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

[heltok]

Last I knew the FSD computer isn't running redundantly either as they've had to (for a couple years now) needed to use cross-node compute since they'd run out of compute on a single node quite some time ago.

When it's working correcly it's not running redundant, that's true. But we don't really how they handle/will handle cases where 1 neural processor fails. I see a few different options:
1. Have a 50% smaller neural network ready to be loaded for these situations, run of the previous output for 1 cycle while the new network catches up ie run blind for ~0.06s instead of 0.03s when a neural network processor fails once.
2. Keep running the same neural network but instead of 36FPS only run it on 18FPS. Sure 0.06s reaction time is worse than 0.03s, but it still beats humans and is ~cat reaction time.

They only need to do this until they can safely stop the car on the side of the road. So the slightly increased risk of accident per mile will not be for millions of miles but for single digit miles thus the net addition of risk is insignificant in the scheme of things.

 

[Linkmeup]

Gotta post the gif man!

How I feel sometimes on this forum about ICE vehicles

 

[ggr]

I ask again, can the moderators please install a "" option in the "" reaction menu?

And I will reiterate that the moderators have no control over the forum software, and even our suggestions are often ignored.

 

[2Pearls]

There’s a sucker born every day.

But seriously, did you really have your eyes wide shut back then? I mean, you had to know for starters that the company was likely to go boobies up when you bought that S. Never mind FSD, which btw you paid a whole lot less for it than later purchasers.

Most people who bought an early Tesla were intellectually aware that they’d likely just kissed all the purchase money away. And while hindsight can certainly make you regret a lit of things, most people also were intellectually aware their investment in TSLA likely was throwing money in a fire. And for years it was.

I get it. But so far nobody has ‘FSD’. Everyone has some incomplete version of it. Since I paid about half of current pricing, perhaps I’m only entitled to half functionality of the final product at current pricing? Or maybe I’m only entitled to the first version of it, or did I buy when they started the second version of FSD? I know I paid way before this 3rd or is it the 4th remake Tesla is on?

Point being. Let it go. Be happy with the years you’ve had driving emissions free. And fun driving it has been, no? Be content in the knowledge your money helped keep the company afloat and that they didn’t waste it but put it to good use. Be proud of that and don’t let hindsight and regret rule your head. They are ugly bedfellows that want to destroy happiness.

And what about people who have the Model S, like my family, who, since early 2013 bought Model S to support the cause and be an early adopter (on our 4th Model S now), and yet, we still do not have active FSD as part of the 30 day trial. Tesla states they are working on a fix and hope to get this corrected soon. We have a Plaid, purchased in 2021. We would love to have a trial, yet, this only seems to be going out Presently to model Y and 3 owners. You would think they would support the people that supported them first, but that just never seems to be the case. No, not complaining, I can wait and will continue to support tesla.
But, certainly looking forward to your snarky response.

 

[Usain]

Aha! Another child masquerading as an adult.

Indeed!

 

[Knightshade]

When it's working correcly it's not running redundant, that's true. But we don't really how they handle/will handle cases where 1 neural processor fails. I see a few different options:
1. Have a 50% smaller neural network ready to be loaded for these situations, run of the previous output for 1 cycle while the new network catches up ie run blind for ~0.06s instead of 0.03s when a neural network processor fails once.
2. Keep running the same neural network but instead of 36FPS only run it on 18FPS. Sure 0.06s reaction time is worse than 0.03s, but it still beats humans and is ~cat reaction time.

They only need to do this until they can safely stop the car on the side of the road. So the slightly increased risk of accident per mile will not be for millions of miles but for single digit miles thus the net addition of risk is insignificant in the scheme of things.

As noted earlier, citing to a post by Green, the two nodes are not equal HW on most cars--- B node does not have network if A node is dead.

So your ideas can potentially work if Node B dies- but not Node A. Redundancy that only works for half your failures isn't really redundancy.

IIRC HW4 fixes this by giving each node a network port, but no way to fix this in HW3 cars, it's a limitation of the board.

 

[Krugerrand]

And what about people who have the Model S, like my family, who, since early 2013 bought Model S to support the cause and be an early adopter (on our 4th Model S now), and yet, we still do not have active FSD as part of the 30 day trial. Tesla states they are working on a fix and hope to get this corrected soon. We have a Plaid, purchased in 2021. We would love to have a trial, yet, this only seems to be going out Presently to model Y and 3 owners. You would think they would support the people that supported them first, but that just never seems to be the case. No, not complaining, I can wait and will continue to support tesla.
But, certainly looking forward to your snarky response.

I addressed all of that in the post you quoted from me.

 

[Singuy]

Last I knew the FSD computer isn't running redundantly either as they've had to (for a couple years now) needed to use cross-node compute since they'd run out of compute on a single node quite some time ago.

Some folks believe they'll eventually be able to trim back the NNs to fit back into a single node again (though usually it reads pretty strongly of hopium when they say it)-but currently there's no redundancy in compute on FSD at all.... there's also a few things not-perfectly-identical in the HW ability/connections on HW3 so even then you don't have true redundancy (see Greens tweet below)... HW4 might be better in this regard but is a minority of the fleet today

https://twitter.com/x/status/1572003268909203458

James Douma was talking about how NN software optimization has 10000x in speed using the same hardware. Google used to dedicate an entire data center to run a 1 billion parameter NN and now James can run 4 billion parameter NN on his laptop. Hardware did not 10000x but the software did. He mentioned that the FSD bottle neck was the CPU as the 300k C code utilize the CPU while all the NN stuff runs on the NPU which is around 1000x-10000x faster. Now they have free up the CPU by deleting all the C code that dealt with control and threw them into the NPU by making them into NN. He said it is difficult to get the software to run on the GPU and even harder to get them to run on the NPU while CPU is the easiest to program for.

So who knows if a TI 89 calculator can run FSD in the future. Texas instrument certainly price them as if they could.

 

[Usain]

How do you know they were poached? Maybe they were ready for a change and approached xAI themselves.

And of course, nobody knew who they are, or was talking about them, until somebody saw that they left Tesla.

People have been moving between Tesla and SpaceX for years. It's no surprise that the same thing would happen with xAI.

Technology workers move around when they see a new project that looks like fun. Totally normal.

 

[Knightshade]

James Douma was talking about how NN software optimization has 10000x in speed using the same hardware. Google used to dedicate an entire data center to run a 1 billion parameter NN and now James can run 4 billion parameter NN on his laptop. Hardware did not 10000x but the software did.

So who knows if a TI 89 calculator can run FSD in the future. Texas instrument certainly price them as if they could.

I agree nobody knows if some future version of FSD will ever fit back into a single HW3 node or not.

But none of that solves the lack of full hardware redundancy that exists in HW3 cars. Which was the bulk of the post you replied to.

 

[DarkandStormy]

I’m not trolling. I do not believe for a second that you are posting from a place of honesty. I’ve seen you before. Bursting through the door, come to save us all from ourselves. You can’t believe you’re the first.

I believe you’re here to try and convince long term shareholders that they should sell. You cherrypick data and then frame it in the most negative way, sometimes suggestively.

I’ve been here for over a decade. Tesla doesn’t have struggles today. They have blips, speed bumps, kamikaze squirrels.

I expect you to do a vanishing act, like so many before you, when those in control decide they’ve made enough churn money at the lower end of valuation and let the SP run again. I’ll be here, regardless of the ending but know that we’re all screwed, including you, if you’re right.

"if I'm right"

Where have I posted any predictions about the stock price sinking further?

Seems like you've had a lot of run-ins with bears before and are projecting. Sad.

 

[dedicatedtek]

And I will reiterate that the moderators have no control over the forum software, and even our suggestions are often ignored.

What ?! Mods ignored ? Say it ain't so.

 

[dedicatedtek]

If FSD is out of Beta and Optimus runs on the same perception stack...
Bobs gonna ask: "What is it you say you do here?"

Turtles

xkcd.com

 

[Linkmeup]

Auntie Cathy sure knows how to YOLO

https://twitter.com/x/status/1775518932088045748

 

[Singuy]

I agree nobody knows if some future version of FSD will ever fit back into a single HW3 node or not.

But none of that solves the lack of full hardware redundancy that exists in HW3 cars. Which was the bulk of the post you replied to.

Is hardware redundancy needed for robotaxies? The government doesn't seem to care.

 

[bpjod]

Why does this notion that Tesla doesn’t consider cold climates persist? It is obvious that vehicles need to be designed for all possible environments, so obvious that you thought of it without even being a full time employee. Tesla has been doing extensive cold weather testing since the Roadster days. This was one of the main reasons they went down the path of OTA software updates decades ago. They would do tests on the mule in Scandinavia or wherever and then ship software updates from home base in California to enable faster iteration cycles. Also, Tesla recruits engineers from all over the nation and the world. It’s not like they’re all California natives who have never seen snow. I mean the CEO went to college in Canada. And Norway is one of Tesla’s biggest markets.

The "the stuff that makes Tesla, Tesla - the California/Texas company" clearly needed an emoji to make it apparent I was joking around by drawing on an old criticism of the company. Sorry about that ommision on my part.

IF the low, recessed headlights are prone to becoming blocked by snow/slush though, it could be a genuine problem. I'm willing to wait a winter to see whether or not it's an issue or not as what I'm REALLY waiting for is a towing solution that is better than my Model X and the current CT is not it. The ability to book/rent the extended battery and have it installed/removed at a service centre (would love that option) or a model with the originally announced 500 mile range (option B) would cause me to order as soon as I get an invite but having more or less the same range as my current Model X doesn't do anything to improve my range while towing. Furthermore I'm not sure I could convince my wife to trade in her Model X for a Cybertruck and I'm angling for either a Model 3 Ludicrous or Roadster for myself whenever I pass my current Model 3 on to one of my kids...

 

[Zero CO2]

"if I'm right"

Where have I posted any predictions about the stock price sinking further?

Seems like you've had a lot of run-ins with bears before and are projecting. Sad.

you are a bit tiring with your constant negative bias ...dark and stormy indeed ... .. you have been here nearly 3 years and your post are predominantly bearish/negative ... i certainly don't agree with all things Tesla does but they are more right than wrong with their decision making ... and i find that you eventually catch up to understand how they work

for example my FS CT order , no communication , no VIN , no test drive, confirmed I am shareholder wit proof after email from Martin V , no response , still waiting for VIN... when they are ready to deliver your truck you might get an email , but the app will definitely prompt you what to do next ... you follow tasks pay $ and get your truck ... efficient , no unnecessary phone calls , no special treatment , no getting over ... I love it

 

[M|S|M MYP]

Auntie Cathy sure knows how to YOLO

https://twitter.com/x/status/1775518932088045748

We got bears saying its going to $14 or $0 or whatever and then we got auntie Cathie's PT at $2k. Good chance $TSLA lands somewhere in between.

NFA, do you own research

 

[Linkmeup]

We got bears saying its going to $14 or $0 or whatever and then we got auntie Cathie's PT at $2k. Good chance $TSLA lands somewhere in between.

NFA, do you own research

Tbh - If Starship actually turns out to be scalable (not only a success), then $5T isn't a big deal to get to.

 

[Zero CO2]

Auntie Cathy sure knows how to YOLO

Tesla stock going to $4,000 — that would be an increase of 1,100%: Money manager Catherine Wood

"If we're wrong, and all they do is electric, our bear case is $600," Wood says, a level that's still a double.

www.cnbc.com

throwback ...back in 2018 they called her Catherine.... CNBS had respect back then... funny thing is back in 2018 i would have been happy with the bear case