Also have to consider that once your EV burns, you'll be accessible to bears, wolves, and mountain lions.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable
- Thread starter AudubonB
- Start date
First, the design power goal for the entire FSD computer (not just NN chips, but the entire board power including powering the sensors through the board with redundant power supplies) was around or less than 100W because they needed to be able to retrofit it into existing cars (while Model 3 could have easily handled more thanks to it's connection to the liquid cooling loop, S and X are air cooled with fans and heatsinks inside the dashboard, which is very suboptimal for cooling), plus power draw impacts range. They specifically said that of the 72W observed in real life operation, 15W is being used by the NNs, the rest of that is *everything else* (power supplies, RAM, ARM cores, GPU cores, etc etc). Perhaps the NN will use more when NN complexity grows, or perhaps they always fill the NN units with fake work to keep them operating at 100% (they do say it continuously loops running code) and that's the peak power it will ever hit - we can't know unless Tesla reveals more details. Tesla specifically said it was the full AP stack so that leaves room for more complexity to use more power when running FSD (since it seems likely FSD is not the same as "the full AP stack"), but worst case must remain around 100W total, so at most ~28W more power for NN processing (this does not mean simply that they're already using 1/3 of available TOPS as static power consumption of the NN units even if not doing anything, assuming they can don't keep them stuffed with work and instead idle them when doing nothing useful, will likely be part of that 15W). In any case, irrelevant considering the following points.
Second, TOPS/W efficiency won't matter if the NN doesn't fit on the chip/processing unit, or has too many stalls due to cache misses due to lack of local SRAM (I'm not sure if EyeQ5 has comparable local SRAM or just L2 cache due to lack of detail I could find, but it's clearly a win over NVidia's designs). Tesla determined they needed a minimum amount of NN compute ability and memory, and designed their NN to meet or exceed this, while still being more efficient than NVidia's designs, which were the only available designs on the market that they could do as they wish with (see below point). EyeQ5 may be more efficient but the claimed TOPS per chip may be insufficient for Tesla's needs and some things can't be simply split up among different chips well, so just adding more EyeQ5's to meet the TOPS requirement isn't an option necessarily. With only 24 TOPS (EyeQ5), versus Tesla NN unit at 36 TOPS (2 per chip for 72 TOPS), that is a difference of 50% more performance per NN unit in favor of Tesla's NN chip. Add to that the fact that using a second NN unit requires an entire second chip for MobilEye's solution (vs having a second unit of another 36 TOPS on chip) may make all the difference in terms of software complexity. Also, I question MobilEye's EyeQ5 design effectiveness for large NN's since one of the charts MobilEye has posted looks like that's across 18 units (1.33.. TOPS per unit), which may make the software complexity even greater (verus one NN processor that operates on a single large 96x96 data set per cycle for 36.8 TOPS per unit), with corresponding hits to realistic TOPS efficiency.
Third, if Tesla can't write and deploy their own code when they want and how they want, then there's no point to having MobilEye hardware, and they must go their own way as they have. Part of the reason for their split was that MobilEye was too timid to let Tesla continue pushing the envelope.
What frustrates me is that Waymo is just robotaxi company that will either sell FSD to someone else, or have to pay money for a fleet and yet... I wish $TSLA were valued like a robotaxi company
$TSLA market cap ~$44B
Waymo worth $175B
Waymo is worth $100 billion more than previous estimates, Morgan Stanley says (GOOGL) | Markets Insider
Tesla need to show some real time testing videos of FSD on city streets, then some people will start to believe, until then it is just a dream.
Semi-OT
Boy, the Trumpsters are idiots. I knew that in China, 99% of the enforced legal stuff was actually regulations. The law was actually *less* enforced than the regulations, historically (consider China's labor and environmental laws, which look great on paper). Apparently the crack idiot squad at Trump Central didn't have anyone who knew a damn thing about the Chinese government at all. Wow.
Boy, the Trumpsters are idiots. I knew that in China, 99% of the enforced legal stuff was actually regulations. The law was actually *less* enforced than the regulations, historically (consider China's labor and environmental laws, which look great on paper). Apparently the crack idiot squad at Trump Central didn't have anyone who knew a damn thing about the Chinese government at all. Wow.
The Federal Reserve is preparing the American financial system to withstand climate change, Chair Jerome Powell tells Democratic Sen. Brian Schatz
From the article:
The question that comes to mind for me is, just how far behind is the US in preparation compared to the rest of the countries in the world?
From the article:
- The Federal Reserve is taking steps to prepare the American financial system for the effects of climate change, according to Fed Chairman Jerome Powell.
- But Powell notes that climate-related risks do not fit squarely within the central bank’s existing framework for assessing financial stability.
- “As I have commented previously, although addressing climate change is a responsibility that Congress has entrusted to other agencies, the Federal Reserve does use its authorities and tools to prepare financial institutions for severe weather events,” Powell writes to Sen. Brian Schatz, D-Hawaii.
The question that comes to mind for me is, just how far behind is the US in preparation compared to the rest of the countries in the world?
That was Azaz's entire point.
Mark my words, when Musk starts making cautious, quiet statements about robotaxis and starts talking about how "human drivers are underrated", it'll mean they've made enormous progress on robotaxis and have a solid plan for getting there one project at a time.
So, better air conditioning in vaults?
Driving data collection and analysis of said data. Mobileye has 0, nada, none. Some of their partners may have some, but so far, it seems unlikely that any of them have much.
