HW3

[S4WRXTTCS]

That's precisely what I thought but @PoitNarf above implied Elon meant all of 2.x I checked again and the question, though ambiguous, said HW2 family. So did Musk hear 2.x or 2.
If he meant every Model 3 purchased prior to April then I, too, will be BS.

I'm not going to put too much thought or energy into it until we see unique features for HW3. To me that's the real signal that upgrades will commence soon.

As to upgrades one would hope that they'd upgrade the oldest owners to the newest. Like an early HW2 FSD owner deserves it fairly soon within the upgrade windows, and a HW2 owner who just purchased FSD can wait a bit.

They probably won't do it that way though. They'll probably upgrade California based influences first to get the positive dude vibes going. They can't release it in Seattle because then it would rain all over it.

 

[wcorey]

I'm not going to put too much thought or energy into it until we see unique features for HW3. To me that's the real signal that upgrades will commence soon.

As to upgrades one would hope that they'd upgrade the oldest owners to the newest. Like an early HW2 FSD owner deserves it fairly soon within the upgrade windows, and a HW2 owner who just purchased FSD can wait a bit.

They probably won't do it that way though. They'll probably upgrade California based influences first to get the positive dude vibes going. They can't release it in Seattle because then it would rain all over it.

It's a chicken and egg thing. I am sure they already have it as they've referenced it and said it will do things in hardware that currently are done in software, it's what, 10x faster or 100x faster so frames processed/sec radically different and while the existing 2.5 computer is at 85% utilization the FSD computer will be at 15% utilization. I suspect it is required for Level 5. It might even be required for FSD at any level. So FSD end-of-year will be sub par until it can run on the faster hardware and move more functionality to hardware.

 

[S4WRXTTCS]

Apparently the FSD computer itself has lots of issues as well. Someone with an upgraded HW3 computer (so they do upgrade existing customers) lost all of AP and the mobile tech said ended up putting back the 2.5 computer.

I could see it being an Early access person or an employee, but I haven't heard of anyone's computer being upgraded.

It could be that they were testing upgrades, and something went wrong.

I haven't heard any issues with the HW3 computer from owners, but I wouldn't be the least bit surprised if there were some issues as it's brand new HW.

 

[wcorey]

I could see it being an Early access person or an employee, but I haven't heard of anyone's computer being upgraded.

That could well be. It was on here though. But then you'd need an advanced search engine to find )FSD computer or HW3) and "lost AP" and installed 2.5.
It's more work than I want to do right now.

 

[OCR1]

I have confidence that with HW3 Tesla will continue to provide future software updates that will make the current software better and more reliable. But I think it will be many years before any of us will be taking a nap while driving our Model 3’s. Maybe a decade or longer.

 

[Knightshade]

I don’t agree at all. I’m not price sensitive (I bought a P3) but I found the sales pitch for FSD during the test drive ridiculous when the cars started to go berserk on the screen.

Which is utterly meaningless to actual operation, again that's running hw/sw that can't even process all the video frames the camera is feeding it let alone do complex operations on them.

Hence why its capabilities are much more limited compared to what HW3 will be able to do.

I don’t expect FSD today but even HW2.5 (my P3 has HW3 and so did the brand new demo vehicle two weeks ago I assume) should be able to identify objects in a way that doesn’t make the system seem spastically broken.

Over a billion miles have been driven on AP so far, seems to mostly work fine as intended, even if your display looks "funny"

Also currently HW3 is simply running the HW2.5 software at this point, so that's meaningless as well. They're not expected to roll out public versions of the HW3-specific code until late this year. THAT is when you'll see a difference.

Again you're basically saying "the child brain can't do this thing, so obviously it's impossible an adult one can"

If they can’t do this without HW3 AND software then they should not display these ridiculous cars and lines at all. It’s a sure way to lose credibility and was the main reason why I didn’t purchase FSD.

Again- over a billion miles driven on AP. That's got a bit more credibility than "screen looked funny on a test drive"

I myself do a couple hundred miles a week on it and it works very well within its limits.

There's simply no reason to doubt a massive increase in compute capability (both for handling all cameras at full frame rate plus its other inputs and for being able to run much larger/more complex neural networks) won't result in a massive increase in competence and capability.



Your argument appears to amount to "I don't believe a more advanced system I've never actually used will work because a much more primitive and less capable system displayed things a bit oddly during a test drive"



Have you watched any of the stuff from autonomy day earlier this year? Hours of some deeply technical stuff but also-

Here's a demo they released (it's sped up quite a bit but shows a HW3 car running HW3 developer code handling an entirely hands-free drive including stop signs, stop lights, 4-way intersections, oncoming traffic, etc.

