TMC is an independent, primarily volunteer organization that relies on ad revenue to cover its operating costs. Please consider whitelisting TMC on your ad blocker and becoming a Supporting Member. For more info: Support TMC

FSD may require a hardware upgrade...

Discussion in 'Model S' started by boonedocks, Feb 13, 2017.

  1. MP3Mike

    MP3Mike Active Member

    Joined:
    Feb 1, 2016
    Messages:
    2,774
    Location:
    Oregon
    Because the people that paid for EAP didn't pay for the FSD features, so they won't get them. Why would Tesla give those features away for free?
     
    • Like x 5
  2. Max*

    Max* Autopilot != Autonomous

    Joined:
    Apr 8, 2015
    Messages:
    5,773
    Location:
    NoVa
    All cars have had AP1 hardware for a while, and yet, only the AP1 enabled cars were able to use the AP1 features. I'm not sure what you're getting at though?
     
    • Like x 2
  3. lunitiks

    lunitiks Member

    Joined:
    Nov 19, 2016
    Messages:
    659
    Location:
    Norway
    I'm getting at the fact that it seems unnecessary!

    It also seems unnatural to talk about features that EAP-cars get, that FSD-cars dont. FSD-cars should get ALL the things. Then, EAP-cars should stop getting updates
     
  4. Max*

    Max* Autopilot != Autonomous

    Joined:
    Apr 8, 2015
    Messages:
    5,773
    Location:
    NoVa
    I agree with the 2nd sentence, still not sure I follow the first.

    You're assuming everything will happen in a linear progression. It won't.

    Some features get done faster than others. Tesla may decide to bundle those features into EAP or FSD, and send them out accordingly.

    But you're right, one day, when EAP is complete, then EAP will stop getting updates. Until that day, once FSD starts to gradually roll out, you'll likely get mixed updates. Some features for EAP cars, some features for EAP+FSD cars.
     
    • Like x 2
  5. MorrisonHiker

    MorrisonHiker S 90D 17.17.4

    Joined:
    Mar 8, 2015
    Messages:
    2,703
    Location:
    Colorado
    I think everyone agrees that FSD cars will get EAP updates as well. We're just saying that in 3-6 months, FSD cars will start receiving features that EAP cars will not get. EAP and FSD cars will continue to receive updates after those FSD-only features are released.
     
    • Like x 2
  6. lunitiks

    lunitiks Member

    Joined:
    Nov 19, 2016
    Messages:
    659
    Location:
    Norway
    Ok. Phantom disagreement, then.

    Must be because of my pointing out that FSD sounds like the whole shabang, and that talking about "FSD Features" (being gradually implemented) is weird, and conflicts with the word FSD. Guess we've settled it.

    Sorry
     
    • Like x 2
  7. dknisely

    dknisely Member

    Joined:
    Dec 4, 2016
    Messages:
    86
    Location:
    Seattle, WA
    Tesla to make its own custom SoC (System on Chip) for self-driving cars built by Samsung, report says

    Caveat that this is my opinion and reflects a lot of gross generalizations:

    The [VERY POWERFUL!] NVIDIA platform is based on GPUs, which are historically designed to render graphics for applications like gaming. While they offer incredible potential for billions of math operations per second, this is accomplished by massively duplicating rather inflexible processing elements that do math operations in parallel without a lot of additional "logic" processing. Recently, GPUs have been widely applied to machine learning, especially in the cloud side when TRAINING deep neural networks (which require a whole lot of similar math operations without a lot of logical operations). Self-driving cars, for the most part (this is an extreme over-simplification) don't perform neural network training, but rather execute the PLAYBACK of previously trained (in the cloud) networks downloaded to them (many nets concurrently). They are also characterized by a lot of ordinary computer vision and complex algorithms (e.g., strategic decision making and just plain "intelligence") that require a lot of flexible logical operations running complex code that isn't all parallelizable onto a GPU.

    While GPUs are easy to design because they just duplicate a small functional block many, many times, along with a lot of complex data movement into/out of the processing blocks, what is really needed is a lot of custom logic that is designed precisely for neural network playback, along with a whole lot of general purpose programmable logic (e.g., many CPU cores). Absolutely for sure, NVIDIA is working toward repurposing their GPUs to be amenable for neural net applications, but this hardware is an early adaptation. That's why Tesla and many other companies are working on far more advanced purpose-designed architectures for self-driving. I'm not critical of NVIDIA or Tesla's use of the NVIDIA platform as it is probably the most powerful platform available in early 2016 for inclusion in AP 2.0. I just don't believe it is fully up to the demands of FSD, but I have confidence Tesla will get there. That's just my opinion.

    I do believe the sensor and computing platform will be sufficient for something in the Level 4 range, which is GREAT, and I'll be delighted to get it and happy with my investment in FSD and MS. We would never get that much without Elon shooting for Level 5 aggressively.
     
    • Informative x 3
    • Like x 2
    • Helpful x 1
  8. sorka

    sorka Active Member

    Joined:
    Feb 28, 2015
    Messages:
    4,027
    Location:
    Merced, CA
    Please tell me where it says there that it will be 2018 or 2020 before I get the FSD that was demoed in the video?
     
