There will be NO HW4 upgrade for HW3 owners

[lowtek]

Anyone know how many months it took for HW4 production to replace HW3 production?

Just wondering how long it might take for HW5, then HW6, etc, until there's enough processing power to implement a full-suite of neural networks required for level-5 full-autonomy.

I don't think it has yet, we'll know when forum members start taking delivery in March. My date says Feb 15 - March 22. I'm now considering giving up my $250 if I don't get HW4 and reject delivery, although I won't even buy FSD again until it's worth what they are charging for it.

An Apple M1/M2 laptop would be enough for "a full suite of neural networks", it's the software that's going to take another 5-15 years to get right.

That's a good comparison I hadn't thought of. The Apple M1 chip has one of the most efficient neural engines available and they just one-upped it with the M2 chip ... but that doesn't mean the software won't run on the M1, it still does just fine, albeit perhaps slightly slower.

The factor really isn't the computer, it's the lack of what sensory inputs are actually required for Level 5. For example, now they are introducing HD Radar back into the mix.

I doubt Tesla has Level 5 in mind with the current fleet and foreseeable future ... I think right now they are just trying to get FSD figured out ...

 

[Daniel in SD]

I think right now they are just trying to get FSD figured out ...

What is FSD though?

 

[ddeflyer]

What is FSD though?

That is going to be the multi-billion dollar question!

 

[lowtek]

What is FSD though?

All just my personal opinion FWIF ... but Full Self Driving != Level 5 Autonomy at this point ...

"Tesla Will Have Level 5 Self-Driving Cars This Year" ~ Elon (2020)

So yea ... 2022 and we still exist where optimism meets edge cases ...

Telsa has smart people working on this, if it can be done I think they will be the ones to do it.

 

[Daniel in SD]

All just my personal opinion FWIF ... but Full Self Driving != Level 5 Autonomy at this point ...

"Tesla Will Have Level 5 Self-Driving Cars This Year" ~ Elon (2020)

So yea ... 2022 and we still exist where optimism meets edge cases ...

Telsa has smart people working on this, if it can be done I think they will be the ones to do it.

Haha you didn’t answer the question. I think we can all agree that FSD is not currently capable of being deployed as an L5 system (though I think it is an L5 system, it’s just in beta still and very likely won’t get out of beta with either HW4 or HW3.)

 

[lowtek]

Haha you didn’t answer the question. I think we can all agree that FSD is not currently capable of being deployed as an L5 system (though I think it is an L5 system, it’s just in beta still and very likely won’t get out of beta with either HW4 or HW3.)

Here's what I do know ... it won't be until people grow up with self driving cars before they will trust them completely ... As a software engineer who works with machine learning, I will never trust "AI" software with my life if I have a choice ... but at some point Level 5 will be thing and "manually driving" will be as archaic and laughable as horse drawn carriages

... but somebody has to lead the way

 

[hoeschjp]

I'm not sure if you mean this jokingly but you're right, HW3 is more than capable, it's the software that needs to evolve. They've already re-engineered the stack at least 3 times as far as we know.

That might be true in theory, but it only matters if they actually evolve the software. At some point HW3 releases will have little to no focus, with only maintenance support, and my main worry with HW4 (and the new sensor suite) is that this point may be quite soon. Once we have an initial park-seek implementation, I'm not sure why they'd spend much effort on legacy hardware.

 

[Daniel in SD]

Here's what I do know ... it won't be until people grow up with self driving cars before they will trust them completely ... As a software engineer who works with machine learning, I will never trust "AI" software with my life if I have a choice ... but at some point Level 5 will be thing and "manually driving" will be as archaic and laughable as horse drawn carriages

... but somebody has to lead the way

From what I hear people very quickly get used to the driverless Waymo and Cruise vehicles. I don’t think it will be an issue.

 

[wwu123]

An Apple M1/M2 laptop would be enough for "a full suite of neural networks", it's the software that's going to take another 5-15 years to get right.

That's a good comparison I hadn't thought of. The Apple M1 chip has one of the most efficient neural engines available and they just one-upped it with the M2 chip ... but that doesn't mean the software won't run on the M1, it still does just fine, albeit perhaps slightly slower.

Your Apple analogy might be an apt predictor of what has happened and will happen to HW2, HW2.5, HW3, HW4 owners. I have an original Apple iPad, 1st gen, at home (was built like a tank), as well as a 1st gen iPad mini - both still charge just fine, power on just fine, still run their version of iOS. Both stopped receiving OS updates, security updates, browser app updates, Youtube updates, years and years ago - and most of their existing app versions were deprecated.

Both are basically useless, not because they run slower - will never know if they can run current functionality slightly slower, as their software was deprecated years ago.

