The bigger reason is just the nature of automotive development. Every part is developed by a specced or sourced by different teams and outsourced to a different Tier 1/2/3. Things are designed to be modular, so if you pick Option A for something, it might use ECU A, and if you pick Option B, they might pack a whole different ECU B. Also, every ECU is costed down to only support the intended application, so ECU A likely doesn't have the extra bandwidth to add on extra features for Option B.

Also, none of this is decided or developed at the same time, so you have all the different features and ECUs developed throughout. Some things are changed mid program. Some things carry over from previous programs, so it's usually just easier to go your own way and not work with other teams on combining features.

Every automotive OEM has a person who comes up with the brilliant and cost saving idea of combining ECUs. I know 4 such people from different OEMs but they have all failed for some of the reasons mentioned above.

You don't build semiconductors for that purpose by using 10nm process then adding a lot more redundancies when all you need is a simple opamp. You instead build the thing beefy and resistant to the environment.

And all the excitement over new chip plants by TSMC etc, won't do a thing for these automotive applications.

That is basically the operating range of the average graphics card these days. Not many people go sub-ambient with their cooling but no graphics card would complain if they did. The actual silicon is perfectly fine with such temperatures. The only real issue is the external cooling rig, a classic engineering problem that any car company shouldn't have a problem solving. A water cooler tied to the vehicle's coolant loop would easily cap the upper temperature range at water boiling points. A car engine has plenty of power available for fans or even air conditioning if necessary. Hide the controller chips somewhere in the passenger compartment and the humans will die of heat before the silicon.

Also this means, no solder joints or any other component won’t act funky.

A transient error in your GPU is a one pixel blip. In an ECU, that’s a power loss event in best case.

(This is also done for fire safety reasons. Any metal that could possibly short if wet could also ignite a fire during/after a crash.)

B6 Audis have a design problem with their sunroof drains. When they clog up, unknown to you, the water will pool in passengers footwell area. It takes a while to notice this as a person, as there is 4-6" of foam padding between the metal floor of the car, and where you rest your feet . The TCM lives between these layers, and eventually, in the lake that will form, submerging your TCM -- leaving you with yet another reason to hate your Audi.

This is early 2000's. Late 200* a8 models had I believe 30+ different "computers" living in the same over metal under foam situation.

An aside, if you ever wash the engine bay of a car with low pressure water, I'd suggest covering the blatant electronic bits with some plastic bags, and leave the car idling when you do so -- that way if/when it starts to sputter or just dies, you have a general idea of where it all went wrong.

They really are not as water tight as we all assume.

This is early 2000's. Late 200 a8 models had I believe 30+ different "computers" living in the same over metal under foam situation.*

I now have secondhand PTSD from reading this, and thinking about the Audi owners who have found out the hard way.

Dead leaves caught in your sunroof can make your transmission flake out. I'm sure the owner's manual, on page 935, instructs you to ream out the drains with a pipe-cleaner every 5,000km so that your car can shift properly. The only word I can think of for this is..."perverted".

Story time:

Around that time they saved a few bucks by switching the thermostat housing from metal to plastic and not changing the design at all so that it could be a drop-in replacement. With that distinct lack of engineering effort, they were prone to splitting at corner-induced stress lines in the plastic.

In an ordinary car this wouldn't be an issue; thermostats fail regularly, so their housings are readily accessible -- kind of like how most cars make changing the oil filter a manageable task. In an Audi however, the process is to take the front half of the car off to expose the engine, take the front half of the engine off, and then replace the now easily spotted housing.

So anyway, I don't know of a better place for the thermostat housing to split in half (leaking coolant out faster than you can pour it in) than the middle of fucking nowhere WY. By this time mobile phones were pretty common, but I swear to God they were sharing a single 1Mbps connection across ten counties; only about every third Google search would succeed without timing out, and as a precursor to learning how to take apart my car it was quite literally faster to code up a crude strip-the-garbage proxy (paginating compressed text and nothing else to send back) and deploy that to a home server than it was to try to surf the modern web over at best a 1k connection.

