Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Musk Can’t Dodge Tesla Cash Woes Any Longer

bloomberg.com

Post reply on HN
Print view View original

71–80 of 152 posts

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#71

Post by Robotbeat »

dragontamer wrote: post #39
DennisP wrote: post #34

Earlier quoted context omitted.

There's more to it than feature size. Other companies have more general-purpose chips. Tesla basically built an ASIC for self-driving.

NVidia is the only chip that seems a bit heavy on GPU features. But the Volta architecture has tensor processors with full FP16 and INT8 support, so it seems useful for self-driving as far as I can tell. MobilEye EyeQ4 and EyeQ5 has no other purpose than self-driving. EyeQ5 is going to be 7nm and is launching 2020. A full ASIC for self-driving as well. And no one knows what the heck Google / Waymo is doing. All we kn…

Launching in 2020 means two years (starting in 2018) of using the more expensive and slower Nvidia chip. By using the 14nm process now, Tesla paid far less on upfront dev costs (we're talking tens of millions versus billions) and saved on unit costs right out the gate, putting them in better position for smaller process sizes in the future.

Volta architecture is much more general purpose than Tesla's.

If you're doing a custom ASIC hyper-focused on your application and you're not selling hundreds of millions of them, it doesn't make sense to lock yourself into paying of dev costs for cutting edge fabrication techniques with risk of yield problems and questionable per-unit economics.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#72

Post by MaupitiBlue »

Anyone with a scintilla of intuition about human behavior would have recognized at least a year ago that Musk was begging for a way out of Tesla. The talks he gave about how overworked he was, the tweets, and the violation of the judge’s orders screamed out that he had given up.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#73

Post by andbberger »

CoolGuySteve wrote: post #20
brianpgordon wrote: post #2

> a fleet of self-driving vehicles that will make their way into the robotaxi service as soon as next year. So... are we living on the same planet? Does Musk seriously think that Tesla will have solved full autonomous driving, with no LIDAR, and have commercialized a functioning robotaxi service, by next year ? This is so far outside of the realm of possibility that it's like saying he's going to Pluto next week.

A robotaxi isn't too far off if you limit where it can operate. It's not infeasible to train an automated driving system specifically suited to the streets of LA or wherever. Some place where it doesn't snow, doesn't rain often, and drivers/cyclists/pedestrians are fairly predictable (at least compared to someplace like Asia). Alternatively, a robobus, that is to say an automated system that drives from one place to…

> It's not infeasible to train an automated driving system specifically suited to the streets of LA or wherever

You don't know that.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#74

Post by gamblor956 »

WhatIsDukkha wrote: post #55
gamblor956 wrote: post #47

Earlier quoted context omitted.

The researchers found that analyzing the captured images from a bird’s-eye view, rather than the more traditional frontal view, more than tripled their accuracy, making stereo camera a viable and low-cost alternative to LiDAR. I look forward to the day when cars have giraffe-necks protruding from their roofs, mounted with stereo cameras... But until then, the researchers actually just confirmed that front-mount camer…

I'm REALLY unclear how you decided to post this giraffe comment... The picture and the words in the article show cameras next to the rear view mirror. """ Using two inexpensive cameras on either side of a vehicle’s windshield, Cornell researchers have discovered they can detect objects with nearly LiDAR’s accuracy and at a fraction of the cost."""

That's not what the article originally said, so it appears that it's been updated since then. Good think I quoted the relevant section of the article in my comment...

That being said, existing self-driving cars already have some cameras at that level without displaying the sort of accuracy the researchers found, which relates to the methodology issues other commenters have noted.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#75

Post by Robotbeat »

mplewis wrote: post #66
agumonkey wrote: post #56

Earlier quoted context omitted.

You don't buy his claims that lidar is non necessary and even inefficient at sdv ?

Elon/Tesla have a vested interest in making sure LIDAR isn't required by law for self-driving vehicles. In addition, his NN-camera-powered-cars are still trying to drive themselves into highway barriers.

Requiring a specific technical solution with problems of its own by law is a wet dream for whoever holds the IP to that technical solution. Now accusing others of having a vested interest in MAYBE opposing this hypothetical law that no one has actually proposed is just gas-lighting and possibly projection.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#76

Post by jmpman »

toomuchtodo wrote: post #53
Judgmentality wrote: post #46

Earlier quoted context omitted.

That study has been highly disputed by people in the field; the methodology is questionable at best for the claims they're making.

Anthony Levandowski of Waymo LIDAR fame has said he now agrees with Elon Musk ("a wiser person than me") that LIDAR is a crutch. https://youtu.be/fNgEG5rCav4?t=105

DARPA should have continued their self driving car competitions after their Urban challenge, but limiting the input to just commodity cameras. The field would be much further along.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#77

Post by dragontamer »

Robotbeat wrote: post #64
keithwarren wrote: post #38

Earlier quoted context omitted.

