The link you included says, in big bold blue letters, "FALSE".
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
- Status
The link you included says, in big bold blue letters, "FALSE".
dc_h
Active Member
hpw about an app that delays your tweet for 15 minutes and requires your future self approval.
Model 3 is more efficient than S or X. I do mixed highway/city but mostly highway and mostly at 70+ MPH. Occasionally completely floor the go pedal but not often (just a regular LR here not Performance). Lifetime average is 244Wh/mi over the past ~month.
Well, that is either EAP (enhanced auto pilot) or you may have ALSO purchased FSD, for which there is nothing to show to date. Summons is enabled by EAP for all - if they have it. (almost all, there are legacy S with I think V2.0 AP that don't do summons)
V9 of EAP should bring more features to version 2.0 (and 2.5) AP hardware though, which has been lacking.
And, I think only people who paid 5K for EAP AND 3K for FSD (which does nothing at the moment) are apparently going to get any of the new reported hardware that is 10x better than current hardware at visual interpretation "for free". At the moment though that hasn't been confirmed.
Yup, we're seeing the same thing from (your) Technical and (my) gut perspectives. I might not wait for closing actually, you just convinced me we're spot on.
This was shared with me today by a former Tesla bull who became disenchanted with EM/Tesla/TSLA and moved his TSLA money into other stocks......I still hold a good sized position in TSLA but it does make one understand why diversification is a good idea:
Arun Chopra CFA CMT on Twitter
Arun Chopra CFA CMT on Twitter
Oh... I assumed summons was part of FSD (it's driving itself, no?). I have both EAP and FSD so hard to separate and not clear on several other threads either. Anyway, watch for FSD V9 coming soon then the stock bump.
Exactly. Did you know that Apple has added 1.5 Teslas since they hit 1T a couple of weeks ago.
I said not entirely bullshit. Read the rest of it. I concede that it is mostly bullshit.
He's right - I have EAP, bit not FSD and summons works for me - have actually used it for real, not just for show too
OK, fine, but given slightly different parameters, $TSLA could have been much higher and $AMZN could have flatlined.
It's all very easy in hind-sight.
Ah, the good 'ol "cherry pick the start date" technique...
Johan
Ex got M3 in the divorce, waiting for EU Model Y!
It's true but all about framing. I might look very different in the next 5 years, and you could have choosen some time period where TSLA share price growth far exceeded AMZN.
TSLA is still a young company, AMZN is much more mature even though I agree they have some really huge growth potential still ahead of them. What I've read about the conditions for workers in AMZN facilites would worry me a lot as an investor.
RobStark
Well-Known Member
But Amazon doesn't make much money selling stuff, being a retailer.
They make money with services.
I don't think Amazon will have massive market share as a healthcare provider.
Actually, the Pascal GPU is most likely to be replaced if only one Nvidia element is replaced. However I would not be surprised if they replace both the Parker SoC and Pascal GPU with their ASIC, if it has an ARM core(s) or similar to handle the various necessary support tasks to keep the NN logic running.
While it's true that you could simply remove much of what makes a GPU a GPU and use it as a more power efficient NN processor, that's not the optimal solution, and it doesn't sound like it's the path that Tesla took. It sounds like they built a design specifically tailored to not just NNs but the type of NNs they are going to run. There will be very little in common with a GPU design and a dedicated NN processor design, other than the broad strokes of it processing a lot of data in parallel very quickly.
While building the upgrade as a MXM module is possible, I suspect they'll actually just swap out the whole module as it would probably be easier / faster / less training of service techs to do. Possibly they keep the old ones as spares for non-FSD cars or perhaps they remanufacture them at a central location to have the upgraded hardware by swapping out the MXM board, and then use them to upgrade other cars to FSD. I also expect that they'll only need one ASIC per vehicle, they will have sized it appropriately to begin with. Even at 14~16nm, a dedicated NN processor should have no trouble obtaining the needed performance while still being small enough to have high wafer production yields. Going with 7nm will mostly just lower the per-unit production costs by getting more dies from a wafer, though it will also improve power/performance.
While the upfront costs are probably at least several millions (a couple million initially and then a million or so per respin), Tesla is likely able to keep using this ASIC chip design for a decade or more and they will recoup their costs and then some quite easily.
