Elon Musk's latest brainfart is to turn Tesla cars into AWS on wheels EV carmaker Tesla is considering a wonderful money-making wheeze – use all of that compute power in its vehicles to process workloads for cash, like a kind of AWS on wheels. The Elon Musk-led outfit said in its recent earnings conference call for calendar Q1 that it had noticed its vehicles spend a considerable amount of their …
People of a certain age......
I am amazed to see that it is still a brand on sale.
The sad thing is how many people will be happy to pay extra for something they won't even get to benefit from. I really do hope someone proves we are all just NPCs in some giant version of The Sims. At least then you could chalk this nonsense up to some kind of programmer error or maybe they're doing testing for some kind of fascist techbro expansion pack, and we exist in the test environment.
"And now that we have already paid for this compute in these cars, it might be wise to use them ..."
No, you've paid a bit for the hardware, the user has paid a lot more for that hardware, and the user is paying for the power. I wonder when reality will register with the muskrat (or is it just that he's always looking for his own main chance)?
I think it is also when the reality registers with the users of said vehicles.
There appears to be a strange cult around Tesla that they are somehow light years better than any other vehicle on the road.
They are just a large battery, computer & crummy software in a convenient tin box on wheels.
Sadly I don't see that happening any time soon.
I'm.running with the tales of woe my next door neighbour and three colleagues regularly tell me about their Teslas amd Tesla service, then there's also the evidence ofy own eyes when i see a Tesla with a fishtank in place of lights or panel gaps with enough space to insert a hand.
Then there's one of our directors who threw his back at the lease company because the damn thing kept just powering off on the A406 in rush hour despite it having spent weeks being investigated by Tesla.
Quality control, reliability and Tesla are not close acquaintances.
Unfortunately, people will hear a story about poor build quality which then "sticks" in their mind for all time. Tesla have had build issues but then so have most of the major car manufacturers.
What people tend to forget is they they are still a new car company when compared to most of the competition, so this will be a reason for many of the build issues. My impression is that of late, many of those issues have been resolved, but of course, that tends not to be as "popular" a story!
Do I think this latest idea is a good one? Not sure, but then there is a reason why I don't have a multibillion dollar company, I don't like taking risks but clearly he does and who knows, it may pay off or it may not.
I don't have to stick with a story I've heard... I can look across the road at my neighbours Tesla, or walk past the other neighbours Tesla about 6 doors up the road.
I can SEE the panel gaps from the path as I walk past... I've been up close enough to many more and I've not seen a single one that didn't have the problem.
But ole musky apologists and enablers always have some kind of excuse for him.... oh they're a 'new' car company... give them time... 'oh that's just a story you've heard and you're sticking to it'
Try 'not' drinking the muskaid and seeing past the fanboy worship.
Tesla absolutely has worse build issues than the rest of the industry and on top of that they have a terrible service network that means that when poor build quality happens on a brand new car, people have terrible experience. Hearing stories of brand new cars spending weeks or months at a repair center just waiting for parts is not entirely unheard of in other brands, but downright common for Teslas. I have an acquaintance that bought a Tesla, drove it a day, had to have it towed to a repair center when it died on him on the highway and then spent a nearly a full year driving a loaner before Tesla finally decided they'd just refund his money and take the car back because it had such big issues they just didn't want to bother repairing it.
As to the other point, if it was because they were a new company, they could A: have learned from the "old guard" and the myriad of people they hired from other automotive companies and B: They should have resolved a lot of the quality issues by now. They haven't. They're shit as always. I'm still seeing numbers of around a 1% rejection/rework rate at end of line for Tesla, when 0,0001% (maybe a car a day) is a more accepted industry standard reject rate at the end of the line. I still see downright shit panel gaps in new cars, bad stitching on seats, etc. I don't think any of the more common issues have been resolved.
I had a 1973 Mini Clubman estate, with bored out 1275cc gt engine to around 1331cc I think.
It was well over 20yrs old when I got my hands on it.
It was also quite reliable once you fixed a few flaws.
Flaws like... water getting onto the front plugs & HT leads when it was raining too hard... Original idea for a shroud was ineffective... An old t-shirt wrapped around the centre part of the grill, allowing air to freely enter and cool the rad still... well, that worked a treat. :)
Flaws like... water getting onto the front plugs & HT leads when it was raining too hard..
Our Morris Minor doesn't do that (damp-start is a wonderful thing). What it *does* do is, when it's raining hard and the wind blows from a certain direction, leak like a sieve..
We've tried everything. New door/window seals, new windscreen seal (carefully because the heated front screen wiring is quite delicate), lots of boat sealant round the windscreen.
