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

[Runarbt]

Like charming a snake with a flute:

I for one, am a bit sad it seem like this snake wont make production.

Someone have argued its too complex.. or might we see this one on SuperChargers eventually? Would love to have my Cybertruck charged by a Cybersnake.

 

[MikeAtkinson]

Also there is something fishy about the reported new chip order from TSMC.
It has been widely reported as HW4, but it might not be so, it might actually be the Dojo chip we are speculating.

The news article Ferragu replied to mentioned they use some kind of new “system on waffle” technique to directly build the whole system on waffle without base plate or PCB, then glue it to heat sinks.
A 12 inch waffle only yields 25 systems, and its 7nm.
That’s far too big of an area for next gen FSD computer. Also the order is for only 2000 waffles, so only 50k systems tops.

As some comments under Ferragu’s tweet did point it out.

From what we heard(maybe not supposed to though) from autonomous day, HW4 would be more about iterating HW3 and make it cheaper and faster, I would expect lower costs and energy consumption, also performance improvements within 2x-10x, not the kind of big revolution if the new chip turns out to be HW4.

I don’t have enough knowledge about chips to say for sure, could someone who knows these staff comment on this?

(This is not OT, since implications of HW4 could have Osborne effects closer to its rumored release date. Also, I think Dojo is a much bigger deal for the future of the company than most people realized)

Agree there is something fishy about the reporting. It seems most likely to me that China Times has obtained information from several sources and conflated them together. It is possible that Tesla actually have three chip designs in progress with TSMC, Dojo, HW4 and a Broadcom/Tesla joint project.

The stated number of wafers is also strange, 2000 is too many to be samples (unless the expected yield is very, very low), but it is less than I would expect for a full Dojo system. Full production in Q4 next year is later than Elon's stated timeline for Dojo and it is a much bigger gap than I would expect between initial production and full production, five quarters is an age in semiconductors, and even longer for Tesla.

If HW4 does turn out to be using this "chip-on-substrate chip-on-wafer packaging" 7nm TSMC technology then it would be much smaller (10x or more) and faster (10x or more), use less power, but probably be more expensive (at least initially).

Also agree that Dojo is a much bigger deal than most people realise. In Ark Invest's latest Big Ideas report they are very bullish over Deep Learning, saying that "adding $30 trillion to global equity markets over the next two decades." If Tesla could capture 20% of that, it is of a similar order to self driving. [note: the $30 T figure is not well explained, it might already include self driving].

 

[Oveeus]

There you go. Audi needs to react to Tesla to remain competitive.

Currently, Audi sells a compact electric CUV for what Tesla sells a midsize electric CUV.

Not because the cars are so horrible nobody wants them.

Of course it's a bit of both.

But the reasoning remains the same. The competition can choose to compete in price, which it can't do for long because Tesla has the absolute advantage in battery cost.

Alternatively, it can compete by putting more money into R&D for EV to stay competitive with Tesla.

And Tesla can't do the volume alone. It has the first mover advantage, but ultimately, it needs the entire market shifting toward EV to achieve the economy of scale.

When the whole car manufacturing is EV-oriented, everything Tesla needs, that don't see the need enough to vertically integrate, would become cheaper because it's no longer a special part for Tesla's EV, but for all EVs out there.

As a shareholder, yes, we want Tesla to dominate and ideally be the absolute dominant player.

But really, I'd be satisfied for Tesla to become the leader in EV and Energy as Apple is in the Smartphone industry. You can never do enough to satisfy the entire market. So, Tesla needs its competition to grow. They will fill those markets that Tesla can't or don't want to cater. And only by then, Tesla can achieve its mission. It's never about Tesla becoming a monopoly... it's about shifting the world to a sustainable energy future.

 

[Right_Said_Fred]

Keeping up with this thread is hard enough. Two pages of facts, opinions and anecdotes about immigration doesn't help. So they have been removed. Please think twice before posting or responding to (way) off topic stuff.

