MCU2 - Retrofit

[MXWing]

You're talking MCU1->MCU2 retrofit? Not even close.

I've said it a few times already... for the foreseeable future, no MCU2 retrofit will even be possible due to the number of changes to the car's internals (IC, wiring harnesses, etc). Maybe a few years down the road Tesla might develop a drop-in MCU upgrade for MCU1 cars, but that is way off. And even when that happens, it's not going to be a $500 upgrade like the LTE radio which was just a daughterboard swap on the MCU.

What about adding in an additional computational unit in the form of a tablet?

I'm not even sure $500 would cover infrastructure costs today to swap out even just a motherboard? They are not selling HEPA or tow kit retrofits anymore on the site.

First option was $500 for $50 part (if that). Tow kit is probably 66% or more margin not counting labor and other infrastructure.

 

[HankLloydRight]

What about adding in an additional computational unit in the form of a tablet?

Two tablets in the car? I don't think so.

 

[MXWing]

Two tablets in the car? I don't think so.

Semi has two I believe? Not saying it would be particularly elegant but you'd greatly expand the amount of data you can see as well as allow for upgrades.

I compute with 4 monitors.

 

[CarbonFree]

There have been some assertions made that the new MCU is only slightly faster than the seven year old Tegra 3. That is not correct; the Intel Apollo Lake platform spanks the A9 in almost every way.

- The Tegra supports only a single memory channel, whereas Intel uses 2-4. Each channel is up to quadruple the speed (the channel is both wider, and faster).
- Intel offers double the cache (1 MiB vs. 2 MiB)
- Intel's Goldmont architecture offers vastly superior out of order retirement and speculation capabilities than the A9.
- Intel includes two audio DSPs per chip.

But most importantly, Intel can burst for up to double the clock speed for short periods of time. If the display refreshes at 60 Hz, most of the time the processor is sleeping, which means that for almost all user interaction, the active core will be running at maximum turbo, which is 2.5 Ghz (on some of the SKUs, at least). Even when all cores are active, the chip will run faster than stated speed--at the so called all cores turbo frequency.

For common code, I'd expect the Intel MCU to be at least 4x faster, possibly more. This is evident in the huge speedup one can see in the web browser--it's insane.

 

[evster]

There have been some assertions made that the new MCU is only slightly faster than the seven year old Tegra 3. That is not correct; the Intel Apollo Lake platform spanks the A9 in almost every way.

- The Tegra supports only a single memory channel, whereas Intel uses 2-4. Each channel is up to quadruple the speed (the channel is both wider, and faster).
- Intel offers double the cache (1 MiB vs. 2 MiB)
- Intel's Goldmont architecture offers vastly superior out of order retirement and speculation capabilities than the A9.
- Intel includes two audio DSPs per chip.

But most importantly, Intel can burst for up to double the clock speed for short periods of time. If the display refreshes at 60 Hz, most of the time the processor is sleeping, which means that for almost all user interaction, the active core will be running at maximum turbo, which is 2.5 Ghz (on some of the SKUs, at least). Even when all cores are active, the chip will run faster than stated speed--at the so called all cores turbo frequency.

For common code, I'd expect the Intel MCU to be at least 4x faster, possibly more. This is evident in the huge speedup one can see in the web browser--it's insane.

Do you work for Intel or just in general have some incentive to bash the Tegra 3 as much as you can?

 

[Asterix187]

I can still heat my seats the same today as I could last week.

 

[CarbonFree]

Do you work for Intel or just in general have some incentive to bash the Tegra 3 as much as you can?

Heh ;-)

I don't work for Intel, and in general I'm a big fan of ARM. The A9, in particular, was the first ARM chip which aspired to be "server class." Unfortunately aspired is the operative word. Intel has spent a long time designing chips optimized for typical desktop and server code, and it shows. Intel still does poorly in the power department, but that's not an issue in the Tesla; the Tegra 3 was on a 40 nm process, I believe, and so the Intel chip, which is on 14 nm, likely consumes half the power at full load, if that. Most of the time the chip is in a low power state, to boot.