    • Like x 1
  9. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,812
    Location:
    Los Angeles
    @sorka - It doesn't say 2018 and you can get as mad as you want but if it's 2018 it's 2018. You may force Tesla to buy your car back or whatever but you won't force faster development. I'm taking delivery of a hardware 2 S shortly. It will have less capability than my hardware 1 autopilot - until it has more. It is what it is. I don't expect anything like the demo video til 2018.
     
    • Informative x 1
    • Like x 1
  10. brianman

    brianman Burrito Founder

    Joined:
    Nov 10, 2011
    Messages:
    15,772
    IIRC, Elon said "L5" explicitly in the conference call.
     
    • Informative x 1
    • Funny x 1
  11. brianman

    brianman Burrito Founder

    Joined:
    Nov 10, 2011
    Messages:
    15,772

    It's at the very beginning.
     
    • Helpful x 1
  12. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,812
    Location:
    Los Angeles
    @brianman - Too lazy to listen to entire thing - at what point does he say "level 5"?
     
  13. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,812
    Location:
    Los Angeles
    Nevermind - if I'd actually read what you wrote I'd see you said "at the very beginning." :slaps self:
     
    • Funny x 2
  14. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,812
    Location:
    Los Angeles
    Okay - you're right - he said it. Elon said "Level 5." I'll admit it - he wrote really big check with his mouth. I hope his engineers can cash it...
     
    • Helpful x 2
    • Funny x 1
  15. brianman

    brianman Burrito Founder

    Joined:
    Nov 10, 2011
    Messages:
    15,772
    I'm pretty sure Elon doesn't have anyone working on this.

    Then again, he has had challenges with women and is really scared of certain AI possibilities so maybe....
     
    • Funny x 1
  16. brianman

    brianman Burrito Founder

    Joined:
    Nov 10, 2011
    Messages:
    15,772
    Tesla has also said it's easy to upgrade the battery pack. "Just a swap", right? Great, where do I sign up for a swap to a 100 kWh pack into my Signature S?
     
    • Funny x 2
    • Like x 1
  17. 3Victoria

    3Victoria Active Member

    Joined:
    May 8, 2016
    Messages:
    1,320
    Location:
    Victoria, British Columbia
    Good thread. First GPUs are excellent for processing AI models. There are different kinds of AI, some have preprocessing/AI/post processing, and some just have AI from start to finish. The amount of compute power is relative to the number of channels of info that is getting processed. For the present AP2 hw, the present computing power is likely enough. If it isn't, Tesla has made it easy to swap in a faster processor package. EM seems to be hedging a bit on how soon level 5 will be here. It is possible that it will require additional sensors (say 360 radar), and hence more sensor compute power, and possible additional post-processing power. We are all guessing. I won't be surprised to see Tesla chips and boards.

    OT: (1) Upgrading from AP1 to AP2 requires much++ work rewiring the car and mounting the sensors ... not worth it, just trade-up the car. (2) It seems that it is more work to develop FSD, and specifically limit it for EAP cars, then just developing the FSD.
     
    • Like x 1
  18. calisnow

    calisnow Active Member

    Joined:
    Oct 11, 2014
    Messages:
    1,812
    Location:
    Los Angeles
    Your point is noted. However I will say this - up until now Elon has dealt with - let's admit what *most* of us are who own Teslas - early adopters with a lot of spare cash.

    He knows he's moving to a whole new market with Model 3 - a less forgiving market. I have 7 personal friends with Model 3 reservations. If he tells them "level 5" - it better f*cking work. You don't get to f*ck that crowd over with bullsh*t and snakeoil and then tell them to go buy a new car. It's hard earned money for them, there will be way more of them, and lawsuits will ensue if he doesn't deliver. We tolerate it because the cars are amazing, we are hobbyists to some degree, and we can take the $ hit and keep upgrading to new cars. He won't have the same leeway with the new set of Tesla owners - and there will be an order of magnitude more of them.

    Elon is not stupid and he knows this. I think Tesla will deliver and they have probably built in an easily swappable central board as a hedge should significantly more powerful computing be needed to process the inputs.

    OR it might be the case that what we get right now is less powerful, computing wise, than what the Model 3 ships with - remember it doesn't go into production til July. That's 9 months later than the first "self driving capable" S.
     
    • Helpful x 1
    • Informative x 1
  19. Rarity

    Rarity Member

    Joined:
    Jan 29, 2009
    Messages:
    284
    Some of y'all are crazy. The value in HW2 is in the data that everyone is generating using identical hardware across the fleet. The value increases as the number of cars increases -- a network effect.

    Believe me, once Tesla has a more powerful computer ready to go, they will be begging everyone to come in and upgrade their computer, because they will want everyone to be part of the network of identical hardware. Silicon is dirt cheap.

    As to whether Tesla is ahead, of course they are. Nobody is gathering more data than Tesla.
     
    • Informative x 1
  20. sorka

    sorka Active Member

    Joined:
    Feb 28, 2015
    Messages:
    4,027
    Location:
    Merced, CA
    Not me. I don't have AP2 hardware ;) I do think this is going to blow up in Tesla's face as the average consumer who doesn't read Elon's tweaks or hang out in this forum will be asking why they paid $8K for something that doesn't work after a reasonably amount of time.
     
    • Like x 1
    • Love x 1

Share This Page