No idea where it's coming it. It spent a year with no carpets in the front at one point because the old carpet got mouldy. We tried covering it as well but that just made the whole thing damp and start to get mouldy.
"the old carpet got mouldy"
We used to have an Austin 1100 (steering wheel like a bus, gear stick about a yard long) that had mouldy carpets, we eventually discovered that the rubber seals in the floor panels (where the car was originally dipped) had perished and water was oozing in.
morris Minis
How dare you! BMC Mini!
The mighty Morris was an entirely separate brand under BMC (BMC was formed by a 'merger' of Austin (the hated enemy of Morris) and Morris although the brands were kept separate) BMC eventually morphed (via various mergers and aquisitions) into British Leyland (remember them) who ditched the separate brands.. Especially as the first Chair was the previous head of Austin and, under his tenure, Morris and the other brands got downplayed in favour of Austin kit. Which is why our 1966 Morris Minor 1000 has an Austin A-series engine.
Not an auspicious start - trying to run two separate brands who, fairly intensely, disliked each other and produced direct competitors of each other. Add in all the corporate infighting between the divisons, the management/union fighing and the start of imports from Japan, it's hardly surprising that the whole thing fell apart.
It's actually quite sad seeing the MG marque being slapped on cheaply-built Chinese cars. Or the Mini marque being applied onto thinly-skinned BMW Series 1 cars that are neither cheap nor mini - two of the important design points of the original cars.
"Buying a vehicle = jumping into a cult?"
Apparently so, for people that have a problem with progression!
Look, I don't agree with some of the things Tesla have done with cars i.e., removing stalks on their latest models or removing LIDAR/parking sensors. Yet, some features like the cameras are a great addition.
Not every new idea "works" but generally enough of them do which is called progress.
Though not all progress is good IMHO!
"Yet, some features like the cameras are a great addition."
What exactly do you gain from that? I'm happy with my reversing camera as I have a hard time twisting around due to an old back injury, but mirrors and a sense of caution work just fine in place of a bunch more cameras. If a wing mirror got broken, I could buy one online or nip over to the dismantlers and harvest one in short order very cheaply (my car is really common). I don't have to invoke my EE to sort out why a camera isn't working and then wait for ages to get the parts if the company will even sell them to me rather than require I take the car to one of their certified mechanics a few hundred miles away.
For a few years, they were the 'best' EV on the market... as in there... wasn't really much in the way of competition.
But now there is competition, and every other EV maker has long surpassed tesla in practically every area... especially in design and build quality. There isn't a single good looking tesla out there... not since the original roadster that ole musky had no hand in designing because it was before he bought intot he company and sued the founders to be able to call himself one.
"not since the original roadster that ole musky had no hand in designing "
Elon did toss a spanner into the design of the original Roadster by insisting on all sorts of expensive changes that increased the price and delayed the release. I learned a long time ago that it's often best to do a "good enough" job on Mk1, Rev0 and then start selling. If you want to add carbon fibre trim and lower the lintels so your wife/gf can get in easier while wearing a long dress, that's something to think about for Rev1,2 or 3. A first go at a new product will often be mired in lessons so the idea is to get to it as fast as possible and look for improvements. That's not to say that skipping thorough engineering isn't necessary, but there is a point where you say "done, ship it" and move on from there.
Teslas are as common as muck where I live so I know plenty of Model 3 owners. They all sing the same song -- "Best car they've ever owned". I have no idea why, I've yet to actually drive one (and I've got no plans to buy one). So just dismissing them as a PoS owned by cult members is probably missing something. Sure, I wouldn't expect any to be where I used to live back in inner city England ("If it wasn't nailed down it got nicked, if it was nailed down it would eventually be prised up and nicked") but now I live in suburban California....
BTW -- If you want a proper critique of them then here's some food for thought. They're quite heavy so the eat through their special tiers much faster than a normal car's tier wear. Even a minor fender bender can run up a five figure repair bill and so they're easy to write off in even a minor accident. They don't like being left out in the cold (not a problem here) but their batteries do seem to last (although for how long is anyone's guess -- its such a big part of a car that a bad battery could write off an otherwise usable car). The drive train and electronics seems reliable with very little maintenance needed compared to a normal IC car.
"Even a minor fender bender can run up a five figure repair bill and so they're easy to write off in even a minor accident."
A windscreen can be $1,200 with a 6 week wait if all of the Tesla staff is re-directed to help get deliveries done at the end of a quarter. A head light can be in the $900 range. The Model S initially had a problem with breaking the little microswitches in the door handle that would tell the car if the handle was extended or retracted. The switches are a couple of bucks, but Tesla only sold an entire assembly for better than $800 (plus the old part back). I fixed a couple and made good money doing it.