 

[UkNorthampton]

This is definitely useful to track, but the ex-Audi executive in one of @avoigt 's recent interviews blamed the lack of e-tron sales on the supply chain, which I don't buy at all. I think they had plenty more capacity in the supply chain, but almost nobody wants that car. Otherwise why has it been discounted so much...

Where has it been discounted, USA, EU or everywhere? I don't think an Etron is compelling/ompetitive in USA, maybe more in EU (import taxes, range less important, brand loyalty, company car choice restrictions). Import tax works against Audi in USA and against Tesla in EU.

Discounts:-
Possibly idiotic Audi (US?) managers who don't understand the difficulties of making the cars and are wrongly incentivised with monthly or quarterly targets.

Possibly dealership chains trying to send a message? Possibly a USA/North America phenomenon?

Stock:-
For UK - I looked at one of the Audi chains and their current stock.

Model Number Link
Etron - 14 - New Audi e-tron cars in stock - Listers
Q5 - 39 - New Audi Q5 cars in stock - Listers
Q7 - 41 - New Audi Q7 cars in stock - Listers
A4 - 63 - New Audi A4 cars in stock - Listers
A8 - 11 - New Audi A8 cars in stock - Listers

I don't know Audis - I just get to experience them being driven badly near me. Presumably Q7 and Etron have similar appeal, but A8 not so much at the moment, or it's always discounted.

Where discounted Audis go in UK? New Audi Deals | Audi Finance Offers | carwow

I'm not sure what the below is saying, but I might revisit this in a few months and compare.

vs

and

 

[kanweg]

I’d expect the SP announced tomorrow. Next week the trade of chairs is a bit hampered by the bill that is attached to each chair sold. That is under the assumption that the index funds have no problems dealing with the bureaucracy involved and retailers have (If they buy next week). So, it would be a buyers week for the index funds and everyone can sell to them without bureaucratic hassle.

But then, the retailers may well be unaware of the bill if they buy. And it may be an administrative burden to the index funds.

As to the effect of the stock split and the rise we have observed, I think that retailers may be reluctant to buy expensive chairs for the following reason as well: If your fridge breaks down, a large sum is locked up. If the chairs were cheaper, you’d sell only a few. With an expensive chair you have to sell one and are left with cash that is no longer in the stock you wanted to invest in.

As this form of illiquidity is resolved in sweat, even those that balk at a high SP for that reason would start buying, as the disadvantage is temporary; resolved in a couple of weeks. So, this dampens the effect of a possible stock rise after 31 August.

 

[jerry33]

Gov. Newsom said the critical hours to conserve electricity are between 2pm and 9 pm. I understand the air conditioner peak use thing but am shocked that California doesn't have enough enough solar panels to cover A/C use (at least between 2pm and 6pm.) during mostly sunny summer days. This suggests the electrical utilities haven't installed enough solar to help keep the rates low for rate-payers. And rooftop solar should be helping out a lot here but maybe not enough homes have it.

I believe you've answered your own question.

 

[jerry33]

If the Tesla chip is faster, uses less electricity and can run the same native code as Tesla cars, they’re going to use it. If Nvidia was so great, why are they waiting to build Dojo? If Dojo comes after the first v4 chips, it seems rational the v4 chip will be used for Dojo.

My understanding is that V4 is a special purpose chip. It's only faster for that purpose. Special purpose chips can be many times slower when used for processes other than what they were designed for.

 

[avoigt]

This is definitely useful to track, but the ex-Audi executive in one of @avoigt 's recent interviews blamed the lack of e-tron sales on the supply chain, which I don't buy at all. I think they had plenty more capacity in the supply chain, but almost nobody wants that car. Otherwise why has it been discounted so much...

To add to that a recent study shows that besides Porsche (average selling price about € 100k) and Tesla no other manufacturers makes any profit per unit sold. BMW did not want to talk about their i3 profit or loss at the ER call and Audi does not do it too.

The heavily reduced prices for the e-tron should be seen in the context of carbon credits as well as the strong marketing push from the VW Group to get BEV market share and try to create the image they are a strong BEV player right after Tesla.

