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

[Sean Wagner]

Concerning German numbers: it requires a lot of effort, great persistence, and also attentiveness to sell meaningful numbers in many countries with a domestic industry. Lexus fell flat on its grill in Europe. Have no doubt there are legions of prospective EV buyers there just waiting to push the button on a VW [which is a good thing].

A towing hook is coming... now, for that wagon.

 

[Fact Checking]

Wow:

Vincent on Twitter

"1/ On April 2, Wu Qing, Deputy Mayor of Shanghai, met with Jerome Guillen, Tesla President of Automotive & the team. Production equipment are expected to install in May, & some production lines will be mass-produced at the end of 2019."

"2/ Both sides conducted in-depth exchanges on the progress of the Tesla Shanghai GF3 project and related topics such as supply chain localization and future development planning in manufacturing and parts supply."​

 

[Pras]

Concerning German numbers: it requires a lot of effort, great persistence, and also attentiveness to sell meaningful numbers in many countries with a domestic industry. Lexus fell flat on its grill in Europe. Have no doubt there are legions of prospective EV buyers there just waiting to push the button on a VW [which is a good thing].

A towing hook is coming... now, for that wagon.

But the German cars sell so well in America. I hope Germans return the favor

 

[Zaxxon]

We are all doing this to ourselves what with all the premature discussion on when the report will come out and trying to read between the tealeaves what it means if it comes out yesterday or today or tomorrow.

Please. The report is not delayed, it's just not dropping at the time most expected by the hive collective here. That's fine. Come back with delays if the report isn't there by next Monday at the earliest. Anything else is self-inflicted agonising over nothing.

Agree that it's not delayed, but Monday would definitely be. Wasn't their original timing promise when they started doing the releases that they'd announce by the end of the third day following quarter-end?

Of course they can always change this practice, but Monday would definitely be a departure from their currently-defined practice (ie a delay).

I expect it sometime today.

 

[Sean Wagner]

I hope Germans return the favor

I don't want to drag the discussion out here, but building a product that ticks all the important boxes works better. And servicing it, and projecting a certain dedication to the task. Also, zero jokes. Oops. I'll never make it on Zwitscher.

 

[SebastianR]

That‘s underwhelming... Having some crow for lunch ...

First: these are still great numbers. For Germany after all!

Second, it is exactly what I cautioned against.

Third, the list is not soo bad: Tesla beat Honda, Jeep, Landrover and came very close to Porsche (2723 vehicles)

 

[mrdoubleb]

I think it's important to keep in mind that the people who received a Model 3 in Q1 in Europe are all reservation holders.

Maybe, maybe not. As far as I recall they`ve opened up the configurator for everyone on January 4th. So technically we can have non reservation holders for deliveries in Q1. I bet many res holder are waiting for RWD LR and SR/SR+.

 

[Singuy]

Demonstration also can diminish ability of a fair amount of reviewers to talk down AP as ~”basically just adaptive cruise control with lane keep assist features we’ve been seeing on cars for some time now,”

and

that “study” that claims Tesla is something like 22nd out of 22 self-driving programs... some brief video excerpts of the day could make the silliness of that “study” dramatically easier to spot for a casual observer.

Let's not get our hopes really high up. Elon said recently tweeted that HW3 actually performance worst than HW2.5 due to lack of optimization for the hardware.

 

[Chocochip]

Maybe, maybe not. As far as I recall they`ve opened up the configurator for everyone on January 4th. So technically we can have non reservation holders for deliveries in Q1. I bet many res holder are waiting for RWD LR and SR/SR+.

Fair enough, but my point was that those who received their cars in Q1 might have, at most, seen a US version of the Model 3 in a Tesla store and sat in it, but definitely none of them got to go for a test drive.

This kinda takes us back to the whole "are test drives necessary to pull the trigger" discussion, but one must consider how much more confidence-inspiring is to see actual owners in your neighbourhood as opposed to just videos of reviews online for the average person, and therefore how vast the potential demand is when delivery records are beaten with just early-adopter, expensive version deliveries. Tesla will NOT have a demand problem for the Model 3 for at least 2 years, no matter how fast they ramp up production!

 

[Fact Checking]

Let's not get our hopes really high up. Elon said recently tweeted that HW3 actually performance worst than HW2.5 due to lack of optimization for the hardware.

Those FSD computer comments by Elon have been widely misinterpreted.

What he said, exactly, was:

Elon Musk on Twitter

"Retrofits will start when our software is able to take meaningful advantage of the Tesla FSD computer, which is an order of magnitude more capable. For now, it’s slightly disadvantageous to have Tesla FSD computer, as our software is more refined for HW2."​

My guess:

Right now the FSD computer is probably only ~4% utilized and power draw is probably slightly higher than that of HW 2.5. Maybe calculation performance of the existing networks is slightly slower as well.

