Tesla infotainment system upgradeable from MCU1 to MCU2

[whitex]

Of course. How in the hell would that get 8 year old tegras.

First, it's been 8 years since the release of the chip, not end of production. Second, the direct answer to your question is PLC (Product LifeCycle) planning. Chips have a well defined lifecycle, with ability to purchase for years, with well-in-advanced end-of-production notices which allow customer to stockpile what they need. There are other automotive manufactures which use Tegra chips too btw, and they manage to do it. Tesla, knowing the failure rate will be 100% within the life of the car, should be stockpiling a whole bunch of them (or should have stockpiled - I don't know whether the Tegras they are using are past end of production yet). Oh no they need a whole bunch of them? Well, that is entirely self-inflicted by Tesla by their design - the Tegra chips are not failing, my guess is their MTBF is on par with other automotive parts.

Tesla is showing such poor planing (or no planning at all - Elon's "best process is no process" attitude). First, they should have ensured proper supply of all parts to support their cars for their lifecycle. Second, the darn Tegra+EMMC already is on a module board, which they should be replacing instead of an entire MCU - considering that every MCU car will need at least one swap in its life, that would be so much less expensive (both to customers and to Tesla, not having to stock such expensive parts)! Maybe this is Elon's agile methodology as applied to product support, solve the problem when arises, don't plan ahead, who cares if it's expensive then, let's cross our fingers it won't happen?

 

[whitex]

With some types of ASICs such as commodity CPUs, it is very difficult to get a new batch made after the initial production has stopped. The chip fabs introduce new manufacturing processes that produce chips that are totally incompatible with the prior chips (voltages are different, timings are different, etc).

On the one hand, the NVIDIA chips were probably made in very large volumes and so there are probably lots of spares floating around, on the other hand there may be other support chips on the board that aren't as easy to get; if tesla wasn't planning on this board having a high failure rate they may simply not have banked enough parts to maintain a reasonable supply of parts. It would probably be completely impossible (IE absurdly expensive, not actually "eat-a-neutron-star-sandwich" impossible) to make a brand-new iphone 3gs using only parts manufactured in 2019, for example, because there are too many highly integrated parts for which there are no manufacturing capacity.

We will very soon know how diligently they've been refurbishing these MCU1s -- the continuum is anywhere from

1. ship untested returns to other customers, repair boards that come back two times
   
2. drop off a batch of dead MCUs and ebay bought eMMCs at a San Bruno phone repair place that will drop ship the fixed boards to service centers
3. send boards off to an aviation electronics specialty shop that re-certifies the part after putting it through extensive testing including vibration, wide temperature ranges, and finally x-raying the boards

If there is a large return rate on replaced MCUs, they're taking options 1 and 2...

As far as other car manufacturers goes -- they tend to farm out the actual manufacture of a part to other companies (delphi, delco, bosch, etc) and those companies tend to focus on producing a stable family of products used to integrate into a wide range of cars and so they have an incentive to have a small family of stable products to drop into different OEM requirements. This approach results in lower integration density and less optimization but a slightly more stable supply chain.

At some point Tesla will need to address legacy support. They will face a variety of issues that taint the brand, things like MCU failures, battery degradation, etc. They can either walk away Nissan style or $tep up like Lexus or Mercedes. Time will tell.

There are other auto manufactures which use Tegras and plan their product lifecycle support just fine (and yes, most will use other companies like you mentioned to handle the design and manufacturing of the modules). Most chips have a well defined lifecycle and planned end-of-production.

 

[ChrML]

$500 core charge? It seems they really want the old one to keep having remanufactured ones, which probably means that they don't have a supply of new ones and don't have a way to substitute with MCU2. Another unique experience of owning a Tesla vs. a traditional car manufacturer which supplies new parts for years and after that aftermarket parts are also available.

Remanufactured. Basically just a nice way for Tesla to say: We gave you an used MCU1 with a new 10$ eMMC chip from another repair that the other customer paid 2100$ for. Oh and you need to pay 2100$ too.