The data collection is not only necessary to train the NNs, it's necessary to figure out what the driving problems actually are and what the NNs need to do. I fully expect Tesla to have to rearchitect their system at least once more; at that point Mobileye will probably still not have noticed the problem which will cause Tesla to rearchitect their system, and will hit it 2-3 years later.
- I'd say better US financial planning preparation for disasters like Puerto Rico from Hurricane Maria that could save lots of lives going forward.
humbaba
sleeping until $7000
I hear you and I get it, but AP is real and widely deployed its the perception* that is held back. With the no-confirmation update, Tesla now has the equivalent of Waymo (safety driver bound) except that instead of being limited to one location on fixed routes it is limited to the interstate network.
Sure, it isn't FSD. But then again, neither is Waymo's (at least, not by any useful measure of FSD)
* which I think was the entire point of autonomy day (trying to alter the perception)
But meanwhile, Musk is talking about colonizing Mars and hasn't even started working on the extremely difficult biology problems. (The physics problems are, frankly, solved. It's the biology problems which make Mars colonization implausible -- I can guarantee that the first colony will either abandon Mars and come home, or die. The biology lessons they learn from their failure will inform a second colony, if there is one...)
The credibility problem is absolutely the overpromising. Musk is totally honest, but his optimism veers into the delusional routinely when it comes to timelines, and occasionally when it comes to other things (they simply aren't getting 10K Model 3/year out of Fremont -- at all).
Artful Dodger
"Neko no me"
Link, or you're just making it up (or I'm just too lazy to go find the Ferengi rules of acquisition for myself
OT
Ferengi Rules of Acquisition
All software patents are illegal and have been for a very long time; software is pure mathematics and it's established that mathematics is unpatentable. Sadly a lot of judges don't understand this. The USSC *has* understood this and has intermittently, but repeatedly, smacked down the corrupt Federal Circuit, which is a fake court owned by patent lawyers. It's expensive to take your case to the Supreme Court but Tesla could afford it.
Unfortunately, the automatic driving patents are for an *application* of math to the real world (unlike software patents which are just about internal computer workings, which are totally unpatentable -- it's gross that any judges have ever allowed them, but it shows that many judges are ignorant fools) so they may survive the end of software patents qua software patents. However, in many cases the application is essentially obvious, so most of MobileEye's patents will probably be thrown out as "never should have been granted, too obvious".
In practice, everyone in the software industry knows this, so with the exception of occasional lunacy like the Apple/Samsung case, they all sign "mutual disarmament" agreements agreeing to not sue each other over any purported software patents. (Hardware patents are another matter.)
Doggydogworld
Active Member
It's not all extra. FCA paid Tesla 300m+ in 2018.
I'd still like to understand the 200m of non-ZEV in Q1.
Fact Checking
Well-Known Member
Also note that the 2020 income might be significantly larger than the $167m/q which is $2b divided by 12 quarters: I believe both @generalenthu, @Doggydogworld and @ReflexFunds agreed that FCA's 2021 assumptions seem very rosy: where are they going to get the EVs to sell, if even top German luxury OEMs have been unable to offer competitive EVs, after half a decade of effort to bootstrap their EV programs.
Also, if @generalenthu's estimate of around $25k/vehicle credits for the first couple of ten thousand units sold in Europe are true, then Tesla might slow-walk 2019 European demand and sell those units in 2020 instead, to maximize the FCA income.
Every 10k units sold in Europe in 2020 would generate above $100m in regulatory credits, if we assume a 50%/50% split in the benefits between FCA and Tesla.
So it's easy to see $300m/quarter and higher levels of FCA income in Europe alone, in 2020 - i.e. around $1.5b/year. To be repeated in 2021 and 2022, unless FCA manages to get their CO₂ - which they frankly cannot with their current product palette, and any future products might not sell or might not sell profitably enough, i.e. it might still be more profitable in 2021 and 2022 to just pay Tesla.
Oh come on, already covered
www.netflix.com/title/80144355
These guys had it harder than the first Martian
The Martian (2015) - IMDb
The communications issues really ARE of existential importance. While they may not be as urgent as other existential issues, they are also a corporate culture matter, which is notoriously hard to fix once it's gone bad -- by not dealing with it up front they have created a problem which will at the least take a lot longer and cost a lot more to fix, and they might be unable to fix at all, and it's the problem which might kill them. I think Musk as an engineer underestimated the importance of human factors.
The 10K/week was existential as well; they managed a quick pivot to an alternative 10k/week strategy by getting the Chinese factory deal super-accelerated, which means they're OK. (And during the failing Model 3 ramp-up, they also pulled every emergency trick they had to get to 5000/week while slashing very important parts of their budget to save cash.) Had Shanghai not been so hell-bent on getting Tesla, or had the Chinese central government been a lot slower to approve the wholly owned factory, the company might be in a very bad position right now.
I hope so. They do seem to be shifting to trying to fix communications, which they absolutely need to do in order for the company to have a long-term future rather than just a 5-10 year future. I continue to watch this, but ridiculous and absurd horror stories, *from trustworthy owners*, are still showing up over in the Model 3/S/X forums -- though less frequently, which is good.
I do not trust that Tesla will take proper care as they have a record of not always doing so. You trust that they will. I certainly hope you're right.
However, as a prudent investor, I will be *watching their insurance rates* and seeing if they make sense to me -- if they're underpricing severely, then I know they're not taking proper care.
I am never going to pay the WSJ, who are criminals. Any new information here? (Specifically, who's on the hook for the underwriting gains/losses? Is Tesla simply branding, or is Tesla taking the insurance risk?)
Similar threads
- Locked
- Locked
- Locked