The HW 2.5 GPU is maybe 90% utilized during peak load.

Once they have the new, much larger networks, the FSD computer will probably be ~80% utilized during peak loads. The HW 2.5 simply won't be able to run those networks at all: it requires ~1800% of the current computing capacity, which is simply not there. Performance would drop from the required 100 fps processing to maybe 5 fps processing speed.

Elon is completely right that networks that fit on HW 2.5 can be run on HW 3 as well but it's "slightly" slower or slightly less power efficient.

But with the new networks there will be no contest.

 

[Carl Raymond]

I think it's important to keep in mind that the people who received a Model 3 in Q1 in Europe are all reservation holders, and they all went for the higher-end, expensive versions. This says almost nothing about the organic demand of the Model 3 in Europe.

People who are not obsessively following Tesla online (and can afford a Model 3) are just now starting to see the cars on the road, or in their neighbour's driveways, or in adjacent parking spots. There will be a lot of "What's this car?", "Is this that new electric one I heard about?" "So, they're actually making these now, and they're selling them here?" "Weren't they bankwupt or something?", and most importantly "How does it drive?", "I should go for a test drive".

I expect the demand to keep increasing throughout this year as more Model 3s become visible in Europe and as more (affordable) versions become available. Call it the Norway phenomenon, where it suddenly dawns on the greater population that EVs are actually practical, and can be charged in many places, conveniently, and they're super-fun to drive.

And then Bjørn can make more money from views on his channel, and he won't have to sell his underwear to keep wifey happy.

Excellent comment. Years ago I read a book “the Tipping Point” by Malcolm Gladwell.

Every day there are predictions on this forum about such and such *event* finally spurring the share price to its rightful place in the stratosphere. They may be right, and no doubt when it does happen people will back track to the immediately prior catalyst, point their finger and exclaim “that, was it”.

More likely, what we have are grains of sand falling on the high end of a set of scales. Each car produced is a grain. Each friend of the purchaser is another grain. Each tourist to Norway is a grain.

Each day the sand flows faster. The weight of FUD on the other end is considerable. But she will tip. Slowly at first. Then with conviction. People will say electric cars were obvious from the beginning. We will be too polite to ask how much stock they held.

 

[DriveMe]

Let's not get our hopes really high up. Elon said recently tweeted that HW3 actually performance worst than HW2.5 due to lack of optimization for the hardware.

For features (current features) that do not take advantage of the new AI chip, there is no performance improvement. However, for FSD features that will take advantage of it, the performance improvement will be dramatic.
Here is the exact quote:

Retrofits will start when our software is able to take meaningful advantage of the Tesla FSD computer, which is an order of magnitude more capable. For now, it’s slightly disadvantageous to have Tesla FSD computer, as our software is more refined for HW2.

Edit: Fact Checking is much faster in typing!

 

[KarenRei]

Those FSD computer comments by Elon have been widely misinterpreted.

What he said, exactly, was:

Elon Musk on Twitter

"Retrofits will start when our software is able to take meaningful advantage of the Tesla FSD computer, which is an order of magnitude more capable. For now, it’s slightly disadvantageous to have Tesla FSD computer, as our software is more refined for HW2."​

My guess:

Right now the FSD computer is probably only ~4% utilized and power draw is probably slightly higher than that of HW 2.5. Maybe calculation performance of the existing networks is slightly slower as well.

The HW 2.5 GPU is maybe 90% utilized during peak load.

Once they have the new, much larger networks, the FSD computer will probably be ~80% utilized during peak loads. The HW 2.5 simply won't be able to run those networks at all: it requires ~1800% of the current computing capacity, which is simply not there. Performance would drop from the required 100 fps processing to maybe 5 fps processing speed.

Elon is completely right that networks that fit on HW 2.5 can be run on HW 3 as well but it's "slightly" slower or slightly less power efficient.

But with the new networks there will be no contest.

IMHO, it's probably a threading issue. A program designed for a low number of threads will generally run faster on a processor with a low number of cores, because those cores are generally faster than the cores you find on many-core processors. But if the program is designed for many threads, it'll run vastly faster on many-core hardware.