For cell phones, where performance per watt really matters, ARM is far and away superior. However, when you can afford to spend ten watts, Intel brings incredible performance to the table.

Well, unless you have Apple's designers. They create amazing ARM implementations!

 

[BigD0g]

Heh ;-)

I don't work for Intel, and in general I'm a big fan of ARM. The A9, in particular, was the first ARM chip which aspired to be "server class." Unfortunately aspired is the operative word. Intel has spent a long time designing chips optimized for typical desktop and server code, and it shows. Intel still does poorly in the power department, but that's not an issue in the Tesla; the Tegra 3 was on a 40 nm process, I believe, and so the Intel chip, which is on 14 nm, likely consumes half the power at full load, if that. Most of the time the chip is in a low power state, to boot.

For cell phones, where performance per watt really matters, ARM is far and away superior. However, when you can afford to spend ten watts, Intel brings incredible performance to the table.

Well, unless you have Apple's designers. They create amazing ARM implementations!

We are pretty sure it’s an intel Gordon peak processor.

 

[CarbonFree]

We are pretty sure it’s an intel Gordon peak processor.

Could both be right: Apollo Lake is a chip, Gordon Peak is a platform/board.

 

[HankLloydRight]

Semi has two I believe? Not saying it would be particularly elegant but you'd greatly expand the amount of data you can see as well as allow for upgrades.

I compute with 4 monitors.

The semi has one center seat for the driver and the screens equally off to each side. It's a completely different layout. The side screens also act as side-view mirrors.

Computing at a desk with four monitors (I have three) is very, very different than driving where one touchscreen is distracting enough.

If you're talking a Tesla designed solution with two touchscreens in the MS/MX? Never going to happen. It would be 100x easier to engineer a drop-in MCU retrofit.

Maybe some third-party inelegant solution? Possibly.

 

[Snowstorm]

Maybe so much has chang d between MCU1 to MCU2 that a simple swap upgrade is not practical. But I recalled that the actual Tegra chip and memory is on a separate daughter board, so is there an option to just replace that to get a faster but I/O compatible processes and faster / more memory? That should be relatively easier to do and would result in incremental improved performance. Is the NVIDIA relationship still intact after the shift to Intel to enable such a move? Or is MCU1 the Mobileye 3 equivalent now?

 

[CK_Stuggi]

The wiring/harness for a MCU2 would be necessary for what?
Power, display and Comms. If you just hook in the old antenna you'd still get 2.4GHz Wifi. No (extra) Bluetooth, but I'd be willing to omit that. Power supply I'd venture is the same as MCU1, as is display. Space available should be no issue. So if you were able to just exchange the PCB/core box there's no physical reason why the new MCU could not be exchanged.
Just as the modems could easily switched out, which, in Germany, made a world of a difference in usability of connected features and was well worth the 550 € they charged for it.

Now whether or not Tesla would offer an upgrade (Payable of course), is a whole other matter and for all the reasons mentioned, I sadly assume they will not.

However, a gazillion people would like to get an MCU upgrade and so far, aside from SeC employees opinions (which usually are not worth much), there has not been any official statement on it by Tesla.
That's why I was asking if there was anything official by Tesla regarding an upgrade.

 

[lunitiks]

We need images of MCU2's backside in order to see the connections. "MaxKW"'s single photo of MCU2's side (with a glimpse of some connectors) suggests that at least some connectors are moved / rearranged

 

[GatorGuy]

The wiring/harness for a MCU2 would be necessary for what?
Power, display and Comms. If you just hook in the old antenna you'd still get 2.4GHz Wifi. No (extra) Bluetooth, but I'd be willing to omit that. Power supply I'd venture is the same as MCU1, as is display. Space available should be no issue. So if you were able to just exchange the PCB/core box there's no physical reason why the new MCU could not be exchanged.
Just as the modems could easily switched out, which, in Germany, made a world of a difference in usability of connected features and was well worth the 550 € they charged for it.

Now whether or not Tesla would offer an upgrade (Payable of course), is a whole other matter and for all the reasons mentioned, I sadly assume they will not.