 

[whitex]

Remanufactured. Basically just a nice way for Tesla to say: We gave you an used MCU1 with a new 10$ eMMC chip from another repair that the other customer paid 2100$ for. Oh and you need to pay 2100$ too.

When Elon said Tesla cost of ownership is cheaper than an ICE car, he meant cars like Bugatti or Ferrari, not all ICE cars.

 

[Nenarek]

What I hope happens is that the FSD computer upgrade is actually a stealth MCU capability upgrade. If rendering can happen on the new computer the MCU could run remote display software as well as whatever else makes sense to keep there. Networking comes to mind.

Why does the FSD chip have 12 A72 cores? How many GPUs does it have? One? One per four CPU cluster? Could one four cpu cluster with GPU be dedicated to running media software?

 

[cduzz]

There are other auto manufactures which use Tegras and plan their product lifecycle support just fine (and yes, most will use other companies like you mentioned to handle the design and manufacturing of the modules). Most chips have a well defined lifecycle and planned end-of-production.

Yeah, I imagine that the CPU itself is likely easy to get; hopefully there are no other high density ASICs on that board, but there may be. Hopefully there is someone at tesla keeping their eye on the supply chain and demands on their spare-parts.

Guess we'll find out, right?!

 

[brkaus]

What I hope happens is that the FSD computer upgrade is actually a stealth MCU capability upgrade. If rendering can happen on the new computer the MCU could run remote display software as well as whatever else makes sense to keep there. Networking comes to mind.

Why does the FSD chip have 12 A72 cores? How many GPUs does it have? One? One per four CPU cluster? Could one four cpu cluster with GPU be dedicated to running media software?

Way too complicated & still reliant on the eMMC so the same risk is there. And their MCU software would then need to run on yet another architecture.

 

[NoMoGas]

What I hope happens is that the FSD computer upgrade is actually a stealth MCU capability upgrade. If rendering can happen on the new computer the MCU could run remote display software as well as whatever else makes sense to keep there. Networking comes to mind.

Why does the FSD chip have 12 A72 cores? How many GPUs does it have? One? One per four CPU cluster? Could one four cpu cluster with GPU be dedicated to running media software?

FSD has zero to do with MCU

 

[StarLog]

Tesla is showing such poor planing (or no planning at all - Elon's "best process is no process" attitude). First, they should have ensured proper supply of all parts to support their cars for their lifecycle. Second, the darn Tegra+EMMC already is on a module board, which they should be replacing instead of an entire MCU - considering that every MCU car will need at least one swap in its life, that would be so much less expensive (both to customers and to Tesla, not having to stock such expensive parts)! Maybe this is Elon's agile methodology as applied to product support, solve the problem when arises, don't plan ahead, who cares if it's expensive then, let's cross our fingers it won't happen?

The problem with replacing the daughter board is the Tesla tech's are nothing more then appliance users. They only know how to replace modules, not down to a chip or small board level.

They do not have the skills to go into the MCU and pull it apart like the videos we have seen by the Calif. and East coast tech shops.

 

[cduzz]

The problem with replacing the daughter board is the Tesla tech's are nothing more then appliance users. They only know how to replace modules, not down to a chip or small board level.

They do not have the skills to go into the MCU and pull it apart like the videos we have seen by the Calif. and East coast tech shops.

I wouldn't want a rework shop in each SC or even each region.

I would expect a robust supply chain where expired MCUs are taken out, shipped to a vendor who specializes in this (or department at tesla), fully re certifies each board, and ships refurbished boards back out to service centers. In a perfect world those reworked boards would be thoroughly reworked including larger eMMC chips, better heat sinks, etc. Practically speaking, hopefully they at least put the reworked boards through a heat cycle or two, x-ray the boards, and put proper asset tracking in place to identify who/what/where/when did the work so they can manage proper quality control.

The service centers need to grind through a broad array of work as quickly as possible.

