Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

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

This site may earn commission on affiliate links.
Wall Street Journal website today. How things have changed!
tslabmw.png
 
I disagree here.

What you are arguing is Tesla needs to split (largely the same) development into two branches, one for HW2.5 and one for HW3. This is neither a standard practice nor aligns well with their economic interests.

Splitting the codebase in two is not exactly what it is. The neural network is going to have two or three versions. But the rest of it is not being split.

But yes. This exactly what they need to do.
 
WSJ’s Tim Higgins with an awful “good” story about the share price rise (paywall).

He sprinkles in article SEC cases, challenges in Model 3 production & that the $420 price was set “to amuse his girlfriend because the number is part of marijuana culture. Mr. Musk further cemented his reputation in drug culture in September of last year when he appeared during a live-video interview puffing a marijuana blunt.”

Tesla Shares Race Past $420 Buyout Figure

Uh yeah that is one wall I'm not gonna pay
 
I’m thinking we aren’t in a squeeze unless borrow fees start to increase; or is liquidity just too high for that to be an indicator? Otherwise a very sharp price increase would be the other telltale sign I suppose?
No, generally the borrow fees go down during the short squeeze. Shorts need to cover, which means they need to buy, which increases liquidity and makes it easier (for total loonies) to borrow shares.
 
I disagree here.

What you are arguing is Tesla needs to split (largely the same) development into two branches, one for HW2.5 and one for HW3. This is neither a standard practice nor aligns well with their economic interests.

You cannot claim that they will freeze HW2.x in place and don't do any more development there because there are thousands of people that paid for EAP which is a huge subset of "FSD" that has not been finished yet.

As such they are doing the most sensible thing - which is develop a single codebase that would work on both. They then went the extra mile to disable some functionality on hw2.5 in the UI while fully retaining it in the underlying system.


I suggested no such thing and it's just your imagination.

They are compiling same/very similar similar NNs for HW2 and for HW3 (which makes sense in light of the above about conserving development effort), but that does not mean hw3 upgrade is pointless (I've no idea how important that is, for all I know they might run the NNs on hw3 at higher fps).
My problems is they seem to be deliberately turning off certain visualizations for no good reason. Even if it's less perfect on hw2.5, what's the problem with letting people see it? Seems to be a wasted PR opportunity to demonstrate "see how much better hw3 performs!" (since right now there's no difference I can feel between the two in any way when in operation)

There is probably a "good reason" even though we don't know what it is...

My guess would be they can't cram too much into a HW2 car without compromising performance somewhere. It is important that HW2 continues to the main job of driving the car as well as possible..

So it would be interesting to look at the latest release, and see if this is just disabled in the UI, that makes no sense to me...

Regardless, Tesla know what they are doing, and they always do things for a reason.
 
  • Informative
Reactions: Artful Dodger
I disagree here.

What you are arguing is Tesla needs to split (largely the same) development into two branches, one for HW2.5 and one for HW3. This is neither a standard practice nor aligns well with their economic interests.

You cannot claim that they will freeze HW2.x in place and don't do any more development there because there are thousands of people that paid for EAP which is a huge subset of "FSD" that has not been finished yet.

As such they are doing the most sensible thing - which is develop a single codebase that would work on both. They then went the extra mile to disable some functionality on hw2.5 in the UI while fully retaining it in the underlying system.

Here you are jumping to conclusions again:

"there are thousands of people that paid for EAP which is a huge subset of "FSD" that has not been finished yet."​

Much of EAP is completed already:

Screen-Shot-2018-08-31-at-11.59.25-AM.jpg

If you read through that list:
  • match speed to traffic conditions: done
  • keep within a lane: done
  • automatically change lanes and transition from one freeway to another: done
  • be summoned to and from your garage: done
  • self-park when near a parking spot: "Smart Parking" feature promised but not yet done.
I.e. most of those promises, with the exception of "Smart Parking", is present on HW2 cars.

It is pretty clear that Tesla is forking HW3 behavior from HW2 right now - with the 2019.40.50 release. They might share the NoA and Smart Summon networks for a time still, but the City Driving features require a level of object recognition that appears to be beyond HW2's capabilities.

Here's a tweet from Elon that states that the compute load on HW2 is around 80% for "Navigate on Autopilot on Highways" workloads already:

upload_2019-12-23_22-46-7.png

Here's a tweet from Elon that states that the compute load on HW3 is only 5% with the same workload:

upload_2019-12-23_22-48-34.png

The fork of HW3 from HW2 is a stated goal of Tesla. It's not my assumption, it has been explained by Andrej Karpathy and Elon Musk as well - both have said it numerous times that HW3 is vastly more capable and that there's going to be an inevitable moment when HW3 functionality will surpass HW2 functionality.

I suggested no such thing and it's just your imagination.

They are compiling same/very similar similar NNs for HW2 and for HW3 (which makes sense in light of the above about conserving development effort), but that does not mean hw3 upgrade is pointless (I've no idea how important that is, for all I know they might run the NNs on hw3 at higher fps).
My problems is they seem to be deliberately turning off certain visualizations for no good reason. Even if it's less perfect on hw2.5, what's the problem with letting people see it? Seems to be a wasted PR opportunity to demonstrate "see how much better hw3 performs!" (since right now there's no difference I can feel between the two in any way when in operation)

I highlighted the explanation to your question: if the results on HW2 are "less perfect" than on HW3, then that alone is a good reason to not show those less perfect visualizations on HW2 systems.

Since these "visualizations" form the basis for FSD driving decisions, it makes no sense to show them on HW2 using imperfect networks that they'll never use to actually drive the car using that particular hardware.
 
Last edited:
Firstly, @verygreen - you should know that @Fact Checking is rarely wrong.

I don't think my track record (which is sadly not spotless at all) is a good basis to end an argument, especially in an area where @verygreen has a good track record and direct experience as well. I always enjoy the videos and explanations of @verygreen, and whether I'm right in this particular argument is yet to be seen.