Keep in mind that the desktop GTX1060 (one of several variants of the GP106 GPU) still sells for in the neighborhood of $200~$400 retail, and Nvidia isn't in the business of losing money. The MXM variants are harder to come by and are often more expensive by $100 or more just because they can be. Some of that is due to the high cost of RAM (thanks in part due to collusion among the RAM manufacturers to keep supply limited), some of that to other necessary things for the GPU board - these things are unlikely to get more than a little cheaper for the Tesla ASIC (possibly less RAM needed, possibly less power regulation needed). After dicing the wafer and removing bad dies, the likely per-die cost to Nvidia for the GP106 is somewhere below $30 (best I found on short notice was pricing from 2016 that put it just over $30, yields and such should have improved since then). RAM costs are probably around $50-70, the rest of the hardware is probably under $50. So per-unit cost is probably $100-150 or so.
That may not sound like a lot of room for cost savings, but I'd bet they can save at least $10-15 per ASIC (on account of being likely smaller and getting more usable from the wafer), another $25-35 on ram (using less of it), and another $10-20 on the rest of the supporting circuity (using less power, etc). Shaving off ~$50 before accounting for Nvidia's profit margin means that they can probably break even on the project after only 50-100k units (assuming 2~5$ million initial costs for tape-out, respins, etc). Depending on how much margin Nvidia was taking on top of that, could reach break even earlier than that. Perhaps a worst case scenario they break even at 500k units if they only save $10 between Nv margins and everything else. If they plan to ships millions of cars a year, and use the design for years, this is a no brainer long term cost reducer.
They're most likely not even running the chips at full potential performance yet as usually the first couple of iterations of your design have issues with yield, and it takes a couple of respins to nail it. Though I wouldn't expect a doubling or anything here, just another 5-20%.
As for switching from ARM to x86, I wouldn't bet on it. Maybe from ARM (cheap licensing) to RISC-V (zero cost licensing), but going to x86 doesn't make a lot of sense when you're already building custom silicon. They can put the ARM or RISC-V core(s) on the ASIC and eliminate both the Parker SoC and Pascal GPU, but if they go x86 they still must have two separate devices (or pay even more money for custom silicon from AMD with NN logic embedded into an x86 SoC - not impossible but likely not the optimal solution here). Plus most likely ARM or RISC-V have more than enough performance for the task of keeping the NNs fed and operating, x86 is overkill.
The x86 CPU on the newer AP HW cars I believe is running the UI/UX, not the AP decision making. I think that's still happening on the Parker SoC. It's possible that Parker isn't enough for FSD decision making, but that doesn't mean they would move to x86, you can get higher performing ARM cores and they may move that into the NN ASIC itself anyways.
SpudLime
Active Member
Cart widths were set by having two horses/cows/goats/ox walk next to each other....
Agree, as a 'weak' long
I may add software practice, logistics to their weak points. Rush of delivery did cause some pissed off customers. We learnt on twitter the captain imposed iron rule on quality control before delivery. This is definitely the right thing to do and positive long term, it may cause lower delivery numbers short term.
All these things are not easy to get right. But they have examples to learn from, existing best practice to follow.
Inventing the future, as they are doing now, is much harder.
don’t forget communications with customers. although this becomes markedly easier as their delivery logistics system begins to take hold.
right now it’s still all over the place (i myself have 2 different stories from 2 separate delivery specialists)...still growing pains.
Last edited:
It's on sale, couldn't wait. Will get more if lower Tues then I'll unload a bit on the way up. I place orders in advance - action is too fast and I don't watch the candles. All gut + knowledge of Tesla/M3 here but I really did expect lower prices this week.
I too thought TSLAP was a sure thing, but still didn't go all in because (as I've shared) we're in a delicate market overall + trade wars, earthquakes, meteor, EM health... What do you know, his health did come up.
I too thought TSLAP was a sure thing, but still didn't go all in because (as I've shared) we're in a delicate market overall + trade wars, earthquakes, meteor, EM health... What do you know, his health did come up.
- Status
Similar threads
- Locked
- Locked
- Locked
- Poll