I'm fine with there being a markup on repair parts, but it can often get very silly. I've been taking care of a lot of little things on my car and it's been nice that the maker is still stocking the bits and pieces so I can repair things properly. The car is getting old so some thing like door seals will be done this year just in case. I'd still be driving my last car, but GM stopped selling door/window seals and all of the little plastic bits that dry out and disintegrate all over the place so when the head gasket blew I didn't see the cost to repair as being worthwhile.
Sounds a whole lot like the crypto miner people applying for datacenter space I ran years ago.
They don't use any power, it'll just sit there, super low bandwidth. It'll be so cheap, to run, you should barely charge me.
My question is why they didn't just put them in their garage then.
Oh, I don't have enough power at home, but you must just have loads at the datacenter you barely pay for.
The base theory of the hyperloop is sound, people have been using pneumatic tubes for around a century or longer. The problem comes in trying to scale that idea over a long distance. Trying to create a vacuum in a tube potentially hundreds of miles long is a lot different from a bank drive up teller lane of maybe a couple dozen feet.
Didn't say anything about any other aspect of the hyperloop except that pneumatic tubes are a proven concept, but one that doesn't scale well. There are a myriad of other issues with it. Like, think how much noise is made by the pump for something as small as the tube at the bank. Now imagine the amount of noise created by a pump large enough to create a vacuum in a hyperloop tube.
think how much noise is made by the pump for something as small as the tube at the bank
Not quite the same thing, but IK Brunel tried vacuum tubes for public transport with great stationary steam engines pumping out tubes. Trains were pulled along conventional rails by a piston in the tube, the attachment being via a rod exiting the tube through a slot in the top, sealed with leather flaps. It did work, after a fashion (changing tracks was difficult), but the leather flap was probably the main issue - it went brittle in the sun and in the cold and greasing it up to help only meant that it became attractive to vermin.
M.
"You forget the fact that traveling in a vacuum at airliner speeds makes these things a prime target for terrorism."
By which token, nothing in the modern world is safe and terrorism has won! The very thing that allows us to communicate over distances and engage in forum post is also posing one of the greatest threats to national security.
And I haven't even mentioned the (seemingly) unstoppable rise of A.I.!
When people start dying by the hundreds, the populace becomes agitated, calling for someone's head to roll. The politicians don't want that because the calls could be for their heads!
Airlines have adapted and there are a lot of security measures involved with air travel, both seen and unseen, which most travelers are blissfully unaware of.
I did an engineering apprenticeship when I left school... the motto in the workshop. 'If the spanner doesn't fit, you need a bigger hammer' and that was supplemented by a shout going out every now and again.... 'BUUUURRRRRNNNEEERRRRRR'
And the guy with the oxyacetylene trolley would appear shortly after to remove something that a hammer could not.
Doesn't that depend on how it got stuck?
If it slowly ground to a halt on spilled beers and wines and an assortment of cheeses, or on the rodents eating said spills, perhaps the deceleration would be survivable.
Until the air ran out. Or the vending machines and everyone reverted to cannibalism.
Have we considered the air? And, perhaps, some jolly pranksters with helium, sulpher dioxide or nerve agents? Or those who eat loads of beans before traveling? Or those delightful folk who live in torrid climes and have yet to hear about showers? Or laundry.
Hyperloops are potentially *lots* of fun. :)
Good point. Department stores were using vacuum tubes to shoot money, and change, back and forth, what, a century ago? Large companies had a similar system for internal messaging. Totally taken the p*ss out of in "Brazil". The limits get kind of obvious when you imagine one moving, oh I don't know, even one person around.
Not simply to be contrarian, but also to share knowledge! The pneumatic tubes for department store/banks is a different concept: A vacuum is created in front of the transported cylinder and atmospheric pressure pushes the cylinder from behind (also the vaccum is very slight). The idea of the hyperloop was that the whole tunnel would be sucked down to as high a vacuum as possible, and the 'pod' (or whatever it is called) has to drive itself via linear motors.
Could we tell Old Musty that the idea would work on the Moon?
It could even be a monorail, above-ground, so the passengers could see some scenery. With, perhaps, underground parts avoiding critical structures.
Musty is possibly malleable enough to see this as a good idea and building it would mean we'd need cities to serve as stops and transit points.
Lunar civilisation built because a bunch of billionaires want to one-up one another with a truly stupid notion?
> Futurama did it first?
Futurama is often a loving homage / piss-take of older sci and sci fi ideas - being first isn't its point.
The idea of carrying everything in tubes has been a sci fi staple since before the Golden Age. In New York from 1897 to 1953, 27 miles of pneumatic tubes were to deliver mail:
https://en.m.wikipedia.org/wiki/Pneumatic_tube_mail_in_New_York_City
An earlier system was trialled in London from 1863, but was unsuccessful.
