Tesla Dojo vs. Nvidia GPUs

So as to not waste precious foolish time I’ve time stamped four parts that I consider better than most Tesla banter. James Douma is well versed in AI

Dojo vs.Nvidia t = 35:05

Mapping t = 46:35

On mapping, consider that the brain is divided into specialized parts. Douma talks about the complexity of the data connection of the Dojo wafers and the importance of mapping. What he seems to have missed is that Dojo could be multitasking which means that the wafers could be running independent of each other.

Hardware 3 & labeling t = 48:40

Free rollout of FSD t = 52:45

The Captain

o o o o o o o o o o o o o o o o o o

At Seeking Alpha someone wrote an article about how a Chinese company (not BYD) could easily eat Tesla’s lunch. I wrote a reply but for some reason I did not post it…

This analysis misses the forest for the trees, it cherry picks the data instead of looking at the whole picture. Let me start by describing how Tesla got to where it is and where it is going.

The best way to describe Tesla’s voyage is organic growth. Let’s build an EV. Lead acid batteries are too heavy, let’s use lithium ion instead. EVs are so different from ICEVs that there is no viable supply chain, we need to make more stuff in house, vertical integration. Roadsters have a tiny market, Tesla needs to work its way into the mass market. Tesla needs to cut costs without cutting corners. Giga castings, over the air updates, shorten the pipeline to customers (the DELL model). Future proof driving, FSD. Need more compute power, Dojo. Need more and better batteries, 4680. Have batteries, build storage and VPPs. Have AI and manufacturing capacity, build humanoid robots. This list is not for the sake of accuracy but for the sake of philosophy. Monolithic vertical integration.

Xiaomi’s entry into the EV market makes about as much sense as Apple’s failed attempt. At least Apple wised up and quit.


This is a great summation, at the same time glosses over some side trips which may or may not have been fruitful. The Roadster, for instance. Was that really necessary to get to the mass market car? (I don’t know, I suspect it was fun and that much was learned along the way, it’s just not clear to me that it was a necessary step.) [Not a criticism; when you are out front in a field you are going to do some things that don’t pan out.]

I put FSD in the same category. Maybe it will come true - someday - but it’s really not ready for prime time today or even in the near future so far as I can see. It might be close, but close doesn’t count when you’re hurtling down the road at 60mph in a 4,000 pound metal container and there are pedestrians and 12,000 pound trucks about.

I must say I don’t see the problem with Xiaomi producing an EV. I mean, why not? There are over 100 manufacturers in China doing so (down from 500 a few years ago.) When the the US industry began there were more than 200 manufacturers, and it wasn’t always “the first” who survived. There were many who came later who did nicely, at least for half a century or so. There were plenty who started early who got knocked out because they chose the wrong path for dealer network or styling or whatever.

Interesting that Xiaomi’s entry is not being built by Xiaomi itself, but is a contract product. At the beginning of the US industry there were nameplates such as Gardiner, Fidelity, Stearns, Chandler, Regal and dozens of others (most were wiped out in the Great Depression). Many of those relied on at least some of the parts (engines, notably, but also others) being manufactured by others in a sort-of contract manufacturing scheme. Eventually (I assume) the idea is to take over everything but as a way to enter the field without all the risk, it seems a viable way to do it.

I am sure that Apple could have sold a bunch of cars just by slapping an Apple badge on them. I’m not saying that would have been a good idea if the car couldn’t maintain the Apple-ness quotient of the brand, but I would still have liked to see what they came up with (the toaster and no-steering-wheel options aside.) Indeed, a competently manufactured vehicle with some engineering prowess behind it could still make for a compelling entry. All of the marketing would be pre-sold.


I can at least see the argument. I believe the intent of the Roadster is to produce an electric car that has better performance than any ICE car past, present, or that can be made in the future. It is to demonstrate unequivocally that EVs are a better technology than ICEs. One could argue that this is an important psychological step for full scale adoption of BEVs.

Xiaomi making cars makes a whole lot of sense if one believes the future of cars is as a software-designed computer on wheels. BYD had the early advantage in China because it is a battery maker, and the battery is what defined EVs initially. The future of EVs will be the software. That will be how cars will be differentiated and that is Xiaomi’s advantage over BYD.

That is also why VW and the rest of the OEMs are in trouble. The few current EV makers who have mastered the software are years ahead of the OEMs. https://europe.autonews.com/automakers/automakers-struggle-create-software-defined-vehicles


As far as Ford and VW, in particular, are concerned, the advantage of having everything software dependent, enables their subscription model, when they can keep dinging owners for fees, for the life of the car, or else owners suffer the loss of functionality of the car, because big brother turns the feature off.


I think it probably was. Martin Eberhard wanted a high performing sports car, but with practical range. To get both it had to be a lightweight two-seater.

