be happy that your $70K car was "half baked" so that Tesla could improve more rapidly?


Interesting parallel question: "Would you be happy if your $800 phone didn't have copy-paste functionality, no 3G, and no app store?"

It's an interesting question to consider because that was basically what the iPhone was when it launched. It was behind the state of the art at the time in many important ways (not having 3G was a big deal), but it was so incredibly far ahead in several very critical ways (multitouch display, 100% touch-based OS, full fledged browser, next level music player) that it didn't ultimately matter.

Now the question is, does Tesla strike that balance? The second question is, can they iterate rapidly and close the gap (which the iPhone ultimately did)?


It's an interesting parallel to consider because iPhone was revolutionary, it was something no one had seen when it came out, it was literally the first smartphone.

So, are you saying Tesla's cars with their design flaws are revolutionary like the first smartphone?


It absolutely was not the first smartphone; Nokia E-series predated by several years.

Nor was it the first keyboardless touch-screen phone, that was the LG Prada.

The iPhone was a clever subset of well-polished features very well marketed. The usual Apple approach.


Yes.

Mass produced viable electric cars, with a mileage of 300 miles.

A bigger revolution than the iPhone.


Yes, go test drive one and see for yourself.


My phone can't kill me, nor others.


Good thing body panel gaps don't kill many people as well!


I would not, which is why I do not have a model 3. People who have Teslas right now are early adopters.

A question back: would you be happy if Tesla did not kick-start electric car ramp up?


I don't own one; don't plan to buy one. So I'm indifferent.


I’d rather the structure were over-engineered where it counted.


They've been in the business, probably are experts at cost reduction engineering. Their business is selling this information to other auto companies so they must be right enough that they can get paid for it.

There are potential reasons why the mentioned flange might be oversized, I'm having a hard time imagining how they tested frequency resonance and decided to just fix it by leaving a little extra material near where the welds go though. Or likewise deciding that the extra material in the weld zone increases safety by some margin. Those possibilities seem like they'd require and extremely nuanced and subtle understanding of the car and the whole process, It's possible, I guess. What's far more likely is that they just haven't done a round of cost reduction as they're just struggling to get the initial manufacturing done for the pre-orders.


Or what's more likely is that Tesla didn't decide to make giant flanges because they're idiots, and they're not going to outright tell M&A what the reasoning is, because this is exactly the information they want to be able to sell to other auto manufacturers.


Munro and Associates was an engineering consultancy known for reverse-engineering not just automobiles, but everything from airplanes to household items like rice cookers, to learn how they work.

A previous tear-down was a lot harsher https://jalopnik.com/tesla-model-3-teardown-by-engineering-f...


Believe it or not, that's referring to the same teardown as this article. It is a multi-month process.


> Is the intellectual asymmetry between creating a design and analyzing it so large that a small teardown team can reliably analyze trade-offs in a design created by a large engineering group?

I don't understand this. So you are saying the reviewer's small team doesn't understand the expertise which might have gone in the car.

But isn't that true for every product review and article out there? Take for example a phone. If a reviewer doubts the CPU benchmark on a phone, are we going to talk about how a single reviewer cannot understand the trade-off a bigger team at Apple/Samsung/Google has put?

Or if a journalist talking about FB's decision on data sharing has the capacity to know the trade-offs made by a much larger technical and business group at FB.

We can only look at the past work. If HN is voting this up then we know these guys have done some good work in the past.

Unless of course we think everyone who doubts Tesla's expertise doesn't understand their genius.


Munro and Associates is the firm that does this kind of analysis in the automotive industry.

It's really fascinating. For the record, it would be hard to produce a design that Munro couldn't find something to talk about, good or bad. They employ extremely gifted engineers with a ridiculously large amount of experience at doing this exact task.

As a software engineer, I wish there were an equivalent firm in the industry that did teardowns like this so that we could, as an industry, improve over time the way that the automotive industry does.

I feel like it may be one of the factors holding software engineering back from achieving true professional engineering standardization.


> As a software engineer, I wish there were an equivalent firm in the industry that did teardowns like this so that we could, as an industry, improve over time the way that the automotive industry does.

I think this thread is a good indicator of why that might never happen. People in software will argue with whatever stuff they can and try to refute a review/teardown - reviewer doesn't understand the design consideration, reviewer don't have numbers, the stuff is revolutionary etc etc.


>As a software engineer, I wish there were an equivalent firm in the industry that did teardowns like this so that we could, as an industry, improve over time the way that the automotive industry does.

There are quite a lot of security companies that reverse engineer other software products and publish their findings. It would be good to see that for non-security stuff too.


Do you want your bosses going to a firm like Monroe.

Because companies will occasionally send in their own product to see if they can squeeze their manufacturers and see what their engineers missed.


Yes, honestly, I do. I would absolutely appreciate detailed and specific feedback across all of my work.

How else are we expected to improve? Even when it's thing subject to opinion, I'd like to hear those other opinions.


I feel like you get a flavor of this for software by reading the mailing lists in the more mature open source project, you can basically get a continuous flow of analysis and build or rebuild. It's not packaged in a nice episodic report form, but you can get a little of that too with some projects.

For the software industry as a whole, the closest thing might be the SEI[1], but that body of work takes some effort to work through, perhaps suffering a bit from a more adademic roots vs a more industry engineering info exchange approach like M&A. For that you might have specialist areas, for example, the Jepsen tests and blog[2].

[1] https://en.wikipedia.org/wiki/Software_Engineering_Institute [2] https://aphyr.com/tags/Jepsen


No they are not. Car companies do not go to Munro, period.

There are far more reputable sources, speaking from someone in the industry. Munro is just trying to get free advertising with this sh*t.


That's interesting. Which sources are in your view more reputable? Where do you work? One of the big three or a Japanese carmaker?


Car companies do not go to Munro, period.

Really? So where do they get their income?

The comment below has some questions for you. This is all information you would have provided if you actually knew what you were talking about. The fact that you didn't leads me to think you are a troller who just likes to say things that make people angry.


> Car companies do not go to Munro, period.

I don't know if you're trolling or what, but the truth is interesting. None of the US domestic automakers buy reports from Munro. They sell information to Japanese, Korean, Chinese and European automakers.


> If a reviewer doubts the CPU benchmark on a phone, are we going to talk about how a single reviewer cannot understand the trade-off a bigger team at Apple/Samsung/Google has put?

The equivalent for the GP's claim would be the reviewer claiming that a phone's CPU is underpowered and that a more powerful one could have been used instead. That assumes a lot of knowledge about the tradeoffs that the phone designer had to consider - technical, logistical, financial, etc.

I don't know anything about Monroe & Associates' expertise, but I agree with the GP that this is a high bar to clear.


You don't need to be an expert to criticize the user experience of a product. A reviewer who says "this phone is slow" is probably right. It requires much deeper knowledge to criticize internal details. A reviewer who says "they used too big a capacitor for C143" would need to understand a great deal about the design for that claim to be credible.

Most of the negatives in the article seemed to be of the second kind.


> f a reviewer doubts the CPU benchmark on a phone, are we going to talk about how a single reviewer cannot understand the trade-off a bigger team at Apple/Samsung/Google has put?

This makes no sense. Yes, if a reviewer doubts it, it means nothing unless they can prove it with numbers.


It's the same ot worse for phones, yes.


> Is the intellectual asymmetry between creating a design and analyzing it so large that a small teardown team can reliably analyze trade-offs in a design created by a large engineering group?

Is it easier to verify a solution to a problem than it is to produce it?

million dollar question.



NP means that a solution can be verified to be valid. Like the reviewer saying “yep, this is a car all right”. But verifying whether a solution is optimal is not generally even in NP.


You can use the decision version of an optimization problem to answer the optimization problem.

For example when you're looking for the fastest route, you're not directly asking "is this the fastest route?" You're asking "Is there a route shorter than length x?" It's a yes or no question, that's easy to verify given a certificate. If the answer is "yes" then the certificate is a route produced that is shorter than length x.

Assuming the existence of an algorithm that can reliably produce a verifiable "yes" with a certificate or "no", you ask "Is there a route shorter than length x?" and if the answer is yes (which you can easily verify), you keep reducing the length of x in the question until the answer to the question is not yes (no certificate can be produced). The certificate of the limiting "yes" is the optimal solution.


That technique isn't as powerful as sometimes believed. In your example, you've assumed the uncomputable thing, an algorithm which reliably produces a solution to a problem for which one exists. Consider, "Is there a proof in fewer than x lines of the Riemann hypothesis?"


I don’t think that a candidate solution to your problem statement can be deterministically verified in polynomial time.

Any candidate solution produced would be subjective to the interpreter who is verifying it as to whether it satisfies the requirements of a proof (maybe they want more definitions, less shorthand notation, maybe in french, references in APA style with full names, etc, all of which would effect the length of the proof).

So whereas a candidate solution to your question is difficult to verify, it is not in this P vs NP discussion.


Verifying a solution is harder than producing one. Identifying a flaw in a solution is easier than either.


Not sure what you are referring to -- at least in the P/NP problem complexity classification verifying the solution is easy, coming up is P/NP-hard, i.e hard.


So if I gave you a flawless solution to a sudoku problem, would you tell me that it was harder for you to verify it than it was for the solution to be produced?


I had to do exactly this in a practical coding interview. Fun was had.


I think that really depends on the problem space. If I had to hire a sales team to increase revenue, then I hire them and check the revenue numbers (with basic accounting for other secondary variables), its pretty straight forward to validate.


Munro claims to have a lot of engineering an tear down experience. I guess that's what they sell. That said I don't expect them to be 100% correct on all the trade-offs.


Wouldn’t those larger flanges provide stiffness for a car that is much heavier, and can deliver to the wheels a boat load more torque, than similar sized ICE powdered cars.

And yeah, an electric motor is going to have vastly different vibration characteristics than an ICE.


agree, and would like to add that businesses can either spend time on figuring out the most efficient flange design or spend that time on improving a feature that will bring higher customer value. I would suspect most people are not buying cars because the flange is not the perfect size, or that there are overlapping panels.

Take any gen 1 product apart from an innovative company, I am sure you will find skeletons.