 

[jerry33]

How/Where to find such advisers?
Does calling Brokerages like ETrade, Fidelity help get you to one of these advisers?

The less you know about something, the more someone from out of town with a laptop appears to be a knowledgable expert.

 

[jerry33]

Why does Tesla keep cutting prices?

The lower the prices, the wider the moat.

 

[vikings123]

Do you have any data supporting that S&P500 inclusion is not allowed before a split?

Asking, as I am doing some option plays to try to have some leverage when S&P500 is announced. If I can delay and save some time-cost, that would be great. ;-)

I think in general people on the twitter verse believe that S&P would not want to deal with the mechanics of a split and make it even more complicated. So expectation is they announce it after the split date Aug 31st.

Be careful with Theta because as an option buyer time is your biggest enemy. Most are expecting flat trading next week but tbh I cannot think of many Mondays and Tuesdays where Tesla does not spike.

 

[ggies07]

I liked this. A lot.
And Search tells me it hasn't been shared here yet.

Shall we dance?

Beautiful.

 

[CLK350]

Since there are no statutory legal rules about S&P inclusion I can very well see the S&P admins deciding to pull a fast one, for example announcing a delay because (FCA credits, C19, split decision of the board, .. whatever). That would sink the SP (temporarily of course), enough to let the MM profit from this, and the index funds or whomever load up cheap. Would not be the first nor the last time this kind of shenanigan has happened. Remember that one page letter from Paulson requesting a Trillion $ bailout? no strings attached, and they continue to gamble illegally (see Wall Street Banks Are Dangerously Evading U.S. Derivatives Rules by Making Trades at Foreign Subsidiaries and that site in general for a running litany of such muffled scandals).
It's surprising the US hasn't gone completely bananas (uh Banana Republic.. ), but we are getting there slowly but surely.
PS. Ergo I sold 10% of my TSLA yesterday above 1900, might regret not having sold more ... but well c la vie ;D

 

[Unpilot]

This seems positive and might help a bit with the macros...

And whatever happened to that stimulus bill they were discussing a few weeks back? Thought that was "urgent"?

U.S. and China to hold trade talks in the coming days, Chinese commerce ministry says

View attachment 578532

I'm no diplomat....but with the election coming up...any leverage the orange man had is about gone.
He needs a win and they don't have to give anything.

Delay and wait would be the most likely Chinese strategy.

 

[Alphacrux]

Also there is something fishy about the reported new chip order from TSMC.
It has been widely reported as HW4, but it might not be so, it might actually be the Dojo chip we are speculating.

The news article Ferragu replied to mentioned they use some kind of new “system on wafer” technique to directly build the whole system on wafers without base plate or PCB, then glue it to heat sinks.
A 12 inch wafers only yields 25 systems, and its 7nm.
That’s far too big of an area for next gen FSD computer. Also the order is for only 2000 waffles, so only 50k systems tops.
The news clearly referred to the chip as an HPC chip, not sure why people insist it would be the FSD computer for cars but not for backend data centers.
It also mentioned the chip was jointly designed by Broadcom and Tesla, to me that immediately sounds like the high speed switch fabric you would need for parallel training. I don’t remember HW3 was jointly developed with them, anyone heard anything different?

From what we heard(maybe not supposed to though) from autonomous day, HW4 would be more about iterating HW3 and make it cheaper and faster, I would expect lower costs and energy consumption, also performance improvements within 2x-10x, not the kind of big revolution if the new chip turns out to be HW4.

I don’t have enough knowledge about chips to say for sure, could someone who knows these staff comment on this?