"Pneumatic tubes work" is hardly enough of an idea to be dignified with the term "theory", I think. Hyperloop is a terrible concept, and not at all a novel one; it was a staple of early science fiction, going back at least to Verne. It's parodied in the opening credits of Futurama, after all.
There are reasons why we don't have a lot of pneumatic-tube subways, and if Musk were a decent engineer, he'd have spent a little time reviewing them before throwing money at a ridiculous project.
"The problem comes in trying to scale that idea over a long distance. Trying to create a vacuum in a tube potentially hundreds of miles long is a lot different from a bank drive up teller lane of maybe a couple dozen feet. "
The Hyperloop and some HSR is pure hype. I'm a big fan of trains and understand the exponential cost of going faster past a certain point. It appears that ~125mph is the major break point. We can build lines to a 125mph spec and while more expensive, it's not prohibitive for many runs. To get to HSR speeds, the distance needs to be long enough between two important points yet not so long that flying and putting up with all of the "not flying" carp isn't too much of a delay to make it effectively slower. There's also going faster by slowing down. HSR in seismically active California where the land moves over time is underway but compromised down to the two end points being nowhere in terms of passenger needs and roughly parallels existing service that could have been upgraded with dedicated passenger tracks to minimize delays waiting for freight traffic and even sped up. An upgrade could have been done much faster. There's also the only US private passenger company, Brightline, that's going to offer service between Los Angeles and San Francisco via an overnight service (the two promised end points of HSR) over existing tracks. Depart at 9pm, arrive in the morning on an all-sleeper service. I think when I worked it out, the speed of the train would be around 50mph, pretty slow in this age, but cheaper to operate, less sudden movements on older tracks and plenty of margin to accommodate hold ups.
Putting a train in a vacuum tube to get more speed will work, it just won't be financially viable and the uptime is going to be horrible with plenty of incidents. There's also little point to striving for the added speed since modern electronic communications have cut down on the need to travel for business as much as in times past. When you need to travel for business, it's important enough to take the time for it.
Muskrat pushed the whole hyperloop idea for one reason only... to delay and try to thwart Californian laws on transport (especially public transport)... that was the entire reason for.
But the idiots with money to throw at stupid ideas bought into it
I'm not saying the 'concept' of a hyperloop is stupid... but there are far better and cheaper options that have been around for decades or centuries that are proven to work... such as trams, high speed rail and so forth.
I've a whole stack of compute power in the house. The thought of putting it to use, and potentially even getting a profit from it, has crossed my mind before. I lumped in with Team Reg on the Folding@Home drive during covid. But the fact of the matter was that, even with actual computers, permanently powered, it was generally more hassle than it was worth. I had to pause the client whenever I needed the machine for actual use, and I gradually just stopped turning it on again. When I replaced that machine, the Folding@Home client was never reinstalled.
So, on a machine that's pretty close to an optimal use case for distributed computing, by a guy who's interested in the concept, the idea still ended up being impractical. Add in the fact that the machine is now embedded in a car, with a questionable data connection, questionable ownership / control of the software and a questionable effect on the battery life of the vehicle, and you may be forgiven for coming to the conclusion that this is an idea not worth pursuing.
Let's not even think about what might happen if the car's computers' response times suddenly take a dive while driving because this "distributed computing" module decides to kick in at an inopportune time. Let's take the time to get the bloody thing working right before we distract ourselves on tangential uses, eh Tesla?
I was always confused by the F@H client as to why it insisted on keeping the workload ready to go when I wanted to use the computer.
Presumably Elon would automate completely shutting down the workload when you switch the car on and loading it or a new workload when you switch it off again? It's easier to tell when the compute is needed for the main purpose in a car than it is on a PC, or especially on a server.
I installed BOINC on a few machines during Covid. It's still running on a couple, with various projects active. With BOINC it's possible to specify how much compute power each machine "donates" so on one multicore machine, for example, I've told it it can only run 90% of the time. It seems to do about 5 seconds of 100% and then pause for ½ a second (just a guess). Doing this means that the processor fan rarely ramps above normal. In normal use (web browsing, word processing, spreadsheets, photo editing) BOINC just keeps running and doesn't seem to affect the computer's responsiveness at all. It occasionally pauses itself when I use the video editor and it pauses for longer when rendering video, and I find it's useful to pause it manually when doing large updates.
It's possible to run it bare metal or in a VM and under Windows some tasks can run on a graphics card (not found a way of doing that in Linux). The best bit is that the BOINC client supports a large array of projects so my computers have been working on Covid vaccines, climate prediction, identification of cancer cells and several more projects. It's even possible to run it on Android phones (though I hate to think what it does to battery life) and Raspberry Pi, even models with 1GB RAM (though it needs virtual memory in that case), albeit slowly, and if a particular computer is not suitable for a certain task, that task is not allocated.
