Tesla, TSLA & the Investment World: the Perpetual Investors' Roundtable

[Thumper]

It is not clear that FCA will be alive to pay for the pool fees for more than a year or maybe two. I hope Tesla does not rely on the $.

 

[EVNow]

I believe it's hypergeometric

I doubt it. Hypergeometric has finite population.

 

[Big Earl]

I haven’t followed Semi specs so much, but I seem to remember reading that 4 AWD-type motors can power it? If that’s true, I find that phenomenal.

Yes, the Semi is reported to be using four Model 3 motors - one for each rear wheel. This should easily provide 1,000+ hp of forward propulsion and ~333 hp of regenerative braking. These figures trump anything that's available from diesel right now, with the most powerful (non-concept) rigs being in the 700 hp range.

This arrangement provides for computer-controlled all-wheel-drive with torque vectoring, which will prove to be a serious asset in bad weather and should nearly eliminate jackknifing.

 

[EVNow]

Insurance should not change much for those driving a ‘horse’ car. But should be cheaper for those with FSD.

As ‘horses’ become scarce, the chance of being run into by an uninsured ‘horse’ falls. The horse driver also enjoys lower accident rate as other cars on the road are increasingly likely to have FSD.

Not if horses keep hitting FSD cars who can prove that horses are at fault every time.

 

[Ratanpara]

Nvidia's claimed TOPS are not realistic for large NN processing at small batch sizes, they are peak theoretical rates using larger batch sizes and likely smaller NN's (perhaps the current NoA NN wouldn't be "too big" but the FSD NN would be). Especially so for the GPU TOPs. Even the "Tensor Cores" are likely not as purpose optimized as Tesla's NN architecture, but at least it's closer to an apples-to-apples comparison. The GPU TOPs from the dGPU and SoC GPU are very apples-to-oranges. All you have to do is compare the claimed performance of the AP2.5 HW versus Tesla's measured performance as reported during autonomy day to see a large discrepancy for GPU compute

An arbitrary benchmark may easily get NVidia's claimed performance but no real application will. Tesla's architecture with unique purpose built and tightly coupled hardware and software should obtain very close to theoretical performance, NVidia's chips will be nowhere near in real applications, which makes the real world performance per Watt even worse. .

Apple processor in ipod and iphone historically always needed less memory compare to samsung for same performance, so custom hardware and software(Tesla) always need less resources then general purspose GPU.

 

[oneday]

Yes, the Semi is reported to be using four Model 3 motors - one for each rear wheel. This should easily provide 1,000+ hp of forward propulsion and ~333 hp of regenerative braking. These figures trump anything that's available from diesel right now, with the most powerful (non-concept) rigs being in the 700 hp range.

This arrangement provides for computer-controlled all-wheel-drive with torque vectoring, which will prove to be a serious asset in bad weather and should nearly eliminate jackknifing.

A 600hp Diesel engine can provide 600hp in compression exhaust braking (jake brake). Based on these numbers it would appear diesel’s have the advantage in this category.

The major difference, engine exhaust braking decreases your fuel efficiency. Regenerative braking delivers electrons into your battery.

 

[Doggydogworld]

Nvidia's claimed TOPS are not realistic for large NN processing at small batch sizes, they are peak theoretical rates using larger batch sizes

Can you explain batch size as it relates to these chips running CNNs?

 

[oneday]

Can you explain batch size as it relates to these chips running CNNs?

I don’t think Tesla uses CNN’s
Edit: I stand corrected. I tried finding my source, but couldn’t. Now I know why.

 

[Alphacrux]

Tesla could go even higher than our $4,000 price target, says fund manager Cathie Wood
Tesla could go even higher than our $4,000 price target, says fund manager Cathie Wood

"Tesla’s plan to raise $2.7 billion worth of capital didn’t phase Wood either, she said Friday in a phone call with CNBC."

“Our bear case has it going to $700 and our bull case is $4,000, but now we think that’s too low,” she said, explaining that Ark’s five-year time horizons for each case already assumed capital raises would occur."

 

[Doggydogworld]

I don’t think Tesla uses CNN’s

98% of their neural network operations are convolutions...

 

[anthonyj]

Tesla could go even higher than our $4,000 price target, says fund manager Cathie Wood
Tesla could go even higher than our $4,000 price target, says fund manager Cathie Wood

"Tesla’s plan to raise $2.7 billion worth of capital didn’t phase Wood either, she said Friday in a phone call with CNBC."

“Our bear case has it going to $700 and our bull case is $4,000, but now we think that’s too low,” she said, explaining that Ark’s five-year time horizons for each case already assumed capital raises would occur."