(This is not OT, since implications of HW4 could have Osborne effects closer to its rumored release date. Also, I think Dojo is a much bigger deal for the future of the company than most people realized)

342 Transistors for Every Person In the World: Cerebras 2nd Gen Wafer Scale Engine Teased

 

[FrankSG]

This quarter too, I compiled a list of the top TSLA shareholders:

TSLA Holders Q2'20 (Top 69)

 

[Artful Dodger]

08:00 a.m. Whistle: Thu, 20 Aug 2020

• TSLA share price: $1872.29 -6.24 -0.33%
• NASDAQ Pre-market Volume: 64,990 "Light"
• NASDAQ-100 Futures: -9.62 -0.08% 08:00:00 ET
• Pre-Market High: $1,883.94 (06:28:23 AM)
• Pre-Market Low: $1,860.00 (04:21:47 AM)

Comment: "Light volume in the Pre-Market w. macros down slightly"

Cheers!

 

[ZeApelido]

I am a bit skeptical, but then it is a lot about semantics. Elon said Dojo was FPGA on his tweet.

Dojo will have two components that both are very important.

1. Generate a huge well annotated dataset
2. Train a neural network on the annotated dataset

1. Will require vasts amount of video processing and use neural networks for inference. The advantage of the 4D labelling is that you can use future data to label current timestep, for example use the side camera 4s into the future to label where the lane marking 100m in front of the car is in 3D space. You can also predict where an occluded car is based on where the radar said it was 1s ago and where it will be in 1s according to the cameras. Here some form of Kalman filter will be useful in combination with the neural network predictions. Also some for of Human labelling and likely some form of AI to predict how a human would label to help the labellers. I think designing this system is the 140+ IQ problem that I look forward to hearing the answer to in a few years and this is where the competition will struggle because there are not enough Karpathys and machine learning jedi engineers out there...

Thus 1 will require a mix of compute, a lot will be very GPU friendly and some will be pure HW3-like inference friendly, but on the server side and no need for real time batch size of one. The good news is that this is very parallelizable and it would be pretty easy to just let AWS do it for you, but maybe they would save money by using their own neural engine chips for it at some point.

2. Will require very large memory or some form of distribution, see for example this video at 6.14:

Imo it seems likely that it somewhere here is where the need for FGPA and extreme bandwidth comes in.

Imo Dojo is all about reducing the cost to run a build of any new code/dataset changes. Maybe they will reduce build time a bit also, but that could probably also have been achieved by throwing more money into the cloud.

I like the last video by GeoHotz where he says that using the cloud is a crutch, it’s better to build your own system as you learn to be more efficient:
Around 1.30.00 into the video he talks about their cluster

(Fwiw it seems Elon has seen at least parts of this video as he commented on a clip of it and he previously offered Hotz millions to work for Tesla).

Interesting take. So I think you agree that their current labeling strategy would not scale as they massively increase the data input? But instead of self-supervised learning off of driver actions, you think they will come up with some intelligent predictive labeler to help speed up / augment the human labeler's productivity?

 

[mongo]

If the Tesla chip is faster, uses less electricity and can run the same native code as Tesla cars, they’re going to use it. If Nvidia was so great, why are they waiting to build Dojo? If Dojo comes after the first v4 chips, it seems rational the v4 chip will be used for Dojo.

This chip cannot be both for production vehicle and training use. Thise are completely different roles with completely different requirements. The training side is like the YouTube/Netflix encoder farm that does the heavy lifting and compresses videos into dense packages. The vehicle side is like an MP3 player or BluRay machine. A simple, low cost, decoder. The HW4 could be part of Dojo for validation, but a Dojo chip woukd be too much (processing, power, and cost) for a consumer vehicle.

That’s far too big of an area for next gen FSD computer. Also the order is for only 2000 waffles, so only 50k systems tops.
The news clearly referred to the chip as an HPC chip, not sure why people insist it would be the FSD computer for cars but not for backend data centers.

It could be for either, but it cannot be for both (while still being an effective vehicle choice).

I don’t think that is the case, if it’s just NV GPUs, why would they simulate it using FPGAs?
I am guess he is referring to the current system which would be replaced by Dojo 1.0 as Dojo 0.1?

Option 1: the Nvidia chips are not available, but the logical design (emulated by FPGA) is.
Option 2: there is more to the system than the GPU chips (memory, physical unit interlinks, higher level proccessing/ control)