[kEiThZ]

My 2006 may not make it to 2022, but I'm definitely considering the used beater option rather than buying new. Or lease for the gap.

- Paul

If I were in your shoes, I'd lease/buy a 3-5 year old used car and ride it to 2025, when there will be fantastic EV options in Canada and charging networks have substantially improved. You won't get the subsidy then, but prices will have improved enough that it won't matter.

 

[mdrejhon]

I've said this before. A lot of the legacy automakers don't realize how much the game has changed. Something like half of Tesla's employees are involved in software in some form. It's single digits for legacy automakers. Recruiting and developing that kind of talent takes years. There's going to be some automakers that will never be able to catch up and make the transition.

Nikkei Business Publications -- a well known japanese financial publishing organization -- did a recent teardown of a Tesla Model 3 and they found electronics were 6 years ahead of a Japanse automaker (Toyota) and an European automaker (Volkswagen).

Tesla teardown finds electronics 6 years ahead of Toyota and VW

Self-driving AI sends shivers through traditional supply chains

asia.nikkei.com

One stunned engineer from a major Japanese automaker examined the computer and declared, "We cannot do it."

 

[mdrejhon]

<Side topic>
The lithium batteries that will be a big part of our transportation lives sooner than expected (thanks to Crosstown LRT, upcoming buses/semitrailer trucks, etc, even if you never buy a BEV).

One concern I have about the lithium battery economy scale up is it's going to be a quite a big mining boom worldwide -- some of it destructive -- the ugly underside of BEVs -- the environmentally-damaging lithium cycle. But there are multiple paths to getting lithium.

We'll need to massively scale up our lithium battery recycling quickly -- European pilot plants are aiming to achive 90% automated reycling efficiency for these modern lithium batteries. That can allow us to get 10x as much batteries from the same mined lithium!

Also, lithium is super-plentiful -- ocean water is 0.1% lithium and some lithium mining is already being done with brine (former saltwater that has higher lithium concentrations). Even the discarded brine of reverse osmosis is already now become a new lithium-mining source too. Eventually, processes will be developed with a much cleaner lithium mining cycle.

Despite the controversies of Tesla and/or Elon, and any missed deadlines they had, they are rather so far ahead in their competitive advantage at the moment. Tesla and its hunger for a seismic-scaleup of lithium battery production -- is single-handedly responsible for the freefall plummet of lithium battery prices -- that made Metrolinx cancel the gas peaker plant and get a battery farm for the Eglinton Crosstown LRT. Everyone worldwide really noticed the big Australia Gamble (In 2017, the world's first 100 megawatt hour battery was built built in less than 100 days "or its free"). That battery wildly succeeded far beyond expectations (paid its capital cost very quickly -- it's fully paid off its capital cost today -- and is now expanding to 150 megawatts. That giant publicity stunt gave the utility-scale battery industry a giant headstart and has also had local repercussions.

The Metrolinx battery, at 30 megawatt hour, while small by today's standards would actually have been the world's biggest battery just five years ago. Right now it's just a tame size, with the Buzen 300 megawatt-hour battery being built. Gigawatt-hour batteries aren't far behind.

We'll need to clean up the lithium manufacturing cycle in a methodical way, in a very MAJOR way.

Much like 1850s coalgas versus today's natgas. The well-off people in 1850s+ who could afford piped gas home illumination -- polluted much more per cubic meter of the ancient "natgas", than today's much cleaner natgas. The lithium cycle may be horribly dirty today but I'm looking forward to seeing the lithium cycle becoming much cleaner very quickly -- our planet needs it.

</Side topic>

 

[rbt]

One stunned engineer from a major Japanese automaker examined the computer and declared, "We cannot do it."

Context is important, as Tesla cannot do it either.

10 million of something per year is far more challenging of a supply chain problem than 500k of something per year. Tesla's manufacturing capacity is constrained largely by parts; not their factory line throughput (they could add more shifts if they had parts).

Toyota's discrete video chip requirements would be equal to Dell Computers; that manufacturing capacity would take a large number of years to build up. Perhaps even more challenging, a number of their existing components suppliers may go bankrupt before Toyota eliminated manufacturing of current lines.

Tesla has good tech but it's their supply chain which is still very much in development that is unique.

 

[mdrejhon]

Context is important, as Tesla cannot do it either.

The other side of "context is important" is that this is a very advanced neural-GPU much more complex than an Intel and AMD processor.

Tesla was using an NVIDIA GPU but they've now created an in-house AI-optimized chip similar to a very advanced GPU, but without the graphics. And there's now a pair of them in all new Teslas being made. Much of these chips are stamped with a Tesla logo, though understandably they would be dependant on the fab throughput (whether it's TSMC or other fab). Nontheless, in any list of top 10 (or even probably top 5) most powerful neural-network chips ever made -- that Tesla chip is now in the list. That's quite an accomplishment, regardless of Tesla controversies. While the operationds per second is a bit different, the optimizations put into the Tesla chip is more for neural networks than for graphics processing, so that gives the Telsa chip an edge (for now).

Tesla's chip is literally a more powerful chip (in neural network performance) than in a ~$1000 GeForce RTX 2080 Ti for the latest and greatest advanced PC graphics for the best games/VR. And there's two of them. Even in the cheapest Tesla.

A pair of these chips, are on that specific circuit board that the "We cannot do it." phrase was applied to.

Sure, mass production is important (Fab throughput), but that massively powerful AI chip is a pretty advanced chip achievement from a horrendously giant fleet of engineers that Tesla started hiring a long ago.

The other vendors, will, possibly, eventually be forced to use garden-variety NVIDIA chips (or other AI vendor) to save costs -- but fortunately NVIDIA is now creating some custom AI chips for the supercomputing industry and other industries that also include self-driving AIs. I did mention a while back that I purchased some NVIDIA stock as an AI play, but I'm quite impressed at how Tesla has managed to usurp NVIDIA with their in-house chip -- an achievement. Not all automakers will do that though, and it's really expected that a massive scale-up of general purpose AI chips is going to continue to occur.

 

[crs1026]

Tesla's chip is literally a more powerful chip (in neural network performance) than in a ~$1000 GeForce RTX 2080 Ti for the latest and greatest advanced PC graphics for the best games/VR. And there's two of them. Even in the cheapest Tesla.

Maybe so, but electronics get cheaper pretty quickly. I bet that engineer meant to add the word "yet" to his declaration.

And, crankshafts and timing gears aren't exactly cheap, either. Stack those two chips up against the cost of a 4-cylinder turbo equipped prime mover and 6-level CVT, and that circuit board doesn't look so expensive.

- Paul

 

[rbt]

The other side of "context is important" is that this is a very advanced neural-GPU much more complex than an Intel and AMD processor.

Sure, similar to a half dozen others out there all being created in small quantities; it's not bleeding edge tech unique to Tesla, though it is unique in the auto space. More importantly, it's an efficiency thing and not strictly required for the calculations themselves; especially if you use a different sensor kit to take more direct measurements.

The Dell example still applies with regard to the issue other automakers have building out electronics kits.

Tesla not being able to do it at Toyota's scale still applies (yet, though they may only be 10 years away from that scale).

...advanced neural-GPU much more complex than an Intel and AMD processor.

Far less complex of a design you mean; the hardware circuitry is relatively simple as much of the complexity (most instructions, and even basic features like branch prediction) aren't added. Since only a small number instructions are required for the specific task (tons and tons of trivial algebra) it becomes far easier to scale (which they did). More components != more complexity, in fact it's typically the inverse (see RAM).

Tesla is doing a very complex task, but the hardware itself is relatively simple.

NOTE: Complexity for me is a function of time & skill required rather than the number of pieces. It's more complex to turn a single block of granite into the shape of a bucket than 1 billion pieces of sand.

 

[mdrejhon]

Fair counterpoints.

You don't need the complexity of 3D rendering for pure AI, and can design a chip that focuses on cherrypicking GPU's neural performance benefits, and optimizing it in a more simple way.

That said, while simpler than an NVIDIA GPU in terms of architecture -- it certainly isn't easy to engineer such a chip. The chip alone, is only some (hard-to-measure) percentage of the years-ahead advantage Tesla currently has -- their foresight to be essentially be practically half IT engineers has contributed to this current situation.

 

[MisterF]

GM set to go all in on electric vehicles with new tech

General Motors (GM) laid out its strategy for ratcheting up its electric vehicle (EV) plans with a modular approach that will include new vehicles, manufacturing its own batteries and increasing charging stations for drivers.

GM wants to stake its own claim in the still-nascent EV market by spending upwards of $20 billion (U.S.) by 2025. The money is spurred by renewed confidence that developing and making its own batteries would give GM a competitive edge in the auto industry.

While the new vehicles were interesting in their own right, the lynchpin of the presentation was the new ‘Ultium’ battery. Developed in tandem with LG Chem, who has supplied GM with lithium-ion batteries and other electronic components, its chemistry is made up of nickel, cobalt, manganese and aluminum.

 

[Towered]

Nobody anticipated the Covid-19 disruption of transport. How will this change things? How will public transit accommodate social distancing measures necessary until a vaccine is developed as the economy slowly re-opens? How many people will risk taking transit on packed vehicles and stations? Where there be a shift towards higher car use again (hopefully only temporarily)? That would not be sustainable.

 

[TRONto]

People are getting used to the existence of COVID. The virus has a mortality rate of about 0.37% and of those 2.5% have no co-morbidities and under 60. I see that people will get used to the existence of the virus and *slowly* transit usage will get back to normal.

 

[jamincan]

Fair counterpoints.

You don't need the complexity of 3D rendering for pure AI, and can design a chip that focuses on cherrypicking GPU's neural performance benefits, and optimizing it in a more simple way.

That said, while simpler than an NVIDIA GPU in terms of architecture -- it certainly isn't easy to engineer such a chip. The chip alone, is only some (hard-to-measure) percentage of the years-ahead advantage Tesla currently has -- their foresight to be essentially be practically half IT engineers has contributed to this current situation.

How does the chip compare to other existing neural-network-optimized chips that already exist on the market like Google's Edge TPU?

 

[Rainforest]

Nobody anticipated the Covid-19 disruption of transport. How will this change things? How will public transit accommodate social distancing measures necessary until a vaccine is developed as the economy slowly re-opens? How many people will risk taking transit on packed vehicles and stations? Where there be a shift towards higher car use again (hopefully only temporarily)? That would not be sustainable.

Short term, higher car usage as a % of all trips, but lower absolute number of trips by all modes including cars.

Afterwards, if the vaccine comes, then basically back to normal. A certain shift to online work / study / shopping should remain in place, but that won't be a dramatic shift.

If all vaccine creation efforts fail, then the reduction in trip counts will obviously last longer. But eventually this virus will become endemic, the % of resistant individuals will be higher, and among those who does get sick, the mortality rate will go down. Kind of another nasty flu, but not worse than that. Not sure how soon the natural transition can occur; would guess 10 to 20 years.

 

[urbanfuturist]

This is why we must halt investment in transit. Any fixed-route, rail-based transit we build now will be obsolete in a few short years. As self driving cars (ride-hailing and ride-sharing) make roads more efficient and cheaper to use, transit will be without purpose, and all that money will have been wasted.

 

[Rainforest]

This is why we must halt investment in transit. Any fixed-route, rail-based transit we build now will be obsolete in a few short years. As self driving cars (ride-hailing and ride-sharing) make roads more efficient and cheaper to use, transit will be without purpose, and all that money will have been wasted.

That's unlikely. Cars still take too much road space, even if they are 100% self-driving and 100% ecological / using renewable energy.