 

[Menifeer]

The major difference, engine exhaust braking decreases your fuel efficiency. ...

How so? Fuel delivery is halted when the Jake Brake is in operation.

 

[oneday]

How so? Fuel delivery is halted when the Jake Brake is in operation.

I didn’t really word it right. If there were no speed limits both an electric truck and a diesel truck could just coast thereby being equal. Any braking reduces kinetic energy. At least ~50% is given back to the battery in regenerative braking.

 

[HG Wells]

How so? Fuel delivery is halted when the Jake Brake is in operation.

Jake Brake in many areas is not allowed due to noise.

 

[RobStark]

Otto-Pilot!!
View attachment 404155

That is not Otto-Pilot.

THIS is Otto-Pilot.

 

[AZRI11]

So, to anyone whose friends are thinking about buying an ICE car: encourage them to buy from the following brands:

• Abarth
• Alfa Romeo
• Chrysler
• Dodge
• Fiat
• Fiat Professional
• Jeep
• Lancia
• Maserati
• Ram Trucks

They'll be contributing via paying the "Tesla Tax"

I won't curse at the Alfa Romeo commercials on CNBC anymore

 

[ZsoZso]

Except the example I cited: on ice the stopping distance is +60% with ABS, and the best braking strategy is to slip intentionally. Which a professional driver won't be able to perform due to always-on ABS.

So yes, ABS is an example of safety technology adoption that is obviously less safe in at least one important driving scenario.

OK, I was wrong about that case. Still, the end result is the same: if your professional driver kills a pedestrian due to too long breaking distance on ice with ABS, the court / jury will never blame that on the technology, the driver still will be found at fault for driving too fast in the first place. On the other hand, if an FSD car kills a pedestrian, that will be very much blamed on the technology with serious wide ranging consequences.

 

[Maitri982]

I listened the autonomy event again and got a few nuggets i missed the first time:

• the FSD teslas will likely have a "phone home" capability if something goes wrong
• there are numerous neural nets being used
• BIG item for me at least...the neural net at this time is only doing object recognition...nothing else...not path planning for example.
• they expect more and more functions to be taken over by the NN in the future from the heuristic code unless heuristics are simple to handle and more computationally efficient than the NN would be
• tesla wants to run the business to cash flow neutral until FSD is available because they want to build as many autonomous units as they can
• the cars drive train etc is built for 1 million miles now...batteries for ~300K miles. Plan to improve battery to 1 million next year by getting to 4,000 cycle support, which they say they already have in powerwalls. But they will optimize on 50kw batteries and thats where they want to push people instead of to 75/90/100 or more. larger batteries means less autonomous units on the street and the goal (given FSD presumption) is to have as many of these cars as possible on the road as that is best for environment and for $$$(profit) and for optimizing manufacturing throughput.
• they believe over time most FSD functions will be done in NN...but as it is now...just object detection
• early FSD cars will still have the passenger sit in the drivers seat and they can take over if needed. i mean for the uber like service this will still be true.
• Elon said (weird to me) that after next winter the path planning of the car will be amazing...not sure whats magical about that time frame
• they are confident that their current lather, rinse, repeat process using the data engine to learn NN weights, fed from annotated data, then finding more edge cases, request more examples and then annotate them and feed back to data engine is the right way to get to FSD.
• they are confident in handling snow as good as humans
• tesla ride service will be much cheaper than uber as their costs per mile are much lower
• 100,000 tesla directed lane changes per day with zero accidents caused thus far
• they do not use the new board/chips for training, but for running the already created NN only. they may change this in the future i gathered....maybe its part of dojo?
• they have testbed to run real world scenarios against the software and see what it does compared to what the actual human did.
• someone asked if they thought about using the new board/chip while it sits idle in the garage as some type of supercomputer service. Elon clearly had not thought about this but it got his attention. Bitcoin mining anyone?

thats about it...hope you find it useful.

 

[SPadival]

None of this is true. absolutely none of it.

Hmm.. very short rebuttal without any substance. Maybe your excuse is that you are writing this in your car while driving?

 

[JRP3]

Yes, the Semi is reported to be using four Model 3 motors - one for each rear wheel. This should easily provide 1,000+ hp of forward propulsion and ~333 hp of regenerative braking. These figures trump anything that's available from diesel right now, with the most powerful (non-concept) rigs being in the 700 hp range.

A 600hp Diesel engine can provide 600hp in compression exhaust braking (jake brake). Based on these numbers it would appear diesel’s have the advantage in this category.

The electric motors should be capable of regenerating as much power as they put out, the limit is the pack acceptance rate. With the assumed huge pack in the semi regen should be more than a Diesel motor.