Even MCU1 owners today who upgraded to HW3, the vast majority still can't get access to the FSD beta.

 

[EVNow]

Even MCU1 owners today who upgraded to HW3, the vast majority still can't get access to the FSD beta.

Thats why legacy auto companies will not even try to be backward compatible yearly.

 

[gearchruncher]

Thats why legacy auto companies will not even try to be backward compatible yearly.

It's also why "legacy" auto companies wouldn't sell you a feature in 2016 via a faked video and no actual technical proof it can be done at all, much less on the hardware they are selling you.
The fact that Tesla has been on HW2, HW2.5, HW3, and now HW4 all before they have delivered anything near that 2016 video tells you that maybe the legacy automakers are onto something by not faking it until they make it. But they're just slow, antiquated, dying companies that will surely be overtaken by automakers that charge $15K for things that don't exist yet. So innovative!

 

[EVNow]

It's also why "legacy" auto companies wouldn't sell you a feature in 2016 via a faked video and no actual technical proof it can be done at all, much less on the hardware they are selling you.

Nissan sold me L3 charging port for $2k that I never got to use because the promised L3 chargers never materialized.

Let’s not even start talking about 8 year batteries that started losing 20%+ within a summer.

 

[mtndrew1]

My only (irrational) hope at this point is that the new radar unit is retrofitted to HW3 cars with FSD.

The new HW4 form factor and inputs make it pretty clear that retrofits aren’t happening.

 

[borland]

My only (irrational) hope at this point is that the new radar unit is retrofitted to HW3 cars with FSD.

The new HW4 form factor and inputs make it pretty clear that retrofits aren’t happening.

With enough pressure on Tesla by the Tesla owners groups, a HW4 processor on a HW3-form-factor (fewer sensors however) could be made for retrofitting.

Just don't get you hopes up for HW4 bringing full-autonomy.

 

[DL_Mike]

Nissan sold me L3 charging port for $2k that I never got to use because the promised L3 chargers never materialized.

Let’s not even start talking about 8 year batteries that started losing 20%+ within a summer.

What is the port type?

 

[gearchruncher]

Nissan sold me L3 charging port for $2k that I never got to use because the promised L3 chargers never materialized.

But the car could actually L3 charge, right? Did Nissan promise to you that if you bought working hardware on the car that they would build chargers?
This is like saying you bought a 4X4 and then never got to use it because your neighbor wouldn't let you drive on their dirt roads. Whereas Tesla is telling you they'll upload a 4X4 system "someday."

Let’s not even start talking about 8 year batteries that started losing 20%+ within a summer.

Did Nissan tell you the battery would not do that? Would it be better if they had charged you for a battery that wasn't even on the car?

 

[M109Rider]

It's also why "legacy" auto companies wouldn't sell you a feature in 2016 via a faked video and no actual technical proof it can be done at all, much less on the hardware they are selling you.
The fact that Tesla has been on HW2, HW2.5, HW3, and now HW4 all before they have delivered anything near that 2016 video tells you that maybe the legacy automakers are onto something by not faking it until they make it. But they're just slow, antiquated, dying companies that will surely be overtaken by automakers that charge $15K for things that don't exist yet. So innovative!

Uh huh.
Folks spin what ever narrative matches what they want to happen or believe. No matter how far off they are ..

 

[M109Rider]

But the car could actually L3 charge, right? Did Nissan promise to you that if you bought working hardware on the car that they would build chargers?
This is like saying you bought a 4X4 and then never got to use it because your neighbor wouldn't let you drive on their dirt roads. Whereas Tesla is telling you they'll upload a 4X4 system "someday."

Did Nissan tell you the battery would not do that? Would it be better if they had charged you for a battery that wasn't even on the car?

You mean, because they didn’t say the battery would loose 20% it’s okay. ?
Lol.
Spin your narrative. No one will change your mind, because you want Tesla to be the bad guy no matter what. So they will.

 

[AndrewZ]

Keeping this to autonomous promises, Tesla was not alone in promising this and that was right around the corner. Other car companies were trying to capitalize on it back to 2016, but I think Tesla is the only company that continuously doubled down in the years since.

Even after breaking up with MobilEye in the early days, going so far as to replicate the software themselves and triple and quadruple down.


I don't think any of this is that much of a concern to the company now. Tesla has produced what, less than 4million vehicles to date and ~400,000 with FSD Beta in the US? That is a drop in the bucket for a company aiming to produce 20million per year by 2030.

 

[DanCar]

..., should we wait for FSD to be complete, ...

How many years do you think that will take? Or are you ready to wait indefinitely?

..., or do you just hope it fails now for some ulterior reason…

I think everyone considers it a failure including Musk, in terms of being Full Self Driving. I'm confident they can turn that around in the 5+ year time frame. 2 years for Level 3.