So long as customers buying those gen 1 products know what they are buying. No company will transparently disclose their skeletons, so I'm glad we have reviewers that do.


It seems as they can't reliably analyze the decisions.

>highlighting areas where Tesla needlessly added weight with things like excess metal flanges and overlapping layers of steel. "This adds weight without value," he said.

it feels rather presumptuous to assume that the added value isn't safety.. which isn't something you can assess in a teardown.

In the next paragraph:

"He was also puzzled by Tesla's unconventional use of multiple welding techniques in close proximity to one another. "I don't get it," he said. "There's a lot of technology [used] here, but what we don't understand is why they used the technology they did."

Not understanding something != bad


For a lay person, sure. But these are experts at automotive engineering. Imagine if you were reviewing a new developer's code, and you were puzzled by a particularly non-idiomatic implementation where you can't see the reasons for avoiding the idiomatic version. Are you going to think that maybe this is some technique you're unfamiliar with, or that the person who'se still quite new to developing might not be aware of the idiomatic verions?


So a good example of this in code would be:

    a = a xor b; b = a xor b; a = a xor b;
In this case it is a technique for swapping two variables without using a third variable. But it really serves no efficiency purpose in most applications (however it is fun to use).


Which comes with the obligatory reminder that it fails when a == b.


No it doesn't. xor swap works fine when the two values are equal. The problem is when &a == &b.


I get where you're coming from, but Tesla engineers have seen those flanges and that extra steel in crash tests where these engineers have not.

Your example would be more akin to assessing someone's code and pointing out potential flaws without ever executing it.


> Your example would be more akin to assessing someone's code and pointing out potential flaws without ever executing it.

You've just described a code review, which is standard practice and not controversial. If an expert doesn't understand your code, maybe the criticism is informed by their experience.


On the other hand: "take note that I have only proven this code correct, not tested it." Code review is performed in the understanding that its results are basically heuristic. Everyone knows that actual behavior trumps anything you get from just looking at something.


They are iterating and optimizing the hardware rapidly treating it just like software. And pushing out versions that work fine to improve stuff later. It’s as if the car is communicating the Silicon Valley ethos with its seemingly random features that are simply optimized towards something that can be shipped ASAP. Interesting how one can deduce culture just by looking deep inside an artefact. One day aliens will try to decipher how early humans behaved by analyzing excavated iPhones.


One of the problems with that is that hardware is not software and at some point it will have to be repaired. It's bad enough when a part changes design halfway through the model year. I can't imagine trying to repair a Model 3 five years from now and finding out that e.g., the right front wheel bearing has 10 different versions and it's not completely clear which one I need.


Don't worry, you aren't allowed to repair your tesla vehicle anyway.


As I understand it, Tesla does not really do model years. They seem to incrementally change minor things as needed and sometimes do a more thorough refresh, but it's not really tied to the calendar in the traditional auto marketing way.


You call Tesla give your serial they will tell you the latest version of the bearing that you should upgrade too. They been d9o ing that for 10 years now.


Oh, that will work so well when they have 10 million of these things running around!


One day aliens will try to decipher how early humans behaved by analyzing excavated iPhones.

Those aliens will be our descendants. And they won't have to excavate. They'll just look up the right shop in Akihabara, or equivalent of the time.

http://kk.org/thetechnium/technologies-do/


Thanks for Radium suppositories! What if humans have long left the Milky Way galaxy and aliens visit us then?


What if humans have long left the Milky Way galaxy

If out descendants become intergalactic spacefarers, then we probably won't ever all leave.

and aliens visit us then?

Those aliens will probably be our descendants. They'll just have to look up the information in their archives. It's not like keeping raw information is going to become so expensive, we have to delete it.


Imagine a universe so old and so populated that a good portion of the matter/energy available in the universe must be devoted to storing the histories of the various civilizations. Wars would break out over data resources, as factions war over access to storage and compute power.

That assumes a future where humans colonize the stars and spread life through the universe. What's more likely for humanity is a future more like the matrix, except instead of being enslaved, we place ourselves there willingly. It seems to me the focus of most of our technology has been to figure out how to poke and prod our pleasure sensors to maximum effect. Forget the stars, we're going to invent our own personal paradises on Earth and just stay here.


Imagine a universe so old and so populated that a good portion of the matter/energy available in the universe must be devoted to storing the histories of the various civilizations.

https://en.wikipedia.org/wiki/The_Library_of_Babel

Perhaps by then, some very clever entity will have figured out a quantum computer implementation of The Library of Babel. It won't actually store all possible books, but it could be used to very quickly search the space by generating all remotely coherent possible books.

What's more likely for humanity is a future more like the matrix, except instead of being enslaved, we place ourselves there willingly.

Facebook.


"Why spend a short life in the real world of sadness and absurdity when you could live forever in paradise?"

https://www.smbc-comics.com/?id=3183


Or maybe present Earth is a single cell in Their giant cellular automaton machine.


At that point the war is for energy.


That's great if you just want to fleece the next round of VCs before you fold but cars last a decade or more and you can't just keep on building half baked junk and expect consumers not to catch on. Just because your lease ends at 36mo doesn't mean how the vehicle performs after that doesn't affect the company's reputation.