 

[Galinette]

Hmmm... let us hope that at some point the logistics of fixing MCU1s become so expensive for Tesla that developing a MCU2b retrofit is cheaper for them

Is the MCU covered by extended warranty?

 

[MP3Mike]

The problem with replacing the daughter board is the Tesla tech's are nothing more then appliance users. They only know how to replace modules, not down to a chip or small board level.

They do not have the skills to go into the MCU and pull it apart like the videos we have seen by the Calif. and East coast tech shops.

They certainly know how to tear the MCU down to replace the 3G cellular daughterboard with the 4G cellular daughterboard. So there is no reason that they couldn't replace the Tegra daughterboard. (It doesn't appear to be much more work than replacing the screen.)

They certainly wouldn't be replacing the eMMC themselves, but they could replace the board fairly easily.

 

[whitex]

The problem with replacing the daughter board is the Tesla tech's are nothing more then appliance users. They only know how to replace modules, not down to a chip or small board level.

They do not have the skills to go into the MCU and pull it apart like the videos we have seen by the Calif. and East coast tech shops.

They can replace an LTE modem in the MCU. Replacing the Tegra module (not the chip) is very similar.

Edit: Just saw @MP3Mike wrote the exact same thing, maybe worded better than mine .

 

[T.R.T.e.s.l.a.]

Is the MCU covered by extended warranty?

It sure is.

 

[cduzz]

It sure is.

"Covered"

I'll reveal a bit too much about myself by reliving this anecdote but I'll do so anyhow. "Back in the day" my parents bought their kids an Intellivision. Sadly it broke in 3 months and we took it to the repair depot and got a replacement; that one was broken in a different way. That one went back in a week and we got another that was broken in the same way as the first and may even have been the first one we brought in. We went back and forth several times more but never got one that actually worked. I took apart the last one and repaired it myself (the controller cables kept coming lose).

It will be interesting to see how Tesla addresses this and other obvious forthcoming supply / product life cycle challenges, especially in light of how their tremendous vertical integration precludes a stream of 3rd party parts.

 

[Jackotai]

One of my advise to easily solved everyone ‘s need for a pre-MCU2 is asking tesla to develop a WiFi Ethernet feature that everyone can spend fews bucks to buy an iPad and hang on. Tons of media resource as well as a separated &larger screen that may serve many needs.

 

[Shateam]

Haven't seen this posted yet.

Interesting analysis on Twitter by @verygreen

green on Twitter

Clearly when Elon said this issue is "much better now" he was not talking about the amount of logging that continues to happen.

 

[S100D]

Haven't seen this posted yet.

Interesting analysis on Twitter by @verygreen

green on Twitter

Clearly when Elon said this issue is "much better now" he was not talking about the amount of logging that continues to happen.

It gas actually gotten worse with the 2019.32.12 releases. Every time I drive the car, Easy Entry is writing to the eMMC tevery time the seat moves.

 

[KyleDay]

It gas actually gotten worse with the 2019.32.12 releases. Every time I drive the car, Easy Entry is writing to the eMMC tevery time the seat moves.

Hate to break it to you, but what do you think streaming music (I.e. the introduction of Spotify) does to the eMMC? It’s not buffered to memory, it caches to eMMC for playback on-demand later on. EMMC is worn far more through music streaming than any easy entry logging.

Only answer is to root your car, setup a ram disk to write logs and other media cache activity to ram instead of storage.

 

[S100D]

Hate to break it to you, but what do you think streaming music (I.e. the introduction of Spotify) does to the eMMC? It’s not buffered to memory, it caches to eMMC for playback on-demand later on. EMMC is worn far more through music streaming than any easy entry logging.

Only answer is to root your car, setup a ram disk to write logs and other media cache activity to ram instead of storage.

Unbelievable!!! I was not aware that streaming is writing to the eMMC. I normally do not stream music but I do stream ABC News while driving in town so it is still being written to the eMMC.

My MCU1 car is still under warranty so I will wait to get the eMMC replaced and rooted.

Thanks for letting me know it is worse than I thought.