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

[Singuy]

BI is now a paid Tesla Shill..lol

MSN

 

[TheTalkingMule]

BI is now a paid Tesla Shill..lol

There's been a clear as day shift in the paid "journalism" world toward positive TSLA pieces. Hedgies(and MM's working the options market?) have switched directions.

It started a couple months ago, but became super obvious with that Lolora Kololodny info-centric piece on CNBC. That was quickly reshuffled to the appendix, but the purchased ad space was already obvious.

 

[Artful Dodger]

For clarification - TX is NOT building 2170 cells. The are building PACKS from 2170 cells that they bring in from GF Nevada.

Do you have a reference stating those 2170 packs are being built in Austin, and not imported from Reno with 2170 cells already installed? Could well be the case, as the math below shows:

Leaked Tesla comments from On Sep 8, 2022 revealed bty pack output from Giga Nevada. It seems they are making enough packs (likely all LR) to provide for both Fremont and Austin MY production.

Tesla Giga Nevada exceeds 6,500 Powerwall per week

Site Leader Eric Montgomery, who noted during the meeting that August 2022 was Giga Nevada’s second-best month of production, coming second only to October 2021. Montgomery also noted that Giga Nevada has to achieve a steady output of 8,800 high voltage battery packs per week to support the company’s aggressive vehicle production plans.

Keep in mind that we heard from public documents, via @carsonight on twitter, that Tesla Fremont is also importing 1,000 LFP packs per week from China (those go to the Fremont Model 3 SR+). And we know Fremont is manufacturing 12K cars/wk currently per the article above, which includes a max of 2K/wk Models S/X.

So 3K/wk S/X/LFP3 leaves ~9K/wk more packs needed at Fremont. And per Eric Montgomery (above), Giga Nevada is building 8.8K packs/wk. So if Austin need another 1K packs/wk, they could well be assembled in Austin. To me, this seems like more effort than just increasing production at Giga Nevada, and also a waste of resources if the near-term plan (or even the mid-term plan) is to switch to 4680 cells.

Cheers!

 

[SmokyPeat]

Do you have a reference stating those 2170 packs are being built in Austin, and not imported from Reno with 2170 cells already installed? Could well be the case, as the math below shows:

Leaked Tesla comments from On Sep 8, 2022 revealed bty pack output from Giga Nevada. It seems they are making enough packs (likely all LR) to provide for both Fremont and Austin MY production.

Tesla Giga Nevada exceeds 6,500 Powerwall per week



Keep in mind that we heard from public documents, via @carsonight on twitter, that Tesla Fremont is also importing 1,000 LFP packs per week from China (those go to the Fremont Model 3 SR+). And we know Fremont is manufacturing 12K cars/wk currently per the article above, which includes a max of 2K/wk Models S/X.

So 3K/wk S/X/LFP3 leaves ~9K/wk more packs needed at Fremont. And per Eric Montgomery (above), Giga Nevada is building 8.8K packs/wk. So if Austin need another 1K packs/wk, they could well be assembled in Austin. To me, this seems like more effort than just increasing production at Giga Nevada, and also a waste of resources if the near-term plan (or even the mid-term plan) is to switch to 4680 cells.

Cheers!

Do we have any idea what the ratio of 4680 to 2170 Model Y cars currently being produced in Austin is?

 

[bkp_duke]

Do you have a reference stating those 2170 packs are being built in Austin, and not imported from Reno with 2170 cells already installed? Could well be the case, as the math below shows:

Leaked Tesla comments from On Sep 8, 2022 revealed bty pack output from Giga Nevada. It seems they are making enough packs (likely all LR) to provide for both Fremont and Austin MY production.

Tesla Giga Nevada exceeds 6,500 Powerwall per week



Keep in mind that we heard from public documents, via @carsonight on twitter, that Tesla Fremont is also importing 1,000 LFP packs per week from China (those go to the Fremont Model 3 SR+). And we know Fremont is manufacturing 12K cars/wk currently per the article above, which includes a max of 2K/wk Models S/X.

So 3K/wk S/X/LFP3 leaves ~9K/wk more packs needed at Fremont. And per Eric Montgomery (above), Giga Nevada is building 8.8K packs/wk. So if Austin need another 1K packs/wk, they could well be assembled in Austin. To me, this seems like more effort than just increasing production at Giga Nevada, and also a waste of resources if the near-term plan (or even the mid-term plan) is to switch to 4680 cells.

Cheers!

Actually, you are correct. It's the full pack being shipped in via rail from GF NV.

Need more coffee, I'm too old for all-nighters.

 

[Artful Dodger]

Do we have any idea what the ratio of 4680 to 2170 Model Y cars currently being produced in Austin is?

Not from any 1st-hand source, but we can speculate based on last known Kato Rd. 4680 production rate. Also, 2170 equipment was delayed from China due to shipping issues in Shanghai, so that's relatively new. I'd also say its a fast moving target, so I just HODL...

We do have 2 more piece of new information. 1st, via Martin Viecha in his GoldmanSachs talk last week, Tesla finally has enough battery cells for both business (auto and energy). So we know that Tesla has their cell supply needs covered.

2nd, we also learned from the Tesla order page that Tesla is no longer taking new orders for LR/AWD Model 3. This implies that all Fremont Model 3s are now LFP/SR+ variants except for the highly profitable M3P LR/Dual-motor (these were about 10% of total Model 3 volume previously). So maybe 600/wk MP3 before, and maybe a bit more now since more people who wanted a 3LR now make their 2nd choice an upgrade to M3P rather than a downgrade to SR+ version. So maybe 1K/wk LR packs to Model 3 after? That's only if Tesla can get 5K LFP packs/wk for SR+ which is unverified.

While it may take some time to work through the backlog of 3LR orders (4-6 wks?), after those ~4K/wk LR packs could be freed up to go to Austin to support Model Y Production (very rough guess; large error bars). But none-the-less, Tesla is clearly prepared to continue the Model Y ramp in Austin w. or w/o 4680s. I think both will be made for quite a while.

Cheers!

 

[Discoducky]

BI is now a paid Tesla Shill..lol

MSN

HA, haven't heard that in like 10 years. Does this person know that every car maker has adopted this approach?

"But its oversized touchscreen and lack of physical buttons isn't for everyone."

Ever heard of voice controls?

"Basic functions like opening the glove box, changing the wiper speed, and directing the air vents require using the touchscreen..."

 

[WADan]

MotorTrend is talking about "Coding the Car" video without mentioning Tesla. LOL.
They even have a new award “ Software-Defined Vehicle Innovator Awards”.

They finally realize it. About 10 years late!

 

[unk45]

The former part of your post was mostly agreeable, but this part I have to disagree with.

Not because it wouldn't be super beneficial to the planet for Tesla to have great competitors, but because Tesla is not going to be pushed faster by any force other than the established mission.

Maybe I was not clear. I was not suggesting that Tesla could grow faster than it can. I was suggesting that the advance of the mission will accelerate only when major competitors emerge, that is to say that if they offer retreaded ICE as BEV they'll not be major competitors. When an otherwise attractive Mustang BEV arrives that has spaghetti connections and excessive complexity that really will not advance the cause.
When Geely keeps the best BEV for China only and uses a platform that accommodates ICE, hybrid and BEV that does not quite make it either. There are copious examples.

The mission only advances when competent and fit-for-purpose BEV arrive in great numbers and variety. They need to do better than Porsche, who eschew regenerative braking because "Porsche drivers want all the feel they have always had".

The advance will only grow faster when the competition develops genuine unapologetic BEV.

 

[jeewee3000]

I've been invested in Tesla since 2015 but never owned one. Today I went to a local showroom and ordered my first Tesla, to be paid for fully with TSLA gains. Part of it is thanks to the TMC community, which gave me the confidence to see through the FUD and to HODL.

Excited about my future blue Model Y! Should arrive by February.

 

[samn]

I've been invested in Tesla since 2015 but never owned one. Today I went to a local showroom and ordered my first Tesla, to be paid for fully with TSLA gains. Part of it is thanks to the TMC community, which gave me the confidence to see through the FUD and to HODL.

Excited about my future blue Model Y! Should arrive by February.

View attachment 853660

Congratulations! It's about time you got yourself a Tesla. Good choice with the blue Model Y as well. That's the first Tesla we bought also in blue. We took delivery August 2020 and now it has almost 50k miles on it. Wonderful car.

 

[wipster]

ah no. It will be like riding in a non-window seat in a plane.

Give them movies to watch or games to play... they'll be happy and at their destination before they know it!

 

[scaesare]

I wrote the specs for some of the code; how's that? And yes it was awhile ago, but they are still using it in some form or fashion as I can still gleen/groc the output behavior.

OK, cool.

So, out of curiosity, what about the output of a new feature like the creepwall indicates to you it's built from previous car data rather than a real-time inference? (not a challenge, honest question).

 

[BornToFly]

Video of Shorty Macshortface trying to get some of my shares in June....

https://twitter.com/x/status/1570919616469405696

 

[scaesare]

you would think cutting off the oxygen would be a better approach to electrical fires.. (i propose a big fireproof blanket)
maybe a chemical based suppressant needs to be developed to rapidly cool battery

The Model 3 packs are backfilled with Intumescent Goo which was supposed to help prevent such fires to begin with. (Or at least they were at one point in their design)

I assume that's why Model 3 fires are rare, given the run rate of the cars.

 

[Discoducky]

OK, cool.

So, out of curiosity, what about the output of a new feature like the creepwall indicates to you it's built from previous car data rather than a real-time inference? (not a challenge, honest question).

Sure, as you can see from the car's real-time vision, the lane lines are moving around constantly (they vibrate, wiggle, appear, disappear...etc), these are being reasoned through computer vision neural networks in non-deterministic pixel space (vs deterministic vector space). Pixel space is to non-specific to place a nice straight non-moving wall that is drawn to be the closest possible space to the lane, without being un-safe. If this was being done in pixel space, the car would most likely end up in the lane in some high percentage of cases or stop way behind the lane in a percentage of cases/scenarios. Only with ground truth (most likely mapped vector space) could that wall be drawn and trusted to such a high degree. And that is being done by reasoning through several ground truth markers in the scene and calculating angles and distances. Then, feeding that ground truth into the trained NN models allow them to be much more confident in their outputs, quicker and have more robust outputs. This also greatly benefits path planning which can take ground truth inputs and be able to plan paths that don't wiggle as the input is deterministic and non-changing as the car moves through the scene, thus turns that are somewhat occluded or blind to the real-time system are now much smoother. Case in point is when the car commits beyond the creep wall and moves to an actionable pose in the median box. This is now smooth as that median space is now trusted as ground truth (vector space to some degree) and the path planner does not need to re-plan near as much as it used to for that kind of challenging maneuver.

Does that help?

 

[MP3Mike]

The Model 3 packs are backfilled with Intumescent Goo which was supposed to help prevent such fires to begin with. (Or at least they were at one point in their design)

I assume that's why Model 3 fires are rare, given the run rate of the cars.

I thought Munro tested the "goo" in the Model 3/Y packs and determined that it was flammable. (What they tested from a different OEM was not as flammable.)

 

[Singuy]

Video of TSLAQ trying to get some of my shares in June....

https://twitter.com/x/status/1570919616469405696

Fitting video since that gun is not even real but a co2 powered airsoft gun...lol.

 

[scaesare]

Sure, as you can see from the car's real-time vision, the lane lines are moving around constantly (they vibrate, wiggle, appear, disappear...etc), these are being reasoned through computer vision neural networks in non-deterministic pixel space (vs deterministic vector space). Pixel space is to non-specific to place a nice straight non-moving wall that is drawn to be the closest possible space to the lane, without being un-safe. If this was being done in pixel space, the car would most likely end up in the lane in some high percentage of cases or stop way behind the lane in a percentage of cases/scenarios. Only with ground truth (most likely mapped vector space) could that wall be drawn and trusted to such a high degree. And that is being done by reasoning through several ground truth markers in the scene and calculating angles and distances. Then, feeding that ground truth into the trained NN models allow them to be much more confident in their outputs, quicker and have more robust outputs. This also greatly benefits path planning which can take ground truth inputs and be able to plan paths that don't wiggle as the input is deterministic and non-changing as the car moves through the scene, thus turns that are somewhat occluded or blind to the real-time system are now much smoother. Case in point is when the car commits beyond the creep wall and moves to an actionable pose in the median box. This is now smooth as that median space is now trusted as ground truth (vector space to some degree) and the path planner does not need to re-plan near as much as it used to for that kind of challenging maneuver.

Does that help?

That does make sense in terms of the precision gained by mapping markers to vector space as opposed to pixel space. And it meshes well with the previous AI day info we have regarding merging the multiple camera sources in to a single vector space used by the NN's.

But your info above seems to be a description of what FSD does with the data once it has it. What I was trying to understand was your statement that: "Creep wall and Median box which is powered by non real-time data that was collected by normal FSD cars and turned into ground truth."

I read that as indicating that the creep wall and median box areas are based on data collected by previous cars (perhaps incorporated in to some map-based data markers used by FSD cars?), rather than gleaned by the current car at the time it encounters that scene.

Are you saying the fact that the creep wall and median box features appear to be vector-space objects that is evidence of previously collected data?

Or maybe I'm mis-understanding what you are trying to say.

(again, thanks for your insights)

 

[Gigapress]

Sure, as you can see from the car's real-time vision, the lane lines are moving around constantly (they vibrate, wiggle, appear, disappear...etc), these are being reasoned through computer vision neural networks in non-deterministic pixel space (vs deterministic vector space). Pixel space is to non-specific to place a nice straight non-moving wall that is drawn to be the closest possible space to the lane, without being un-safe. If this was being done in pixel space, the car would most likely end up in the lane in some high percentage of cases or stop way behind the lane in a percentage of cases/scenarios. Only with ground truth (most likely mapped vector space) could that wall be drawn and trusted to such a high degree. And that is being done by reasoning through several ground truth markers in the scene and calculating angles and distances. Then, feeding that ground truth into the trained NN models allow them to be much more confident in their outputs, quicker and have more robust outputs. This also greatly benefits path planning which can take ground truth inputs and be able to plan paths that don't wiggle as the input is deterministic and non-changing as the car moves through the scene, thus turns that are somewhat occluded or blind to the real-time system are now much smoother. Case in point is when the car commits beyond the creep wall and moves to an actionable pose in the median box. This is now smooth as that median space is now trusted as ground truth (vector space to some degree) and the path planner does not need to re-plan near as much as it used to for that kind of challenging maneuver.

Does that help?

Is this kind of like HD mapping?