There is no indication that they are shipping half-baked junk, though?

I mean, the early S models had drive unit problems and the X hasn't been perfect, but neither have been plagued with major long-term reliability problems. Some established luxury car makers have made far more unreliable vehicles.


>Most confounding to Munro and his team was the body construction of the Model 3. "This car is the heaviest body-in-white I've ever seen," he said, calling the construction "ridiculous" and highlighting areas where Tesla needlessly added weight with things like excess metal flanges and overlapping layers of steel. "This adds weight without value," he said.

Optimizing structural design is hard. There are plenty of examples of talented young engineers with lots of budget who can make a really clever and strong design, but it will frequently also be overengineered and far over weight or cost for its design purpose. An infamous example would be the Juicero juicer[0]. It takes a lot of experience and knowledge to make a highly optimized frame or structural component.

[0] https://www.youtube.com/watch?v=_Cp-BGQfpHQ


Maybe somewhat OT, but the flange review triggers some of the same criticisms I have of AvE.

With Uncle B, any time he gets something wrong (and there are plenty of examples) it's always "I'm just a dude in the shop, having a laugh, etc. etc." But when he's tearing things about it's always, "this design is shit, these engineers suck, they're trying to fuck you over here there and everywhere." There's a huge disconnect in how he presents these critiques as though they're gospel truth, but he won't eat the humble pie. There's the exception here or there, he'll nibble at a particularly big piece, but overall it just doesn't serve the purpose of learning and understanding. He veers into some sort of moralizing about modern manufacturing.

Don't get me wrong, on balance it's great and the juice-squeezer definitely deserved it, but it keeps me wary.

In the same sense, I don't know about that flange. Is it wasteful? Could well be. Or maybe there's a good reason it's there. What I'd like is a good explanation of how they can be sure it's excessive. Tell me why. Maybe they do; I haven't a bit of the autoline video, but I don't see a good writeup.


In order to know the how's and why's of the M&A teardown, you probably have to pay. I'm certain that their detailed analysis includes good explanations of how they know that flange is excessive, but they're not going to tell you for free.


I'm with you to a point on AvE - I mostly watch him for kicks. However, the guy's used a lot of tools for a long time. He might actually have the experience and appropriate exposure to know what does and doesn't fail or endure under regular use.

But yeah. He says it all with virtually no reserve, haha. Take most of it with a grain of salt. He does seem to know a lot about materials, though. Definitely not a dumb guy. His brain is chooching at around 50% at least


The overbuilt body is most likely due to two factors: (i) Tesla does not have the design know-how to do better (ii) Tesla's manufacturing capability is below that of other car makers. As I've recently outlined in another comment, the manufacturing process for these is quite involved and finely tuned. People who do (i) also don't grow on trees, because self-supporting car bodies are basically the pinnacle of that particular engineering discipline; essentially no one outside that industry has a need for those requirements in that combination and manifestation.


> It takes a lot of experience and knowledge to make a highly optimized frame or structural component.

And if you don't optimize your competitors will and they'll deliver better value and take your market share.


If there is anything I want over engineered it is all things infrastructure. Bridges, buildings, planes and boats etc.

Over-engineer them when life is on the line.


Overbuilt and overengineered are two different things.

Overbuild a car and you have mass in all the wrong places. Do it right, and you get passenger protection and crumple zones.

The problems with over engineering are probably more subtle - you want time and money spent effectively.


there's no reason to believe overengineering leads to safer results, yet alone competitive on the market or fit for the problem statement. case in point: airbus a380


I thought the basic problem with the A380 is that it's simply too big for how airlines want to run the majority of routes these days?


4 engine aircraft are technically more safe than 2 engine aircraft. Boeing and Airbus have just gotten so good at engineering safe 2 engine aircraft that companies can run them over oceans without too much trouble.

https://www.youtube.com/watch?v=HSxSgbNQi-g


A four engine aircraft has four times the probability of an engine failure as a single engine aircraft. Of course it still has three engines left if that happens


And the fires


For contrast, here's Sandy Munro on the Autoline Network show talking about his teardown of the BMW i3:

https://youtu.be/uDr4L6BzpP8


And here's him discussing the Model 3 teardown on Autoline (don't see a link to this in the article)

https://www.youtube.com/watch?v=CpCrkO1x-Qo


He is really praising the batteries and electronics as revolutionary. "Other automakers ignore Tesla's electronic at their peril, but Tesla ignored traditional car manufacturing so they are in peril."

At one point Munro says that had Tesla outsourced the manufacture and design of of the car to a company like Magna they would have hit every target and even Toyota would be crapping their pants now. If the traditional mechanical parts of the car would have been even decent, Tesla would have mopped the floor with everybody.


>At one point Munro says that had Tesla outsourced the manufacture and design of of the car to a company like Magna they would have hit every target and even Toyota would be crapping their pants now.

This would lead to hitting short-term targets at the expense of never really learning how to perform such tasks well in-house. Being able to design and manufacture cars seems pretty important stuff for a car maker to know how to do.


They are now risking with running out of money and bankruptcy.

It would have been better idea to make money from outsourced Model 3 and continue R&D.


You make the assumption that people won't give them more money.


Their debt is already rated junk. Any new debt comes with huge interest rate. Tesla's whole future is tied to Model 3 sales. They don't have another change if it fails.


> He is really praising the batteries and electronics as revolutionary.

His opinion on the suspension design was interesting as well. He was so enthusiastic about it that he said the designer could do an F1 car, but in another part of the interview he mentioned that it was a questionable design for a car that was meant to hit a $37k price point.


Anyone with even a passing interest in this kind of engineering owes it to themselves to watch these videos. It's just utterly fascinating. At least one of his interviews with Autoline talked about the business of costing and reverse engineering, which is almost as interesting as the pure engineering.

He has a sense of humor about the controversy Tesla stuff inevitably engenders online.

Craig Cole (on measured gaps and mechanical tolerances): i3 was the best. Where does the Model 3 fall on that scale? Way at the bottom, is it somewhere in the middle... can you share that?

Sandy Munro: Now, I'm pretty sure there's gonna be Tesla fans here so I really don't know if I wanna say the truth, so, I will give you a nice lie. It's in the lower half.


I don't drive and don't care about cars in general, but I'm just riveted by this video.


I just read a book about Henry Ford. They evolved the model T rapidly so models coming off the line weeks or months apart had significant differences. They kept evolving it for years

Reminded me of the model 3. So, I don't have a ton of faith in a single datapoint on Tesla build quality.


Skunk Works did the same. Ben Rich mentions how by the time #9 of a plane rolled off the manufacturing floor they had devised new methods or better way of doing things. With meticulous record keeping they would be able to go back and re-engineer earlier planes.


One doesn't even need to go that far back to find relevant examples. The head of manufacturing at Spacex recently started that no two Falcon 9 cores were built identically.


The Model T cost being so cheap due to supply and assembly line innovation is a big difference.

The Model 3 at $49k starting price messes up the comparison. A shame the $35k version remains vaporware since that would be a more apt comparison I think. People can and should be pickier for that extra $14k.


>The fact that Tesla can simultaneously demonstrate brilliance and incompetence in the same vehicle is not surprising, considering it is a high-tech software company trying to master complex manufacturing processes that took established automakers more than 100 years to perfect.

This is wrong. At the very least it would be since the 60's and probably later. Seeing as 1963 was the first time CNC was used by the auto industry as far as I can tell. While they have been making cars for over 100 years you really can't include the time before technologies like CAD and CAM. And even those took a while to see widespread usage. Add to that the ability modern CAD software has to model stresses bases on the different forces in play makes a hell of a difference.

Even just going back 12 years and the amount of computing you have access to vastly changes. That was when AWS came out. AWS lets you compute massive amount of engineering data without having to invest into a super computer. (Not to say the auto industry did not have those super computers before then.)


> The fact that Tesla can simultaneously demonstrate brilliance and incompetence in the same vehicle is not surprising, considering it is a high-tech software company trying to master complex manufacturing processes that took established automakers more than 100 years to perfect.

>> This is wrong. At the very least it would be since the 60's and probably later.

If you watch the video on Youtube (that someone else posted here) where Munro is talking about the Model 3, (jump to 36m20s, he gets into it about 15-20s later [1]), you'll realize that the "complex manufacturing processes that took established automakers more than 100 years to perfect" are not about advances in tooling/tech, but advances in quality control processes. If you don't want to watch the video, the acronyms Munro drops are AIAG, PPS, PPAP and APQP.

Things get even more interesting when someone on the panel asks him what would have happened if Tesla would have subbed out the design and manufacturing to a company like Magna.

[1] https://youtu.be/CpCrkO1x-Qo?t=36m20s


I think it's reasonable to talk of a culture of engineering (or design) going back earlier than a given technology. A large company has to make sure at any given that a new approach does everything that an older approach does. The company may have access large customers for testing and requirements gathering so they can wind-up doing technology X "correctly" even if they start-out behind on technology X.

With the variety of changes different companies go through (being bought, sold, spun-off, etc), it's certainly an interesting question whether a company retains a given "core competency" or any core competency at a given time.

I know in the case of CNC manufacturing, the devices were essentially programmable lathes and one can talk of a "CNC programmer" but the system was designed to leverage the existing knowledge and population of manual lathe operators and there's no relation to computer programmers and instead the "touch stone" is ordinary lath operation I believe (a CNC programmer makes $22/hour average conveniently). Thus it's reasonable to say manufacturing culture in the US references things earlier than CNC.

You can see a similar thing in the way Photoshop, Illustrator and cousins ape paper and pencil tools - the standards, skills and terminology of layout were carried through the transition (to the point that these industry standard tools have an interface that seems fairly pathological to newbies).


I know in the case of CNC manufacturing, the devices were essentially programmable lathes and one can talk of a "CNC programmer" but the system was designed to leverage the existing knowledge and population of manual lathe operators and there's no relation to computer programmers and instead the "touch stone" is ordinary lath operation

There is a fascinating book "Forces of Production" [0] that analyzes the history of the development of numerical control in the 1960s that argues exactly the opposite. I read it many years ago, but I will try to summarize it TLDR style below:

Before software ate the world there were two competing approaches to automating machine tools:

- Record and Playback: a skilled machinist makes the part and the motions and operations are recorded. Subsequent parts can be made by playing back the recording on a machine with the assistance of a less skilled worker.

- Numerical Control: A white collar worker in an office writes a script (in G Code [1]) to control the tool which is then played on the machine attended by a less skilled worker.

This was a hard problem at the time and technically Record and Playback was easier and worked better. However management, particularly at GE, resented the power of the skilled machinists who had effective unions and were not replaceable in the event of a strike. So despite the advantages of Record/Playback they persisted for years to create the NC technology as it let them replace highly skilled union labor with coding monkeys. The parallels with the present day seem obvious but oddly underappreciated.

[0] https://www.goodreads.com/book/show/625055.Forces_of_Product...

[1] https://en.wikipedia.org/wiki/G-code


That is indeed an interesting story but my searches seem to say that even G-codes are now entered by machinists rather than office workers or specialized G-code programmers ( a G-code programming guide describes it as final thing a machinist learns in their career).

My guess is that management was conflicted in what it wanted - on the one hand, a drop-in replacement system and on the other hand, a system so simple they could hire people off the street to do it.


As I remember from the book, the machinists fought this pretty hard and eventually forced a compromise resulting in the situation you describe.


> At the very least it would be since the 60's and probably later.

So Tesla is a high-tech software company trying to master complex manufacturing processes that took established automakers more than 50 years to perfect? I'm not sure that really changes the implicit argument.


This quote is brutal:

> A Tesla spokesperson said the primary car evaluated by Munro was built in 2017, adding: "We have significantly refined our production processes since then, and while there’s always room for improvement, our data already shows that Model 3 quality is rapidly getting better.”

Sounds like Tesla is agreeing that they sold sub-par vehicles?


Warren Buffett: "I am 7 billion dollars richer in 2017 than I was in 2016."

You: "So you were poor in 2016 then?"


That has nothing at all to do with the changing quality of a manufactured product.


Correct!


Using more resource than necessary to build a product doesn't it has deficiencies from the user point of view.


But they admit it's not just a matter of resources, the quality is changing: "Model 3 quality is rapidly getting better". That means Tesla itself is admitting there are some Model 3s out there better than others, pretty wild for a new car.


> pretty wild for a new car

This is the heart of the question. My understanding is that Tesla Motors, unlike most every other car maker, iterating their manufacturing (hardware) side as quickly as they are iterating on the software side.

For a more concrete example, the initial version of the SpaceX Merlin engine had 76,000 pounds of thrust. That same engine now has 205,000 pounds. I saw an interview last week where an employee at SpaceX said that no two Falcon 9 rockets come off the line exactly the same. Iterative hardware improvements are going on all the time. This is unheard of in aerospace.

This is a huge advantage compared to traditional manufacturing approaches.


How can they be certain the improvements don't have side effects or downsides. If every iteration of software is an improvement, we would have bug free software!


Applying considered and appropriate effort to anything will tend to improve the thing. Its why we have smart phones and cities today.

It's not an absolute; I've seen good software lose its quality with inappropriate and ill-considered effort put into it, but that's not the norm.

More concretely, each of the previous three Tesla car models have improved steadily over the years.


I think it means that they are claiming the production quality of their cars has increased, and will continue to increase over time.


So, don’t buy a Model 3 now. Wait until it’s a lot better.


This sounds like something I tell people when they ask me when I should buy a computer and my answer is "next week". The next computer will always be better than the one you can get today. However, you can't really use a computer you haven't purchased yet, so it's not really better in practice.


Well, why? If they are using more material than necessary, or wasting time due to different soldering techniques, why do you care? What matters is that the end result performs as expected. If they are not hitting their production or profitability targets, it's their problem.


Elon Musk says the quality has vastly improved and continues to improve. That means today’s Model 3 is worse than tomorrow’s and yesterday’s is worse than today’s.


That's true of almost all products, and certainly of most cars. A 2017 Honda Civic is better than a 1977 Honda Civic. But you can't drive a 2017 Civic in 1978.


Sure, but within a model generation (eg E90 BMW, 2007-2012), the cars are largely the same.


I'm no gearhead, but the change from the N54 to the N55 engine doesn't seem inconsequential.


The old saying "Don't install a new Windows version until SP2 is out" is still true :)


When Microsoft treated their customers as beta testers... It was a bad thing.

When Tesla treats their customers as beta tester... It's a good thing?


In this case may be but the statement itself says now is better than previously without explicit admission that previous quality was bad.

It’s like apple, every year iphone is the best iohone yet, but that doesn’t imply directly that the previous model was bad.


It would be huge news if Apple said new iPhone X's are better than launch day iPhone X's...


The automobile market is not like smartphones. In spite of assembly lines, the identical unit is not stamped out one million times and sealed in glue.

Car designs are quite modular and individual components and suppliers are revised during manufacture. There are far more intermediate sub-assembly steps and smaller volumes produced in each batch. It is common to see recalls and technical service bulletins that "patch" cars already delivered to gain some of the fixes devised after the car left the line. Sometimes these are direct replacements of parts,and sometimes they are literal patches of the older design, such as applying an extra gasket or some adhesive to resolve a rattle in an interior trim piece which was designed away in its next revision.