However, a gazillion people would like to get an MCU upgrade and so far, aside from SeC employees opinions (which usually are not worth much), there has not been any official statement on it by Tesla.
That's why I was asking if there was anything official by Tesla regarding an upgrade.

Their executive escalation team has told me and others that there will be no upgrade.

 

[viper2ko]

Their executive escalation team has told me and others that there will be no upgrade.

Would be nice if someone could get elon to acknowledge a tweet

 

[HankLloydRight]

If you just hook in the old antenna you'd still get 2.4GHz Wifi.

Nothing says the old antenna is compatible with the new MCU nor has the same connector(s) or pin-outs.

Power supply I'd venture is the same as MCU1, as is display.

We already know MCU2 has a different display, likely with different power and driver requirements and connections to the motherboard.

Why assume the power is the same as well?

I think it's also a safe assumption that the Instrument Cluster is no longer a separate Tegra computing unit and is now driven by the single Intel CPU. The old IC certainly is not going to be pluggable into the new MCU.

Basically, after 8 years of technological advancement, I would assume nearly everything changed. There's no way to drop in a MCU2 unit into an MCU1 car without significant hacking and re-wiring.

 

[CarbonFree]

There's no way to drop in a MCU2 unit into an MCU1 car without significant hacking and re-wiring.

That wasn't the proposal. The proposal was to keep the existing MCU but with a bespoke daughter module with an Intel chip. Only the daughter would be swapped. It would maintain all existing wiring, power, and display connectors.

 

[barjohn]

I don't think I have seen any definitive information as to whether the wiring is actually different in any significant way. Maybe it is, but it doesn't mean that if there is enough demand that Tesla couldn't produce an MCU2 that was plug & play compatible to address the response issues. If it adds to the bottom line and makes Tesla customers happy, why not?

 

[rnortman]

That wasn't the proposal. The proposal was to keep the existing MCU but with a bespoke daughter module with an Intel chip. Only the daughter would be swapped. It would maintain all existing wiring, power, and display connectors.

It is one thing to suggest taking electronics they are already manufacturing for new vehicles and find a way to retrofit those into existing vehicles; maybe with a wiring harness adapter or two. But now you are talking about engineering an entirely new component only for retrofit purposes. Do you have any idea how expensive it is to engineer, validate, and manufacture a piece of automotive electronics like this? And all so you can offer a retrofit to a relative handful of existing customers that frequent the forums and are jealous of new cars? For what purpose? They would have to change many thousands of dollars per unit to even break even on this. The more they charge, the fewer people will elect the retrofit; and the fewer retrofits they sell, the more they have to charge to make up for NRE (non-recurring engineering) costs and the overhead of lower-volume manufacturing. You have also now introduced yet a new hardware configuration that you have to validate all of your software releases on.

OMG, if I were running the show at Tesla there is not a snowball's chance in Fremont that I'd authorize that. No, I would much rather sell those must-always-have-latest-and-greatest forum-dwelling customers a shiny new car.

When Nvidia quits selling the Tegra 3 and Tesla has burned through their inventory of replacement units, they MIGHT engineer a new circuit board based on a newer chip. They will do the absolute minimum re-engineering required in that case.

The above assumes that it is not possible to do a drop-in replacement of MCU1 with MCU2 with a wiring adapter or two. Which seems pretty likely given what others have pointed out (particularly that the IC is now just a screen and not a separate computer.) If a simple drop-in replacement is possible, then we will find out in a few months when Tesla starts offering it.

 

[MP3Mike]

That wasn't the proposal. The proposal was to keep the existing MCU but with a bespoke daughter module with an Intel chip. Only the daughter would be swapped. It would maintain all existing wiring, power, and display connectors.

Because Tesla has so much free time to develop a bespoke daughter module for the original MCU? Really?

And there would be a limited market for them. (Less thank 300k for sure. But given the cost of bespoke designs, and then ongoing costs for supporting yet another platform combination I would imagine it would come with a bespoke price that would drive the demand way down.)