Neural networks are the ultimate threading task. Our brains have nearly a hundred billion "cores" (neurons)***, but their individual performance is rather pathetic.... with "clockspeeds" measured in dozens of hertz


*** Well... okay, not really. A) Not all neurons contribute to "computation" - for example, white matter is basically a data bus; B) not all "computation" is dedicated to aspects of thought processing; C) many neurons appear to be redundant, in case of damage; and D) the cortical minicolumn appears to be the basic element of "computation", not the individual neurons that make it up; the ultimate neural net chip would emulate whole minicolumns at a hardware level per core, rather than the individual neurons that make them up. We're really more like a 200-million core processor.

 

[Fact Checking]

BTW., the next ~60 minutes would be the rough historic time frame where Tesla filed their "Q1 Production & Deliveries" report.

If they don't file today then we are in uncharted territory.

 

[jbih]

The March numbers for Germany are disappointing, but note that Tesla gets 2000 Euros less per car in Germany - the German EV incentive forces every manufacturer to reduce the price by 2000 Euros per car. (The total incentive is 4000 Euros, 50% from the government and 50% from the manufacturer.) Considering this, Tesla should ship as many Model 3s to other European countries as possible.

 

[Fact Checking]

IMHO, it's probably a threading issue. A program designed for a low number of threads will generally run faster on a processor with a low number of cores, because those cores are generally faster than the cores you find on many-core processors. But if the program is designed for many threads, it'll run vastly faster on many-core hardware.

Neural networks are the ultimate threading task.

Correct, that's my take as well: past leaks of the FSD chip talked about 400 MHz frequencies. The Nvidia GP106 chip in HW 2.5 is running in excess of 1 GHz and possibly has higher "single threaded" performance. [*]

But the capacity of the new FSD computer is probably 10x-20x of the Nvidia chip, so it can run much, much larger networks without running out of parallel computation capacity.

[*] Note that vector CPUs don't really have single threaded mode of operation: all their instructions operate on a large number of data operands at once. But still here's an effective single-threaded throughput figure behind it all, roughly defined by the clock rate of the chip, its style of pipeline and the calculation latency of the functional units.

 

[Tslynk67]

The March numbers for Germany are disappointing, but note that Tesla gets 2000 Euros less per car in Germany - the German EV incentive forces every manufacturer to reduce the price by 2000 Euros per car. (The total incentive is 4000 Euros, 50% from the government and 50% from the manufacturer.) Considering this, Tesla should ship as many Model 3s to other European countries as possible.

Good point. They should also delay M3SR until they have healthier margins too (or out the official price up by 2k in the country)

 

[j6Lpi429@3j]

Neural networks are the ultimate threading task. Our brains have nearly a hundred billion "cores" (neurons)***, but their individual performance is rather pathetic.... with "clockspeeds" measured in dozens of hertz

Ha. I'm literally reading this on one monitor whilst debugging some stupidly complex code on the other monitor that ONLY uses eight threads. The thought of more gives me a headache

 

[Chocochip]

IMHO, it's probably a threading issue. A program designed for a low number of threads will generally run faster on a processor with a low number of cores, because those cores are generally faster than the cores you find on many-core processors. But if the program is designed for many threads, it'll run vastly faster on many-core hardware.

Neural networks are the ultimate threading task. Our brains have nearly a hundred billion "cores" (neurons)***, but their individual performance is rather pathetic.... with "clockspeeds" measured in dozens of hertz


*** Well... okay, not really. A) Not all neurons contribute to "computation" - for example, white matter is basically a data bus; B) not all "computation" is dedicated to aspects of thought processing; C) many neurons appear to be redundant, in case of damage; and D) the cortical minicolumn appears to be the basic element of "computation", not the individual neurons that make it up; the ultimate neural net chip would emulate whole minicolumns at a hardware level per core, rather than the individual neurons that make them up. We're really more like a 200-million core processor.

Right, I've concluded you simply know too much, or you access information too fast, to not be an AI.

Also, were you an actual human (which clearly you are not), I would develop a terrible inferiority complex (which is unnecessary, as you are in fact an AI).

 

[Sean Wagner]

I expect the demand to keep increasing throughout this year as more Model 3s become visible in Europe and as more (affordable) versions become available.

And that should also help sell more S & Xs, although they are purdy wide machines for the local scene. I did manage to take a Cadillac DeVille through a Swiss drive-thru once [wasn't mine either], and vowed never ever to repeat that mistake.

Edit: Swiss March numbers for Tesla are great.
Tesla 4.3% total market share // 1'243 up from 322 in March 2018
BMW 6.5% 1'881 // 2'037 -7.7
Volkswagen 9.8% [top seller], Skoda another 8.4% for VAG, Seat another 4.8%, Audi another 7.5%, plus Porsche 0.8%

Source: http://e3.marco.ch/publish/autoschweiz/941_892/2019-04-02_Auto-Markt_nach_Marken_Maerz_2019.pdf