Now, do I expect L5 driving 5 minutes after the HW3 swap? No way. Regulatory speaking that wouldn't happen no matter how good the HW or SW.

I do expect HW3 should be, (with the HW3-only SW), capable of at least L3 or L4 driving on highways, and at least competent L2 driving on city/local roads and maybe higher... and capable of all of that within a year or so of general public release (again pending regulatory- because until that's worked out everything will be officially L2 no matter HOW capable it actually is)

If FSD delivered nothing BUT L3 highway I'd be thrilled with what I paid for it, more is gravy.

 

[Octo]

Look, I’ve worked on complex SW projects since I graduated from University in 1996 and have seen plenty of promises go up in smoke.

I saw the FSD tech pitch from Tesla and while it’s impressive it’s essentially trying to emulate a visual cortex with a neuronal network - the actual driving part doesn’t seem to be nearly that ambitious. It came across as “we’ll have an awesome vision system and the rest will follow”. We shall “see”.

My point was that it’s a bit much for Tesla to claim they’ll have robotaxis in 2020 if today they can’t even get the product that they sell today to display whatever the car detects without hilarity.

If they want $6000 for a promise, they better not release the ridiculous dancing cars. I’m a Tesla customer so I’m already invested and on board with their products (except FSD) but it’s a severe mistake for them to have this silly implementation they have today out in the wild. Whenever one of my friends who view Tesla skeptically see my P3, I wish I could disable this as to not embarrass myself with this toy level “crap”. If this is possible, please let me know.

 

[dmurphy]

Which is utterly meaningless to actual operation, again that's running hw/sw that can't even process all the video frames the camera is feeding it let alone do complex operations on them.

Hence why its capabilities are much more limited compared to what HW3 will be able to do.

I liken the HW 2 -> HW 3 upgrade to replacing an iMac G3 with a 2019 iMac.

Sure, they’re both iMacs, and both play movies, but that G3 is going to choke and fall over if you try feeding it a 4K copy of Infinity War.

That’s how significant the difference is between HW 2.5 and HW3. The “point five” version number difference is .. uh ... misleading.

 

[afadeev]

Calling BS on that now. Maybe a few token people but not mass upgrades for those of us whose thousands of dollars they've been sitting on and laughing at.

Don't be too harsh on Elon.
He said "End of Q4, most likely".
He did not say what year!



a

 

[SammichLover]

, they better not release the ridiculous dancing cars.

That display seems to undersell how good the information is underneath, how much the system's model understands of its environment. If you go to the Model S forum there's videos of the data ripped direct from the model and overlaid on video that seems a lot more stable and accurate (plus shows things that don't appear on that screen). You can't get to that info in the Model 3 because its isn't reliably hacked to root but nigh certain it is in there, too.

While not full on "yes they can do this", I will say I was far more a skeptic before seeing the progress of the last 9 month while owning the car. I'm beginning to suspect I greatly underestimated how fast they are going to be able progress.

P.S. I'll see your '96 and raise you 7 years more experience. Not that that matters as much here, more experience might actually lead to "we couldn't have done it, so they can't" thinking. This isn't programming like it used to happen, and mostly still does happen, where people generally needed to fully grasp exactly what was happening and will happen to be able to code it.

 

[dingyibvs]

That "billions of miles" driven is pretty useless. Most people quickly find out the limitations of AP and keep its operation within those confines. A student who does a thousand additions a day and nothing else will never learn calculus. Garbage in, garbage out.

These types of real world data are useful when failure is an option. In fact, it really only learns from failures. As such, they need to be tested in a controlled environment where the consequences of failures can be mitigated. Tesla cannot hope to save money and expedite AP development by skipping or skimping on what every other self-driving company is doing, which is to let the system fail time and again and not result in disasters because of trained professionals behind the wheel.

This discussion about HW2 vs HW3 is also plain silly for multiple reasons. For one, we simply don't know what level of computational power is needed to emulate human behavior. It may be HW15 with quantum computing capabilities for all we know. For two, with the advent of 5G, it's not unthinkable that computations will be done centrally at a remote location, and all a car needs is to be a sensor/transceiver.

Lastly, as with all tech attempting to emulate human behavior, the last 10% will probably take longer than the first 90%. IMO, if you're paying for FSD today, you're contributing to the development of self-driving as well as Tesla's coffers. Nothing more, nothing less. To expect to actually have your current car self-drive at level 4 or 5 is foolish. Your car will have long retired before that time arrives. This is, again, just my opinion.

 

[insaneoctane]

I corrected your statement in bold

I have confidence that with HW3 Tesla will continue to provide future software updates that will make the current software better and more reliable. But I think it will be many years before any of us will be taking a nap [on purpose] while driving our Model 3’s. Maybe a decade or longer.

 

[PoitNarf]

That "billions of miles" driven is pretty useless. Most people quickly find out the limitations of AP and keep its operation within those confines. A student who does a thousand additions a day and nothing else will never learn calculus. Garbage in, garbage out.

From what I've read, autopilot features are always running in "shadow mode" if they're not currently in use. Tesla is able to gather this data and look at what the car would have done in certain situations had autopilot actually been engaged. So the system as a whole is constantly learning from data it collects at all times whether AP is active or not.

For two, with the advent of 5G, it's not unthinkable that computations will be done centrally at a remote location, and all a car needs is to be a sensor/transceiver.

This doesn't seem like a good idea to me at all. It's another point of failure to add into the system. There can always be lag even on a fast data connection. What if you're driving in an area with spotty or no cellular data coverage? What if there's a server outage? The current system is able to react to potential dangerous situations in a fraction of a second. I wouldn't want to extend that reaction time by adding in the delay that would come with transmitting the current environment around the car to some central server and then waiting for a response back on how to handle the current situation. This would also make for a rather attractive target to some malicious actors. Want to take down autopilot across the entire Tesla fleet? Just DDoS their servers.

 

[Knightshade]

That "billions of miles" driven is pretty useless.

I mean, it's demonstrably not, since Tesla uses data from those miles to improve the system.

They spoke about this at great length during the autonomy day presentations.

Further my mentioning it was in reply to the claim that the system was not "able to identify objects in a way that doesn’t make the system seem spastically broken."

The guy was thinking that because the display sometimes shows dancing cars, it meant the AP system had no idea where anything was- the fact it has managed to drive (or, driver-assist) owners for over a billion miles of driving suggests that his fixation on the display might not reflect the actual capabilities of the functional parts of the current system.

Most people quickly find out the limitations of AP and keep its operation within those confines

Do they?

Because almost daily there's a new thread on here with folks insisting tesla needs to "fix" some aspect of the system, and 9/10 times it's a problem happening becuase they're using the system outside those confines... (on local roads, with 2-way and cross traffic, with intersections, etc)

And usually when someone points out the system isn't actually designed to work in those situations at all they argue about it.

These types of real world data are useful when failure is an option. In fact, it really only learns from failures. As such, they need to be tested in a controlled environment where the consequences of failures can be mitigated.

It already is. That's why the driver is supposed to be hands on wheel always ready to take over.

See the recent thread where a woman thought AP was about to steer her into a gore point on a highway fork, so took back over. From her perspective AP "failed" but since she was using the system correctly the consequences of that failure were mitigated into nothing more than an elevated heart rate.

The idea you need a "professional" to do this is nonsense- and it's why Tesla has a TON more real world fleet data than anyone else.

This discussion about HW2 vs HW3 is also plain silly for multiple reasons. For one, we simply don't know what level of computational power is needed to emulate human behavior.

Tesla isn't trying to do that.

They're trying to develop a significantly safer system than human behavior gives us.

But hey, if nothing else we know HW3 has more than enough power to process all the cameras at full frame rate and resolution, so that should at least be able to clear up the "dancing cars" issue the other guy apparently finds the sole reason not to buy FSD!

For two, with the advent of 5G, it's not unthinkable that computations will be done centrally at a remote location, and all a car needs is to be a sensor/transceiver.

This is probably a worthwhile path, years down the road, for things like automatic transport truck platoons travelling known paths with excellent cell signal.

For a personal car that might well go places with NO signal, perhaps not so much.

 

[Octo]

Further my mentioning it was in reply to the claim that the system was not "able to identify objects in a way that doesn’t make the system seem spastically broken."

The guy was thinking that because the display sometimes shows dancing cars, it meant the AP system had no idea where anything was

My comment was from the view of a customer who Tesla tried to pitch FSD for $6000.

If the underlying system is much better in detection of objects as you claim, why is the visualization, arguably magnitudes simpler to program than a freaking vision system for FSD, so terrible?

At best it’s a marketing disaster because people like me who see dancing cars where non are in reality, won’t fork over $6000. If the AP core is truly better than what the car shows on its screen, then a fix should be a 1 week SW project for an enthusiastic intern, right?

I suggest fixing this ASAP if they want to upsell this feature. Because outside of hardcore engineering fans like you seem to be it’s a ridiculous marketing boondoggle.

 

[SammichLover]

why is the visualization, arguably magnitudes simpler to program than a freaking vision system for FSD, so terrible?

It isn't 'magnitudes simpler' to make a human-useful display to display all that data. The core difficulty doesn't have anything to do with programming directly, rather an issue with display and human limitations.

Just go look and the videos and I think you'll understand why. There is so much data there. That ultimately isn't really that useful to a driver because you'd need to look at it. As I driver I generally ignore that section of the screen. I will glance if there's an audio telling me something on it is red but I'm not sure that's even useful to me as a driver directly in direct action? Just more of a "okay, that's what the car is on about". More data would create an even less useful thing as you'd have to sift through more visual data, which would probably have even finer details because of that, to find info that might matter to you as a driver in an actionable way.

 

[Octo]

Ok, honestly, do y’all really think that the current product isn’t problematic from a sales point of view?

Tesla wants a significant percentage of the cost of the car for future FSD. And what they present today shows obviously erratic behavior.

You can ridicule this as “the AP core is totally different, why do you focus on the display” but the fact is that the car shows non existing objects that can damage trust in the underlying system. And I don’t care if this is due to <HW3 limitations. If they cannot make this better today, it should be removed.

How something like that can ever pass quality control is beyond me. Truly bizarre.
I guess some people just have low expectations.

Today when I merged manually from the leftmost lane to the adjacent lane, the display freaked out showing a frantically rotating red “truck” that was supposedly in that lane. There wasn’t one.

LOL. Please how can I disable that silly feature?!? I’d be pissed if I were in the FSD team at Tesla and whoever owns that crappy visualization damages the reputation of FSD.

 

[gilscales]

Ok, honestly, do y’all really think that the current product isn’t problematic from a sales point of view?

Tesla wants a significant percentage of the cost of the car for future FSD. And what they present today shows obviously erratic behavior.

You can ridicule this as “the AP core is totally different, why do you focus on the display” but the fact is that the car shows non existing objects that can damage trust in the underlying system. And I don’t care if this is due to <HW3 limitations. If they cannot make this better today, it should be removed.

How something like that can ever pass quality control is beyond me. Truly bizarre.
I guess some people just have low expectations.

Today when I merged manually from the leftmost lane to the adjacent lane, the display freaked out showing a frantically rotating red “truck” that was supposedly in that lane. There wasn’t one.

LOL. Please how can I disable that silly feature?!? I’d be pissed if I were in the FSD team at Tesla and whoever owns that crappy visualization damages the reputation of FSD.

I actually find it comical when I look at the "Dancing icons" I understand that this is and has been a "Beta" product since release and have come to know its limitations as I have put 27k on my car in the first 15 months of ownership.

I do not trust autopilot to work as advertised as I have had many "mishaps" that scare passengers, I have never been scared to use it though as I use it as it supposed to be used and pay attention with one hand on the wheel but in many, many instances I will choose to turn it off as I do not like the jerky action of how it operates most times, I find it most useful in stop and go slow traffic and especially when on a long road trip and I'm out on the freeway away from traffic it is a good stress reliever when traveling 500 miles or more in a day.

I did buy FSD when it briefly went to 2k not because I'm a sucker for discounted vaporware (although I am) but because I wanted to insure they would install the HW3 computer as I was hoping as new software is released it would provide a better autopilot experience than the HW2.5 could deliver, we will see if it pans out that this could be the case or not most likely in the next year, I bet 2k of my money it would.

 

[Silicon Desert]

...... But I think it will be many years before any of us will be taking a nap while driving our Model 3’s. Maybe a decade or longer.

While you might be right, history has shown that even the most knowledgeable technologist underestimate the rate at which technology progresses. Let's just hope the same is true with your guess

 

[dmurphy]

Ok, honestly, do y’all really think that the current product isn’t problematic from a sales point of view?

No issue here. Didn't unsell me at all. Even my wife, who's admittedly a bit of a technophobe, doesn't have an issue with it.

Today when I merged manually from the leftmost lane to the adjacent lane, the display freaked out showing a frantically rotating red “truck” that was supposedly in that lane. There wasn’t one.

Which is better? Displaying a possible object, or ignoring it?

The answer is displaying a possible object, every single time. There's zero cases where it's better to ignore an object that may or may not be there.

Until the NN gets better, this is the limit of operation. Better to give the driver a "false bogey" than to miss one. Every single time.