While marketing maintains "model years" as a nominally uniform offering of a particular car, it is not uncommon to find odd half-year or worse hybrids which mix together cosmetic and mechanical features of different years. These happen when the manufacturer drains out certain supply chains and integrates unusual combinations of the different modules into specific cars. People focus on the visible parts like cosmetic body trim or interior features, but this also applies to mechanical and electronic parts buried deep inside the car.


Would it be news if Apple said that new iPhone X's cost less to build than launch day iPhone X's ?


No, because that doesn’t affect the customer. Tesla’s most loyal and gung-ho customers are punished for their enthusiasm by getting the worst Model 3s that will ever be made.


What? they had plenty times defects were found in early production model and fixed down the road.


> Everything that's below the floor pan is amazing. Everything above it is what it is.

Honest question: Is everything below the floor pan designed and built by Panasonic?


No, only the battery cells: http://img.ismap.com/nimg/6c/67/e370d981467b8ba457e178916006...

The battery modules (a group of cells) and the battery packs (a group of battery modules) are completely engineered by Tesla.

I recommend this series of videos about Tesla's battery modules: https://www.youtube.com/watch?v=P7GR5fERXNY


Are you sure? https://www.tesla.com/blog/tesla-and-panasonic-collaborate (emphasis added):

“The parties intend for Panasonic to begin PV cell _and_module_ production at the Buffalo facility in 2017. […] Shuuji Okayama, Vice-president, Eco Solutions Company of Panasonic, added, “_Panasonic_PV_cells_and_modules_ boast industry-leading power generation performance, and achieve high quality and reliability.””

That first phrase leaves open that Tesla designed the modules that Panasonic produces, but I can’t read that second statement other than that Panasonic owns the design of the modules, and likely also designed it.

Edit: https://www.tesla.com/presskit#gigafactory says “The Gigafactory is producing a Tesla/Panasonic designed and engineered cell called a “2170 cell,” as the diameter is 21mm in x 70mm in height.”, implicating that Tesla even had a hand in designing the cells. They don’t make it easy to get a clear view of how ownership of the technology is divided between Panasonic and Tesla.


The Buffalo facility makes photovoltaic (PV) cells and modules, thats solar panels. The 2170 cells are made in the Nevada gigafactory by Panasonic and integrated into modules by Tesla.


He was very complimentary of the circuit boards, saying he has no idea where they were sourced and basically saying they were the only company using cutting edge circuitry of a quality and density similar to the latest cell phones.


Tesla's former VP of Autopilot Software&Hardware (oversees all low-voltage hardware, aka everything else besides High Voltage battery drivetrain) was Jim Keller[1][2], who also designed Apple's System on a Chip A4 mobile processor and an early AMD chip architect. So that makes a lot of sense.

[1] https://electrek.co/2018/04/25/tesla-autopilot-jim-keller-le...

[2] https://en.wikipedia.org/wiki/Jim_Keller_(engineer)


Jim Keller didn't have anything to do with developing the hardware for hw2. The board is from Nvidia, it's a stripped down variant of their Drive PX2 computers.

https://en.m.wikipedia.org/wiki/Tesla_Autopilot

https://en.m.wikipedia.org/wiki/Drive_PX-series#Drive_PX_2


No. That board uses a reference design from Nvidia. They redesigned Nvidia reference designed removed one of the fpgas, changed interfaces added GPS directly on board. If you are not familiar a reference design would be selling your prebuilt computer and telling you this works. Afterwards you come in throw the case out and change anything but try to maintain the relevant electronic connections between components. Remove the ones that you don't need and add required ones. Plus design and layout a circuit board with your new requirements and parts.


Which quite surprised me. Aren't circuit boards a large, already well researched and optimized field since they're inside everything? Is the autopilot board better than that of a laptop or smartphone?

Sure, hardware is hard, but it doesn't seem like differentiating knowledge in the tech industry (while maybe the other car makers are a few years behind in optimisation).


If you put a smartphone circuit board into a car it won't make it 5 years until failure from sheer mechanic stress. There is a reason you can buy automotive grade components, and it's not that car makers like to pay more or have less volume.

And ultimately, of course, circuit board weight or dimensions are not really a metric worth optimizing for something that goes in a car. If you can make it cheaper by only having 6 layers instead of 18 for a smaller version, auto makers will choose the penny pinching option every time.


Circuit board layouts can be a nightmare. Automatic routing algorithms suck for networks of realistic sizes, so a lot of the work is done by hand and that is a mixture of tedium and art. If you are designing high power or high frequency circuits you also need to observe a lot of constraints in your layouts as well (distances between parts, potential cross talk between parallel traces, signal travelling time along the traces belonging to the same bus...).


Sure. Doesn't seem to change much for cars though. It doesn't seem like something Tesla has a significant competence in.

Even in high-stress environments, unless I missed something.


The boards he was complimentary of when he was talking about circuit density are just standard off the shelf Nvidia DRIVE systems, not designed at all by Tesla - https://developer.nvidia.com/drive/hardware


The board in the video that Munro is holding up looks bigger and different than those two.