There are loads of other citizen science projects around if Folding@home or SETI@home is not to your taste though some of them require a considerable investment!
M.
Only 8-hours commuting!? Lucky!
Yeah - some days it takes me a whole 8 minutes to get to work!
Mind you, that's when I actally go into the office. A normal day it's about a minute (that includes "do not step on the always-underfoot dachshund pup" pauses and the obligatory "push the black cat off the keyboard before waking up the Mac" time).
Working from home. It's the way to go.
He might - shock I know - have thought this through, and only have the cars do the computation when they're connected to mains power.
Assuming that when the car is at home it's on-charge, and not when at work, then being available for about 2/3 of the time is in the correct ball-park
Will it earn Tesla $30,000 dollars per year per car?
Just like the Robo-taxi? LOL. I mean, when the cars are that profitable post sale, why even bother selling the cars in the first place and just keep them for yourself along with all the annual profit, becoming the world's biggest global self driving taxi firm?
That's what I'd do, if it were even remotely true. But then I'm not a tech genius.
Hmm ...
I sometimes take a taxi from the Town Centre to my home. Ten minutes or so at a cost of about £10 or one pound per minute. Assuming you could average 50% usage on a taxi over the year, with it charging and doing distributed computing projects the rest of the time, that's about 60x24X365/2 = £262,800 per year. That is well worth having and more than the cost of the car, charging and inevitable random fixes.
It could be that Tesla-taxis may charge less due to them not having a human driver to feed or licence or they could possibly need that slack to cover lawsuits so let's run with 200,000 per car, per year.
With, say, a million cars, that's an income of about 262,800,000,000 per year, not all of it profits. That is slightly more than Musty owns at present. A million taxis isn't a lot if you scatter them over the entire planet and serve the entire population. If they become cheaper to build, cheaper to run, cheaper to hire and far more numerous, the profits on such a fleet could be astonishing.
If Muesli has ever been in a taxi and paid for a journey (is that likely?) he might have run those simplistic numbers. They are very attractive numbers. He might even think that he could stop *selling* cars and instead use them as taxis.
Then he could pay for his rockets to Mars.
Theft Act 1968, s.13: "A person who dishonestly uses without due authority, or dishonestly causes to be wasted or diverted, any electricity shall on conviction on indictment be liable to imprisonment for a term not exceeding five years."
And that's not including additional wear on batteries and (particularly) flash memory, inflicted by all of this. Presumably, Tesla would be responsible for network charges, so there's that, I guess.
If Musk is serious about this, he'd better have a really good deal for owners in exchange for having their cars used as four-wheeled data centres, and a cast-iron opt-out if they don't want it. He'll also need a very strong way to ensure client data can't be intercepted and observed by tech-savvy car owners, because it's going to have to be processed "in the clear" (unencrypted) by the car's processors. (Will they be TEMPEST-shielded? I think not.)
> it's going to have to be processed "in the clear" (unencrypted) by the car's processors.
I don't know enough to judge the the current state of homomorphic encryption efforts, but the remote processing of sensitive data is an area of active research:
https://en.m.wikipedia.org/wiki/Homomorphic_encryption
Sure, but so far no one has worked out a good way to encrypt data so that it can be worked on by an even slightly untrusted computer, and not waste a lot of compute power on just the "keeping it safe" part.
It's always easier to compute things that don't have to remain encrypted during the process.
As someone once mentioned, a trusted computer is (also) one that can break your security policies.
The idea that running a workload on an otherwise idle CPU is "free" needs to die. It consumes power. Very likely, more power than you'd expect, because if you're doing this thing, your workload is going to be some highly-parallel crunch that maxes out the CPU in a way that regular usage doesn't even approach. Power is not free.
Do it for charity, for some research project you believe in, but don't do it to give someone else your money.
The fact that it runs off a battery, a component that is subject to wearing out with usage, is just adding injury to injury.
It sounds a nice idea, but it's not something they designed the compute to do. (I'm assuming they're talking about the autopilot computer, which has meaningful compute power)
1) The autopilot computer is watercooled, which seems neat and efficient if you're driving the car and need to temperature control drive motors, battery packs, passengers, etc. But running the big waterpump just to cool the compute is a huge power overhead.
2) Waterpump and cooling system wear and tear, was it designed for 24x7 use? Is Tesla going to provide an extended warranty for those who (presumably) opt in
3) Flash use - was the MCU or the AP computer designed with the wear cycles for getting a constant stream of new data to crunch, or was it intended to receive an updated model about once a month (I suspect the later) and is Tesla going to provided an extended warranty
4) Charging - those with home charging might not be particularly fussed if their car chews through battery while sat on the driveway and recharged overnight. Those without home chargers, or away on the road may feel differently... I'd say charging cycles on the battery is unlikely to be a huge issue
5) Age of chips. Yes, they're mighty fine computers, but Telsa refreshes only about every 5 years, is the 5 year old AP chip in a 2 year old car actually worth the energy cost?
There might be some who would happily accept this for a small cut, but it's definitely one of those cases where owners would need to be able to opt in/out, and I doubt much serious thought has gone into the idea, the aim is simply to distract from the declining sales.
Steve 53: please stop calling it autopilot. It’s not an “autopilot”. A mate of mine paid 7K euro a couple of years ago just for the option on his Tesla, he believed Musky McMuskFace when told that “full self-driving” was only a few years away.
Other than that : you made the effort to actually state possible issues. Why ? Based on Tesla’s past “roadmap” it’s pretty clear that they want to boost stock price with unfeasible announcements. Or maybe they’re just looking for ways to finance the manchild’s payout…
Understood Steve. And your argument was very coherent, I learnt today that Tesla’s guidance systems are somewhat inefficient regarding power usage ;-)
Then again, I have no idea what the same features on my car “cost” regarding power consumption. When I got it I was mainly focused on trying to dial all the privacy-related knobs to zero… And I’m still not sure I found them all.
Cool.
They're actually a pretty efficient way of doing things when driving.
The octovalve is a level of over-engineering that other companies can only dream of. It's extremely good at shunting heat around, eg in the winter the heat from the AP computer, motor inverters, etc will end up being dumped in the cabin and/or battery.
The problem comes when you want to run the compute on it's own, for example when in sentry mode or when charging (That has a 300w overhead as well, because it needs to run the waterpump to cool the onboard charger).
And, privacy settings on a Tesla, I mean, there are settings available.. But if you want things like... navigation.. to work, you're going to to adjust the privacy settings.
>>Autopilot is a fine descriptor for a basic system
The majority of the public believe an autopilot to be a black box that flies the aeroplane and handles everything.
They apply that thought to anything labelled "Autopilot".
There is, therefore, no wonder that Teslas with "Autopilot" are crashed by meat sacks assuming the software was going to handle everything, when, in reality, the "Autopilot" is just a cruise control with collision mitigation (providing conditions aren't too complex) and lane keeping bolted on.
I humbly submit that "Autopilot" was a terrible descriptor for the system.
"Autopilots are very stupid, and only work in aircraft because so much of the flight is both very boring and the risk of collisions is very small, since the enforced separations keep planes apart."
The modern versions of the FMS (Flight Management System) can do some simple flying of the aircraft so if you have a crossing restriction (altitude and speed) at a nav point, the FMS will work out at what point it needs to start descending and manage the throttle. This frees up the pilot to "stay ahead of the flight" with things like programming in the next radio frequency and monitoring the weather ahead and at the destination. The system may also chime to tell the pilot that it's taking the rest of the flight off and they need to do some real flying from there down to the ground. Yeah, it's advanced cruise control, not anything too smart.
"4) Charging - those with home charging might not be particularly fussed if their car chews through battery while sat on the driveway and recharged overnight. Those without home chargers, or away on the road may feel differently..."
Elon was banging on about 1kW of computing power in each car and invited us to cast our eyes towards a glorious future, but that 1kW doesn't include the electrical overhead of just the car being "on". Teslas already have some of the highest parasitic electrical usage when plugged in which really shows in the US if you try to use a granny charger (good luck with that). If I'm being pessimistic, I'll call it 2kW total draw. If you only have 6kW to route to the EVSE (charger), that's a full third of power going to these compute tasks rather than charging the car. Over the course of a month, it will be really easy to spot the increase in the electric bill.
Not really, the compute is about 300w*. A typical domestic charger is 7kw
* Approximate draw when running the compute stack for sentry mode, which means running coolant pumps, AP compute and MCU. Actual AP processor is reportedly about 72w, although it's hard to verify that from decent sources, or confirm if that's for both of the processors or each.
....it just needs to be done right.
#1 Fair. If power and compute is being used, then equitable compensation is warranted. This could be as low-cost to Tesla as giving away their self-driving license in exchange. To balance out the extra 'wear and tear', toss in extended warranty coverage on hardware and significant discounts on battery replacement.
#2 Transparent. The amount of power consumed is reported to the vehicle owner. They know exactly what their costs are.
#3 Opt-in / Opt-out. This is off by default, and no sneaking it in to tricky legal documents either. The vehicle owner can fully and completely opt-out at any time with no penalty.
Anything less, and this idea is dead on arrival.
"Example #567287 proving that Xitler doesn't know WTF he's talking about."
For those with some schooling, we know better, but it's fine for hoaxing the yokels. (reference to J. Harshaw)
A lot of what Elon says is hard to parse through the blizzard of stammering and incomplete sentences. Compare the full recordings with the de-stammered versions that Common Sense Skeptic post. On the last earnings call it did seem like Elon had memorized some rehearsed phrases, but would quickly go off the tracks immediately after them. Compare all of those with how polished Steve Jobs was on stage. That wasn't pure talent, he spent time with a team to nail the words and the timing. He didn't improvise or talk about anything that wasn't in the script.
"Elon has a habit of throwing out wild ideas when things aren't going well to distract the punters and energize investors."
Well all I can say is if I was an investor and this is the best thing Musk can come up with to improve Telsa's current situation, i'd be selling my shares now and get out before the bubble bursts, as hes clearly a snake oil salesman.
in deflection... This master is currently on trial in NYC. Just look at how he deflects from how badly his case is going by whining about the temperature of the courtroom. He has an overcoat... why not wear it? Oh wait, it won't look good for the cameras.
Like his comb-over image is vitally important. Gloss over substance.
Both want to be their own Big Brother [see icon].
Tesla would be effectively using electricity that the car owner has paid for to run any workloads while it is idle, so would they get a cut of the money generated?
Yes, it seems. CFO Vaibhav Taneja, said "the capex is shared by the entire world. Sort of everyone owns a small chunk, and they get a small profit out of it maybe."
Note that weasel word.
...that "watts" is not, cannot be and never has been a measure of computational power?
"I think you could have on the order of 100 gigawatts of useful compute, which might be more than anyone"
This is beyond gibberish. I have a 2 kilowatt electric fire with an analogue mechanical thermostat. How much computing power does it have? My laptop pulls at most about 100 watts. How much computing power does it have? Suppose I have a 60KWh battery in my EV. How much computing power does that mean the car has?
El Reg are apparently just as ignorant:
"Of course, all this compute capacity isn't sitting conveniently clustered together in a datacenter"
What "compute" capacity? All we've quoted so far is electrical power draw, which means absolutely nothing.
《This was my thought exactly when i first read this, glad to see you voiced it. I thought computer power on this proposed scale would be measured in giga flops or similar, not wattage》
I was biving musk prat the benefit of the doubt (silly me) as his envisaged workload was AI/LLM which probably has it own lexicon of bullshit and assorted twaddle.
LLM workloads probably don't come down to convenient FLOPS, MIP etc - I believe parallel ops on vectors of rather short floats are more the thing.
So it would not surprise me if a member LLM fraternity afflicted with moment of lucidity developed an equally dubious measure from information theory, statistical thermodynamics that fiddles with entropy and free energy that give a magic number with the units of joules/second.
As an aside I guess you could assign the same measure to the human brain. :) A fair number of pico-watt brains out there and now I can see why the quecto prefix was required.
" I thought computer power on this proposed scale would be measured in giga flops or similar, not wattage.."
He could have used COWFLOPS, Complex Orthogonally Weighted Floating Point Operations per Second. (I can't remember where I remember that from since I filed off the serial numbers long ago).
Specified benchmarks are the best we can do.
Compute energy: quantity of benchmarks completed
Compute power: benchmarks completed per time
Compute efficiency: electrical energy consumed per benchmark completion
It's all a bit hit and miss- much like a Tesla.
You're missing the point that this is not a serious suggestion. Tesla is trying to boost its value by suggesting that the cars contain hidden assets. The solution would be impractical and probably illegal but it makes Tesla sound like a leader when it comes to AI. In fact, it's years behind other car makers who've been collecting data from all their fleets for over a decade.
"Tesla is trying to boost its value by suggesting that the cars contain hidden assets. "
It was a poor attempt. He could have said that the car could manage a smart home and since the car is always connected (big stretch), that no matter where it was, it would have a complete inventory of the fridge updated every few minutes so you could have a constantly updated shopping list on your phone when you pulled into the car park of the grocery store, fish monger, butcher. If you programmed the sat nav for a pub, it would phone ahead to make sure a pint of your favorite was sitting on the bar waiting for you (some limitations apply). A small group of marketing dweebs could come up with stuff like this by the skip full on a daily basis. None of it very worthwhile, but it could sound great. Everybody want to save time and not be wasting time shopping when they could be home watching the latest drivel on their 75" big screen TV.
El Reg hasn't gotten it wrong. There is compute capacity, and they haven't attempted to quantify it. We know that the total of all the computers in Tesla vehicles everywhere would make a relatively powerful parallel computer. Thus, there is capacity. The only problem is that the computers aren't Tesla's to sell, you can't actually use them as efficiently as you could a set of servers in a datacenter, actually trying to use this would put a lot of wear on parts of the car, the power usage would be much less efficient and would lead to problems from users who don't need to waste their electricity on this, and about twenty more things. So El Reg isn't wrong to say that there is a lot of compute capacity, nor are they wrong to explain why it isn't as useful as the raw compute numbers (which they don't have) would make out.
I upgraded my ‘87 Dodge with transistorized ignition + computer throttle body injector, I wager that Betsy’s probably sporting about 12% more raw compute power than NASA sent with Apollo moon missions!
If Tesla refuses to include Betsy in their stupid compute farm, I humbly suggest they can go suck an egg !
Boom! I knew someone would have remembered SETI @ Home. Elon is yet again inventing something that has already been done. IIRC a cancer charity in the UK also had a screensaver that used local processing to train image processing algorithms to spot cancer. So harnessing spare compute is nothing new. It's just a bit complicated and not that reliable.
"Would this be even legal? The vehicles, computers in them and the power used would be the property of the owner of the car, not Musk?"
It would be easy enough to sign that away through a "work made for hire" contract clause in the EULA you sign when you get the car. As long as they don't insist on exclusivity, they can avoid copyright issues.
"Because it's not the company's car for a start."
Ah, Ah, ah, you think that but is it true? Tesla was one of the first companies to have their cars so entwined with the mothership that if Tesla were to fold, a whole bunch of what makes a Tesla marginally useful would also go away. No Superchargers with their proprietary plug (in the US), no app, no satnav updates, no service since the docs are not readily available and they don't have a big third party parts supply as many other brands. Somebody like Ford will use all sorts of parts over and over again in many different models all over the world that a company such as Dorman who specialize in replacement parts will carry reasonable quality knock-off parts. That's not to say that Ford isn't copying the worst traits of Tesla with their bespoke screens in the cars and things such as that.
I bought a Tesla in 2016. Even then the company's figurehead was notoriously mercurial and prone to megomanaical claims. "Full self-driving by the end of the year', anyone?
But, do I want 2024 Tesla to have access to the computing equipment I bought from them and paid for? No, not under current leadership and not without some serious indemnification.
Why not?
1. Their onboard computers (Linux) log too much s***t into /var/logs/whatever, and therefore wear out their SSDs faster than they need to. They've fixed this they say. But it has caused field failures. This is huge hurting hunk of metal, not a game console. It's drive-by-wire and the wire better freakin' work. Every time. It has one job, keeping my family safe. Please don't use my expendable parts (SSD) without working out a schedule to proactively, and for free, replacing them.
2. The cell network (LTE in my case) rig in the car doesn't have much bandwidth. Certainly not enough for ML tasks. Barely enough to show me maps.
3. Power costs money and releases carbon. Don't use my power without my permission.
4. Cybersecurity? Safety? Presumably they would push native code. If they can push it, so could a cybercreep.
Now, maybe he's just saying provocative things. Hope so. But he needs to be careful not to let his loud mouth damage all his household brands. For the customers of household brands have more power than the most billiony billionaire.
"2. The cell network (LTE in my case) rig in the car doesn't have much bandwidth. Certainly not enough for ML tasks. Barely enough to show me maps."
The $200 Android head unit I recently put in my car has the maps onboard and with WiFi, I can apply updates when I'm at home. I didn't pair my phone as I see that as a big gaping security hole so it's not using the cell system for anything. I could use my phone as a hot spot if I wanted to use the larger screen for something rather then the tiny one on the phone, but there's no need for that full time.
The SSD is buried in the tear downs I've seen. Maybe they've fixed that, but it shouldn't be any more difficult to replace than a cabin air filter in a Toyota or Honda. If they are also standard drives, users could proactively replace them on a schedule and upgrade them as well as prices come down.
There's more and more interest in hacking cars at the "Cons". I'm hoping to have the time and money to attend DefCon this year. They've been hosting a car hacking village and it's been very interesting. Rooting a car to use it as a crypto mining rig would show so that's not viable, but people don't even think twice about pairing their phones with the car and also doing their banking on the phone, one click shopping and all sorts of other stuff. Teslas can be a big target since clueless ViPs have them and figuring out how to download all of that pair information, Nav history and other stuff could be very profitable. If Tesla puts driver monitor cameras in the cars, it might be interesting to see what certain people might get up to in their car especially when 'parked'.
Since ViPs often feel they are above the law, they should consider that the data their car stores can be subpoenaed and used against them. Makes me glad my car is so old it was carved out of a block of stone and doesn't have any data stores. There's little I can do about CCTV and number plate readers, but I can avoid ratting myself out.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