In retrospect the coding wasn't necessary since speeds increased to something manageable in the wee hours of the morning when everyone else was asleep and since it was absolutely not a 1-day job, but it was a bit of a fun experience anyway.

The plastic coating though, I have seen. Such coating completely changes the thermals of the electronics making the OPs claims that a home GPU could just be strapped into a car very unlikely.

Well, it takes some parameters, goes into a lookup table and decides how much timing to retard, or fuel to give, and probably has a o2 sensor, checks knocking, etc. All done on pretty barebones 80's tech.

And, as a person who's doing simulation coding, I tend to behave to him about that.

The relevant part starts here: https://media.ccc.de/v/32c3-7331-the_exhaust_emissions_scand...

Lets not exaggerate :) ECU reads a bunch of sensors, does some mild compute, uses that to do a table lookup, interpolates and slaps the result into actuators. 8bit 10MHz is enough to accomplish this task.

It's same with automotive industry. You have a well defined and certified stack from Delco, Bosch, Delphi, etc. and you can trust that hardware. Integration makes you return back to square one.

I remember Toyota tried unifying some control units into a single box, and 5-6 years considered realistic if not a little optimistic.

The old practices are just not good enough.

Falcons get refurbished after short flights. That reduces the emphasis on durability. Redundancy defends against random errors, reducing the need for radiation hardening. It does less against wear and tear.

True, but also think about the path dependence of the old companies. They have withered a lot of ups and downs, they are not the innovators, they are the very slow incrementalists that despite this release a new boring model every few years dressed up as the new best thing ever. (If you have seen a car from one manufacturer you have seen them all from them for the past decades too.)

The whole car industry is a relatively high volume & low margin & medium risk business. (The unit economics is great, but any risk kills profitability, so there was a lot of consolidation and convergence.)

Basically the car industry is like the "iphone industry" except there's some actual price competition and worse fundamentals, plus Apple is relatively young (and Steve set up a pretty high pace, and expectations, plus the fundamentals of the components - eg. semi industry, radios, displays - are really moving forward, whereas internal combustion engines, and the material science of other car components are not).

On the flip side, cars now tend to run more efficiently till they break fully and go into limp mode or won't run - a lot less tuneups and tweaking, no adjusting the valves and carb as the car ages

Went to my service guy, he put down his tea on my engine, replaced it and his tea was half (~10 minutes). Also put in a new set of plug wires since my set was ~18 years old and gone hard at that point.

I went to office, lost 30-40 minutes at most.

The second one happened when the car was at park, at night. Started it in the morning, it was running off, so went to service. Same story. 70Km/h speed, 10 minute change, sans the cables this time.

Newer cars tend to use one per plug. It's just a disconnect, pop-off, push in, plug, go. 2-3 minutes at most.

One interesting tidbit is the coil packs run a bit of current over the plug when it's not sparking to measure ionization to determine engine conditions (i think mostly knock)

That’s interesting. My coil pack is also integrated, but it’s designed to fail independently. Maybe it’s the design of the coil, but not an inherent shortcoming of the system itself.

> One interesting tidbit is the coil packs run a bit of current over the plug when it's not sparking to measure ionization to determine engine conditions (i think mostly knock)

Didn’t know that. I knew that engine knock was measured with vibration sensors, crank position relative to other things and engine output measurements via load sensors (irregularity), but ionization measurement is new to me. Have any sources, so I can read more?

Yeah... no standard knock sensor, which is the normal way to do things, but it's much improved because a regular knock sensor on a 900 can't really tell, what cylinder is knocking (and no cylinder behaves the same, regardless of what you do), so on a distributed 900 with the Bosch systems, it'll hold boost back or (with ezk) retard knocking for a few full cycles.

With ionization it can determine -which- cylinder is knocking and adjust timing and fuel on that one specifically

Here's a good article

https://eeuroparts.com/blog/tech-corner-saab-trionic-and-ion...

What other thing should it have? An ICE? Then Nissan Leaf, Ford F150 electric, and BMW i series are not cars either.