You're 100% sure those are the same?

Munro acknowledges that the chips are from NVidia, but claims the board to be unique.


They are not the same you are correct that board is hand laid by Tesla electrical engineers following loosely Nvidia electrical schematics but removing uneccesaary electrical subsystems and adding ones like on board GPS receiver.


False. NVidia is not supplying off-the-shelf DRIVE systems. They supply the chips that go on them but the board design and manufacturing is by Tesla.


Those boards are from Nvidia. They're stripped down variants of the PX Drive 2.


I thought Panasonic only made the cells and that tesla assembles the batteries, cooling system, electric drive train etc.


https://youtu.be/CpCrkO1x-Qo 90 minute discussion of Tesla 3 with Munro and others on the excellent autoline network youtube channel ('Really heavy, really expensive')


87k for a benchmark report. That’s something. Car makers manufacture in millions so 87k is a tiny expense for a high quality competitive analysis by a 3rd party unbiased firm.

I love tiny business that are the forefront leaders in a small profitable nice.


There's was also this article a few months ago on some glaring flaws in Tesla vs i3

https://news.ycombinator.com/item?id=16305520


Are they working with Tesla to provid this feedback into their production flow? There is no mention of it. Does anyone know?


It would seem odd that as Elon has his experience with SpaceX in the background, and SpaceX engineers are consulting for Tesla, that the metal construction was done overly clumsy without a reason. Equally likely is, that they had a clear plan when designing the metal work on the Model 3 - but it is not obvious to people trained in the way the automobile industry works, as Tesla certainly is thinking differently.


"The fact that Tesla can simultaneously demonstrate brilliance and incompetence in the same vehicle is not surprising, considering it is a high-tech software company trying to master complex manufacturing processes that took established automakers more than 100 years to perfect."

High-tech software company?

Tesla is an automaker, not a software company. Sure, Elon comes from the software/tech/payments industry, but putting Zuckerberg in charge of Exxon wouldn't change the fact that Exxon is an energy/oil company.


I work in one of the big 3's. we do crazy things too, its just that gizmodo isnt interesting in any of it :D I tend to walk in at 9 or 10 am and walk out of 7 or 8 pm. Our culture isn't any different than tech industries (maybe a little exaggeration there) I find it a little intriguing when people label(mislabel) a particular industry.

edit - one of the big 3 automakers


> I tend to walk in at 9 or 10 am and walk out of 7 or 8 pm.

Sounds like a shitty place to work.


How so?


It's fine if you work 4-day weeks. (36 to 44 hours)

With a 5-day week, you're working 45 to 55 hours. That wears on you. It could be OK if the pay is unusually good.


I'm a big advocate for work life balance, but since when did 40 hours/week became the single universal limit for everybody? Humans are not that standardized.


40 hours isn’t the standard. France does 35 and gets better productivity per hour because of it


It's a bit awkwardly worded, but I get their point (I think) - take your example and swap it, for instance: if you were to exchange all the employees at Facebook for employees at Exxon, the company structurally and culturally would still be an energy/oil company even if the product they were making was software. Likewise, Tesla runs, structurally and culturally, like a software company, even though they are building cars.


You definitely don't want to "Move fast and break things" at an oil company. They have very defined procedures for everything. Now days safety is paramount. You can't just change procedures even if its better because they can't afford a blowout or leak or accident. At Exxon you have a manual for everything and deviating from it is frowned upon.


To carry the analogy from above, Exxon moved fast and broke the Prince William Sound. Safety is important obviously but innovation has real costs and software companies seem to learn a lot more and innovate more when they break things vs energy companies-- at least that is the meme.

Also, energy companies move slow and still break things like the environment.


IIRC The alcoholic master handed over navigation to a very inexperienced 3rd mate and left the bridge for him to navigate out the channel by himself. That's an individual screw-up that can happen in any organization's culture. I wouldn't say the whole organization (Exxon) moved fast and broke things.


The boat had a collision avoidance radar which had been non-functional for over a year because Exxon wouldn't spend the money to maintain it.


It collided with an underwater reef.


There was an above-water radar reflector in place specifically so shipboard radar could warn of an impending collision with the reef.


...and an inexperienced 3rd mate was alone on the bridge.


I'd prefer my car not to be broken, even at the cost of innovation. This is a product that protects my family going down a highway at 70mph, not an electronic device used to play candy crush.


I think this is it. Same reason traditional car companies are notoriously bad at software. Different mindset, different culture, different skills, different history and institutional knowledge.


Automotive companies don't usually write large amounts of their own software for infotainment, vehicle controls, machine learning, factory systems engineering and the actual embedded programming for the electronics. Yes Tesla is a software company. They hire a huge amount of software engineers.

Is Apple not a software company because they make phones and give away all their software for those phones for free like Tesla does?


At least they didn't say Tesla was a rocket company.


Another point of view is that Tesla isn't either of those things but rather an energy company.


They're mainly a car maker, energy storage is just about a tenth of their revenue.


for now...


That manufacturing a car isn't something a tech company excels in even though they can build a good battery doesn't surprise me one bit.




Join us for AI Startup School this June 16-17 in San Francisco!

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: