-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Software Update 2018.39.x (Version 9)
- Thread starter PoitNarf
- Start date
Knightshade
Well-Known Member
I haven't seen Tesla state drive-on-nav was an FSD feature- have you? If so, where?
On ramp to off ramp is absolutely an EAP feature, and still is....and drive on NAV explicitly is supported in AP2.0 and higher per the V9 release notes.
The only 2.5-only feature I'm aware of is the dashcam one.
Folks who paid for FSD will be getting a free processor update sometime in the next few months (or longer given Tesla Time) with roughly 10x the processing power, so that should take care of that.
The interesting thing about the 3's in the TeslaFi fleet is that 96.5% are on the same firmware version, 36.2. For S/X vehicles there is a much wider variation, no single firmware version is higher than 30%, with the majority of vehicles spread out between five different versions. Makes sense due to the different hardware configurations in the S/X where the three is a single config (for now).
Lets hope that is good news for the 3. With 2000 beta testers that should be a large enough population to validate and de-bug. Hopefully once Tesla is ready to release V9 to the M3 population it truly is a "wide release".
Lets hope that is good news for the 3. With 2000 beta testers that should be a large enough population to validate and de-bug. Hopefully once Tesla is ready to release V9 to the M3 population it truly is a "wide release".
Last edited:
I think that won't be nearly enough. It is very difficult to guess how much will be required, but I would think that 10x improvement is still going to be at least 100x short, and quite possibly much more. Each frame from each camera has to be analyzed in real time. Every object in each frame that could affect driving decision-making has to be CORRECTLY identified, and connected to whatever objects were identified in previous frames. Other sensor data (radar and ultrasonic) also has to be analyzed many times per second. That data has to be correlated with the data from the cameras to further build up the object list. Then another process has to project those objects into various possible futures, and from that a devise a driving strategy. That's a LOT of work. A lot a lot. I'm sure that Tesla processor could run a neural network that could beat a grand-master at Chess, but that's nowhere near what full self-driving would require.
You've put some thought into this! I'm not sure I'm convinced you're right, but who knows. I didn't get FSD because I thought just getting the regulations and testing would take at least 7 - 10 years in some locations. My guess is that the FSD will work in some places at first and may never work in others. It's one thing to have it work on a well defined, well marked, and well mapped road/route and quite another when things aren't "normal". Just one example: My Tesla gets the speed limits wrong in many parts of my city. One street it thinks the speed limit is 40 mph when it's actually 25 mph. Another example, on my way to my brothers house it thinks it should drive through a sheep pasture on a non-existent road. On that one my guess is that some county old map has a road that used to be there or was drawn there, but never built.
Nope! You get the update when you get the update. Absolutely no way to predict when that might happen. Keep your car connected to your home WiFi whenever possible to increase your chances of getting the update earlier; that’s the best any of us is able to do.
I'm just thinking ahead to update 2025.100.1.Anyway
Back to the topic at hand
Sorry.
Some people have reported that after driving with the new update for a while (50 to 100 miles) the renders became much more stable and accurate, and speculated it might have been the cameras calibrating. Have you noticed any improvements as you drive more, or is it just as jumpy as when you first installed it?
Knightshade
Well-Known Member
Yes, each frame needs to be analyzed in real time.
That's 60 frames per second times 8 cameras.... 480 frames per second to analyze in real time..
The current CPU can analyze 200 frames per second doing that.
The new CPU can handle over 2000 with redundancy. It was explicitly built from the ground up for the AI neural net Tesla is developing.
If the new capacity were insufficient, why would they bother doing the upgrade?
EAP is already near feature-complete (once drive on nav is enabled) so it's not needed for that. FSD would be the only reason.
diamond.g
Active Member
I thought there was a screenshot of the AP settings page that showed camera calibration %. But maybe that was explicitly for Nav on AP.
There's one screenshot that's a dev screen, but the public UI seems to only say calibration is in progress (not specifically which camera and how much).
But for whatever reason, they seem to immediately allow surround view to start rendering, even before calibration is complete. I wish there would be some UI indication that it is calibration and won't look as glitchy in another 50 miles.
No longer random updates. Sorry to keepers of this thread that want to keep it on point. So at least we know a way to get ahead of the line. Not just limited to 3 performance any longer.
Hopefully rewards come sooner than the others.
Might be a way to keep the calls to service about updates to people that really have car issues with being stuck on updates or actual mechanical issues.
Hopefully rewards come sooner than the others.
Might be a way to keep the calls to service about updates to people that really have car issues with being stuck on updates or actual mechanical issues.
Not true, I pick every time I get the update, I wake up and say ooooh, todays the day, let’s update!!
Much depends on what you mean by "analyze". Tesla currently has cars that hit the brakes because of shadows on the road. "Analyze" and "accurately analyze" are not at all the same thing. The more things you need to identify in a frame, and the higher the degree of accuracy required, the more processing power is required. "Self-driving" requires identifying many, many kinds of objects, and to a very high degree of accuracy. The job is vastly more complex than what AutoPilot is currently capable of.
And with this, I will let this drop rather than cause any more thread digression than I already have.
I’m on 36.2 and speech recognition is mostly broken for me. It worked great for the first week, but the last few days it works maybe one time in ten. Doesn’t seem to correlate with LTE signal strength either.
FSD will end up requiring AP HW 3, but that's fine because everyone who bought FSD for AP HW 2.0 or 2.5 will be getting the upgrade of the AP brain to HW 3 (the real difference) as part of the cost of FSD (i.e., you don't pay more, and Tesla already is expecting this "hit"). So if you want / bought FSD, don't worry about it. As long as EAP can be fully feature complete on HW 2.0/2.5 then no big deal.
If of course they decide that they need more performance for improved safety features and/or EAP improvements, then maybe they start upgrading all the HW 2.0/2.5 to HW 3 brains. Though if that happens I expect it to be a year away or so, to drive down the costs of the hardware (most of the costs are fixed costs to do with designing and laying out the custom silicon and such, once you're producing the chips the incremental cost of a new brain is pretty low, and easily absorbed by the price of EAP, for a minor retroactive hit to GM for older vehicles). For future vehicles it won't matter since they'll all ship with HW 3 even if only EAP is purchased.
I'd call a likely incremental cost (with the fixed costs already covered by development costs of FSD and HW 3 which is happening regardless) of replacement for AP 2/2.5 brains to be somewhere in the $300-500 range including labor, up to $1k at most worst case per vehicle. Tesla will probably have sold around 400-450k vehicles equipped with 2.0/2.5 HW brains, before the shift to HW 3 happens. So that might seem bad at up to $120-450M to upgrade EVERY AP 2/2.5 brain to 3, but considering the amount of cash they'll be bringing in over the next year and that EAP pricing probably brought in around $2B in the time frame that HW 2/2.5 was being installed... no big deal if they decide to go that route.
But if that happens, don't expect it for at least a year.
TL;DR : Don't worry about AP HW 2.0/2.5 vs AP HW 3. If you bought FSD you're getting the new hotness as a "free" upgrade when it comes out, and if you didn't you probably don't need it for EAP (not expected to). And if Tesla decides it's needed for EAP after all, the costs can be absorbed without too major a hit to the bottom line when they get around to doing it. Nothing to see here, move along.
If of course they decide that they need more performance for improved safety features and/or EAP improvements, then maybe they start upgrading all the HW 2.0/2.5 to HW 3 brains. Though if that happens I expect it to be a year away or so, to drive down the costs of the hardware (most of the costs are fixed costs to do with designing and laying out the custom silicon and such, once you're producing the chips the incremental cost of a new brain is pretty low, and easily absorbed by the price of EAP, for a minor retroactive hit to GM for older vehicles). For future vehicles it won't matter since they'll all ship with HW 3 even if only EAP is purchased.
I'd call a likely incremental cost (with the fixed costs already covered by development costs of FSD and HW 3 which is happening regardless) of replacement for AP 2/2.5 brains to be somewhere in the $300-500 range including labor, up to $1k at most worst case per vehicle. Tesla will probably have sold around 400-450k vehicles equipped with 2.0/2.5 HW brains, before the shift to HW 3 happens. So that might seem bad at up to $120-450M to upgrade EVERY AP 2/2.5 brain to 3, but considering the amount of cash they'll be bringing in over the next year and that EAP pricing probably brought in around $2B in the time frame that HW 2/2.5 was being installed... no big deal if they decide to go that route.
But if that happens, don't expect it for at least a year.
TL;DR : Don't worry about AP HW 2.0/2.5 vs AP HW 3. If you bought FSD you're getting the new hotness as a "free" upgrade when it comes out, and if you didn't you probably don't need it for EAP (not expected to). And if Tesla decides it's needed for EAP after all, the costs can be absorbed without too major a hit to the bottom line when they get around to doing it. Nothing to see here, move along.
Similar threads
