Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

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

This site may earn commission on affiliate links.
This concerns me.

Our society has normalized over-the-speed-limit driving. Seems like NHTSA is slowly making FSD unappealing by legal nannying it to death.

Are Blue Cruise and SuperCruise put under similar constraints? For that matter every cruise control mechanic on the planet. Most of modern bersions are aware of speed and this constraint could easily apply to them. There is no speed limit above 75 MPH in the US…

The obvious conclusion is that every car capable of speeds in excess of the speed limit must be limited to the speed limit, regardless of whether cruise control, Autopilot or manual driving. Because in all cases the driver is ultimately responsible for the speed. The fact that the driver can set a maximum speed in software to the actual speed limit should be all that's necessary to make it analogous to a manually driven car or one on cruise control.
 
It is.... Tesla ran out of compute on a single node on HW3 over 2 years ago now and has been forced to use the compute of both nodes for a single instance of the full stack since then- no redundancy if either node crashes.

From the HW4 description they've almost (but not quite) doubled the compute on each node, so presumably they'll be able to fit the whole thing into a single node there if they don't run into a compute-requirement-wall again like they did on HW3.
Yes. However, does the entire software have to continue running after failure of a uP node? I expect all that is needed for L3 is to hand control back to the human driver, and for L4 all that is needed is to safely bring the car to a stop in a safe place. Neither of those tasks should require the full FSD software to continue to be run, IMO.

GSP
 
Tesla offering $5k if you trade in S/X with unlimited supercharging

Weird thought here.

It’s likely a big percentage of cars with the previous FSD agreement which promised full Robotaxi, also come with lifetime unlimited Supercharging.

If HW2.5 & HW3 cannot support Robotaxi, Tesla can quietly remove these cars from the road, switch out the FSD requirement so it’s just ADAS, remove the lifetime Supercharging, and put them back on the road and retire all that liability.

At some point, Tesla is going to have to settle with the remaining owners who have these older cars with the broader definition of FSD. This may be a way to preempt that cheaply.
 
I'm been rethinking my "definitely hot" comment about today's PPI numbers. If you average together the last two PPI number, -0.2 and 0.7, you get 0.25. That's a drop from the previous three months and not too far from to the Fed's desired 2% annual inflation rate. Seen in that context, not so bad. It also sets us up for a nice month over month on the next PPI, should January numbers have been an anomaly rather than the start of a trend.

We need to realize there is noise in the data, and that is especially hard to see when you have sparse sampling without any smoothing.

Meanwhile if you were to look at this graph, you'd better be able to extract the signal from the noise and see that inflation is still heading down, albeit slower than it was in Oct-Nov. That's okay, because if it kept that pace, we would be in deflation by the summer.

My guess? We hit 4.x% YoY inflation in March, and 3.x% inflation in May. After that, who knows? But 3.x% inflation isn't a terrible trajectory for now.

Screen Shot 2023-02-16 at 11.10.07 AM.png
 
This was already done at the last earnings call. And I think they squared that circle pretty sufficiently:

"Hardware 3 will not be as good as Hardware 4, but I'm confident that Hardware 3 will so far exceed the average -- the safety of the average human. So how do we get ultimately to -- let's say, for argument's sake, if Hardware 3 can be, say, 200% or 300% safer than human, Hardware 4 might be 500% or 600%. It will be Hardware 5 beyond that. But what really matters is are we improving the average safety on the road."

Yes, HW3 is inferior to HW4, but both will be safer than human drivers. If Tesla can deliver FSD with HW3, there are no broken promises and no need for an expensive retrofit.
They need to charge more for HW4 cars then to make this look different. If the FSD package is the same price on HW4 than HW3, but HW3 can do less they are going to get rightfully sued. I think they should at a minimum offer a route to refund the FSD piece to Enhanced AP or something less.

Maybe HW4 cars will only be subscription...
 
They need to charge more for HW4 cars then to make this look different. If the FSD package is the same price on HW4 than HW3, but HW3 can do less they are going to get rightfully sued. I think they should at a minimum offer a route to refund the FSD piece to Enhanced AP or something less.

Maybe HW4 cars will only be subscription...
Sued for what?

Put another way: Does Apple get sued because the iPhone 14 is more capable than the iPhone 13?

Increasing capacity of technology products over time is as inevitable as the tide.
 
Stock seems to be dropping because of this. 362,758 cars being recaled 'by over the air update' as required by NHTSA because FSD may cause a crash due to allowing the vehicle to "exceed speed limits or travel through intersections in an unlawful or unpredictable manner increases the risk of a crash".
If the “fix” limits Auto-Pilot and FSD to the speed limit, it could affect demand………and certainly cause accidents in many areas of the US.
 