The Roadster made EVs sexy instead of stogy. The Model 3 is hip only because the Roadster was hip.*

*For you youngsters, hip is what the older generation call “cool.” Which is what the younger generation call “phat.”

1 Like

And when was ‘the bees knees’?


1 Like

As someone who enjoys driving just for driving; I certainly hope not.



I think perhaps it was. They needed a very low volume car to prove (to themselves and others) that using hundreds/thousands of small tubular rechargeable batteries all soldered together could even work in the first place. People (including me and including most engineers I discussed it with at the time) thought they were absolutely nuts to tie together so many small batteries to power a car. I thought for sure they would have to create a large battery pack, or a group of large battery packs with a completely new form factor.

The model 3 isn’t really hip anymore, it is “mid” (I think that’s what the kids today use to say “generic”). I own one and I love it, but it isn’t “special” anymore, it’s just a generic 4-door sedan.

The primary, indeed the only function of a car is transport. The Wheels are the important part, not the electronics. All the rest is just making the wheels go where you want them. Nobody is ever going to buy a car that doesn’t take them someplace.

“The feature”? Heck, they can turn the whole car off, by remote control, and you have no remedy. I can’t wait until there’s a police chase and they just call up the company and say “Turn off the xyz car” right now - and it happens.

Well you are not alone and I apparently am. I’m still not so sure. Lots of inventions or improvements went straight “from farm to table”, so to speak, without a middle step, but I’m not going to argue the point since it’s not that important in the story anyway.

1 Like

Available now, on a Subaru "“stolen vehicle immobilizer”. Part of Sube’s “Starlink Security Plus” package. Buy a new Sube, and you get a free trial, for six months, after you give them your credit card information. After six months, for the next thirty months, it costs $4.95/month, plus tax. If you want it after the first three years, then the price escalates to $14.90/month, or $149.90/year. I doubt those prices are good for the life of the car. As I have seen with anti-virus update subscriptions, prices escalate whenever the company wants to make more profit.



This has been well documented many times by Musk in the Tesla masterplan, IIRC.
They needed a proof of concept car that was very expensive (high margins) so that they could afford the cost of developing the Model S and a factory to make it in.
They didn’t design the whole roadster, it was made on a Lotus chassis and basically hand assembled


1 Like

Absolutely correct! I’m surprised so many have bought into the whole “computer on wheels” thing.

The only function of a camera is to take pictures. Yet people have overwhelmingly switched to digital cameras over analog. Smart cameras are preferred by consumers over dumb ones. The same is happening to cars.

A software-designed vehicle means designing the hardware so that all components can be controlled by a single software. This means more efficient performance. It also means being able to improve that performance simply by improving the software, which can then be sent to car owners over the air. Going digital also makes car production more efficient, shortening the time for new model development and, as Tesla is demonstrating, allowing modifications to production to be applied very quickly. And even if you believe self-driving will never happen, the improvements in driver-assist software is making human driving much safer.

In short, software-designed cars do transportation better than analog cars. A MacPro computes better than an abacus. An iphone communicates better than a rotary dial.

It’s progress.

1 Like

Thank you, Goofy! Of course not all panned out. One I was very much against was bailing out SolarCity. In the old TMF there are several posts of mine saying why it had a terribly bad business model.

Here we disagree. Heuristics never was and never will be a solution to AI. Some problems are just too complicated for programmers to think of and to resolve every potential case. This statement is based on years of experience writing code. All of FSD up to and including v 11 were heuristics based. Test drivers kept saying that FSD drove in strange ways, unfamiliar to humans. The most important comment about FSD v12 is how human like it behaves. That is a leap across the Grand Canyon! One of the criticism of neural networks is that we cannot trace the logic behind the results. Of course not because the process is not boolean, not rational. What neural networks do is pattern matching, if this pattern then do that on a statistical basis, it tends to be the best outcome most of the time. Expect perfection and you’ll fail, perfection is not possible, 99.999% is possible.

Someone asked, “When will FSD be done?” I replied, “Never! Learning never stops.” The leap from boolean heuristics to neural networks AI is a conceptual chasm that will be very difficult for many people to manage. Paradigm shift on steroids.

When Martin Eberhard wanted a sexy EV there were lots of EVs in use…

on golf courses!

The Captain


After about 2 days of using it, I have to agree. It (V12) just feels so much more human than V11 did. As a trivial example, V11 didn’t slow down for the speed bumps and V12 does. But slowing down for speed bumps is critical for a proper human experience (and perhaps to protect the car somewhat). Also, the general flow if its driving is so much smoother, and more human-like with V12. Is it perfect? No, not yet. But it’s way better than the version I tried a few months ago (and described here somewhere).