Asserting that Tesla is behind because they are fab at 14 instead of something smaller is so laughable you cannot be serious. Thermal profile and power, instruction set and a dozen other things are more important. Watch the autonomy day video on how the chip was built and why, will help a lot

Frankly, it'd be insane to invest in 10 or 7 or 5nm at this point for a custom design of maybe a few million units. The investment required scales insanely with reduction in feature size (Billions of dollars, potentially), and you'll never get your investment back. That effectively locks you in to an old design for several years. And the difference in unit costs or efficiency between 14 and 10nm is not big enough to…

Intel / MobilEye is making a 7nm chip for all of its customers. And part of that is because MobilEye has Ford, GM, BMW, Audi...

MobilEye is deployed in 313 car models across 27 manufacturers. They're going to have a 7nm chip next year, and are easily going to leapfrog Tesla's 14nm capabilities.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#78

Post by dragontamer »

Robotbeat wrote: post #71
dragontamer wrote: post #39

Earlier quoted context omitted.

NVidia is the only chip that seems a bit heavy on GPU features. But the Volta architecture has tensor processors with full FP16 and INT8 support, so it seems useful for self-driving as far as I can tell. MobilEye EyeQ4 and EyeQ5 has no other purpose than self-driving. EyeQ5 is going to be 7nm and is launching 2020. A full ASIC for self-driving as well. And no one knows what the heck Google / Waymo is doing. All we kn…

Launching in 2020 means two years (starting in 2018) of using the more expensive and slower Nvidia chip. By using the 14nm process now, Tesla paid far less on upfront dev costs (we're talking tens of millions versus billions) and saved on unit costs right out the gate, putting them in better position for smaller process sizes in the future. Volta architecture is much more general purpose than Tesla's. If you're doing…

> Launching in 2020 means two years (starting in 2018) of using the more expensive and slower Nvidia chip.

You mean a cheaper and more industry established MobilEye EyeQ5 chip, with LiDAR capabilities, Tensor Units and all that good stuff.

------------

Let me ask you this: how many vehicles do you think Tesla will make in 2018 and 2019? A million? Two million?

Do remember, 2018 Tesla made only 244,920 cars. 2019 Q1 only had 70,000 cars, so Tesla is currently producing cars at a rate under 300,000 cars/year (maybe growing to 350,000 by the end of the year). Maybe Gigafactory 3 comes up in early 2020, and Tesla then makes 500k cars for 2020.

So... just about a million cars between 2018 to 2020, and I'm being quite generous on Tesla's growth potential here. Remember: not all of those cars having FSD or autopilot installed.

I'm not seeing how this 14nm chip is going to be a financially sound move for Tesla. They just don't make enough cars and they weren't watching out on industry standard chip upgrades.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#79

Post by Robotbeat »

ergothus wrote: post #32
mpweiher wrote: post #18

Earlier quoted context omitted.

A true engineer! :)

I realize you're joking, but this is actually kind of bothersome to me. Most of us give overly optimistic delivery dates without intending to. Over time we learn that we have this habit, so we pad in some time we think is unnecessary...and still give overly optimistic delivery dates without intending to. If you add someone INTENDING to give such poor estimates, the problem moves from the "expected and occasionally pr…

He's not intending to be over-optimistic; his goals are possible, he sees they are possible and says so, even if they're unlikely. And he repeatedly states his tendency to be over-optimistic, which RARELY is acknowledged by pretty much anyone else promoting a new advancement.

"I think we can do this by XX. Granted, I've been over-optimistic in the past"... is refreshing honesty, not "lies."

It's amazing to me that so many people interpret acknowledgement of one's own fallibility as lying.

Re: Musk Can’t Dodge Tesla Cash Woes Any Longer

#80

Post by DennisP »

gamblor956 wrote: post #74
WhatIsDukkha wrote: post #55

Earlier quoted context omitted.

I'm REALLY unclear how you decided to post this giraffe comment... The picture and the words in the article show cameras next to the rear view mirror. """ Using two inexpensive cameras on either side of a vehicle’s windshield, Cornell researchers have discovered they can detect objects with nearly LiDAR’s accuracy and at a fraction of the cost."""

That's not what the article originally said, so it appears that it's been updated since then. Good think I quoted the relevant section of the article in my comment... That being said, existing self-driving cars already have some cameras at that level without displaying the sort of accuracy the researchers found, which relates to the methodology issues other commenters have noted.

The researchers described a new way of processing the data, claiming a dramatic improvement over previous methods.
Post reply on HN