It is.... Tesla ran out of compute on a single node on HW3 over 2 years ago now and has been forced to use the compute of both nodes for a single instance of the full stack since then- no redundancy if either node crashes.

From the HW4 description they've almost (but not quite) doubled the compute on each node, so presumably they'll be able to fit the whole thing into a single node there if they don't run into a compute-requirement-wall again like they did on HW3.
Tesla has not run out of compute. You need to watch videos by James Douma. Just for one example, for testing purposes, Tesla sometimes runs parallel neural nets with duplicate functionality, just to be able to compare results.

And also fail-over to the surviving processor, does not need ALL the functionality as another poster alluded.
 
Sued for what?

Put another way: Does Apple get sued because the iPhone 14 is more capable than the iPhone 13?

Increasing capacity of technology products over time is as inevitable as the tide.
No but apple doesn't offer a massive software package for an iPhone with "future" capabilities. They would need to modify what FSD can do and charge more for that with HW4 otherwise people who bought FSD are getting less product for the same money. My point being is Tesla needs to be aware of this perception and allow those that don't see 10-15K worth of ability to get their money back.
 
I think that recall nailed it. We have had issues with all those things. Especially being in the wrong lane at an intersection (although since a couple updates ago it’s not 100 percent wrong all the time now). But the speed limit thing will get you a ticket for sure when speed limit changes to a lower number.

Rolling stops.,, meh,, sometimes. Yellow lights can be on the edge sometime but not a biggy.

Still. Glad to see they were forced to fix the wrong lane thingy. Kinda liking this NHTSA over watch. 😊.
 
Yes. However, does the entire software have to continue running after failure of a uP node? I expect all that is needed for L3 is to hand control back to the human driver, and for L4 all that is needed is to safely bring the car to a stop in a safe place. Neither of those tasks should require the full FSD software to continue to be run, IMO.

GSP


There are certain things only one of the two nodes can do (IIRC there's at least a couple things only one PHYSICALLY CAN do as not every connection is redundant on HW3- For example only Node A can fire triggers, node B can not)--- in fact Green specifically called out the fact they "fixed" some of that in the HW4 stuff... so if that was the node that crashed it'd be a problem.

Further, given the way the compute is split, there's no practical way to have one be "good enough" for either scenario you suggest- since parts of the perception and driving systems are split in both cases so the "fail safely" requirement would be impossible currently.... plus L3 requires a non-zero amount of warning to the driver to take over before the vehicle gives up on it, which a node crash wouldn't allow. Not to say they couldn't potentially do a major re-write to somehow fit a "good enough to fail safely" set of code into a single node but it couldn't be done with what code they have now... and you'd still have the problem of one of the nodes failing would HW-wise be a redundancy problem due to the 2 nodes not being exactly equal in capability.


All that said... it's maybe a precursor of things to come on the "offer more for trade in" thing just posted about FUSC that Tesla might offer something similar to pre HW4 FSD buyers to encourage getting those cars off the road and limiting liability if they can't deliver L4 on the older HW.


Quick reminder though this sub-forum exists and is prob. a better place to get into the weeds on this stuff:

 
It is.... Tesla ran out of compute on a single node on HW3 over 2 years ago now and has been forced to use the compute of both nodes for a single instance of the full stack since then- no redundancy if either node crashes.

From the HW4 description they've almost (but not quite) doubled the compute on each node, so presumably they'll be able to fit the whole thing into a single node there if they don't run into a compute-requirement-wall again like they did on HW3.
I am aware they did, but that was at a time where full redundancy was not required, and it still isn’t.

But for fully unattended driving it will be required, and it is my belief (and it was also mentioned) that a lot of optimization will be performed by then - bringing it back to one node.

But I do not know of course - either way.
 
  • Like
Reactions: BrownOuttaSpec
No but apple doesn't offer a massive software package for an iPhone with "future" capabilities. They would need to modify what FSD can do and charge more for that with HW4 otherwise people who bought FSD are getting less product for the same money. My point being is Tesla needs to be aware of this perception and allow those that don't see 10-15K worth of ability to get their money back.
Tesla tells you what you are buying when you purchase FSD (linked image). That is what they need to deliver.

So long as they deliver what they have on the package they should be fine.
 

Attachments

  • 1676576224680.png
    1676576224680.png
    90.8 KB · Views: 80
Last edited: