Buckminster
Well-Known Member
My MY engaged emergency automated braking today at ~5mph. Wasn't strictly necessary. Felt like hitting a brick wall. Impressive.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Investor Engineering Discussions
- Thread starter MC3OZ
- Start date
Buckminster
Well-Known Member
Did you save the video clip by chance?
Also, I didn't know that it is active above 3MPH, used to be much higher...
In the context of reducing Model 3 cost 3k-7k by going from stamped to cast, the areas that can be changed:
BOM cost
Amortized equipment cost
Appropotioned factory cost
Amortized labor cost
BOM cost: with existing tooling, stamping cost is roll stock and handling, cheap.
Casting requires gigapress, install, heat source, and ingots. Cheap once amortized. Casting would also cut maintenance and welding/ bonding/ riveting consumables.
Employees on the line: The shorter Y line is due to 1/3 reduction in body line robots per casting. However, that does not carry over to employee count. From a 2018 Bloomberg report: “Tesla says the Model 3 body line is now 95 percent automated, including the transfer, loading, and welding of parts.”
Maybe they reduce people on the stamping line, but then it would be underutilized.
To cut $2k per car with fully burdened wages of $100 an hour requires 20 hours per car personnel reduction, doesn't seem reasonable.
Throughput: this assumes the body line is the bottleneck as opposed to GA or paint. If Tesla is 30 hours of labor per car and could double the station/ operator speed, that's only $1,500 in labor savings.
Non-cycle limited equipment amortization would also halve, but assuming 500 million in cost for a line that builds 250k a year and has 5 year life: 500 million/1.25 million cars = $400 per car. Doubling speed only saves $200. If line were $2 billion, it's $800 in savings.
Also, takt time doesn't depend on number of stations (unless subdividing task to reduce it).
Factory space gain: Only reduces assigned factory cost if it is proportional to floor space versus unit production. New capacity helps Tesla, but doesn't directly reduce the 3 cost. SG&A may be unit proportional. Might take a full line reconfiguration to be usable. Even then, body line is only a potion of the total vehicle line.
If space is not utilized then cost shifts to all other vehicles.
Rework: do we have numbers for the current cost per vehicle? Can't be reduced more than that.
Implementation cost: switchover means taking the line down, especially to optimize factory footprint. The cost is amortized, but still reduces gains.
Yeah, I allowed for reuse to offset CapEx cost. However, the line is already 5 years old so the robots have been mostly depreciated by now anyway. Also seems a new line would want longer life new equipment, not to mention logistical impacts of developing the new line with old robots.
bkp_duke
Well-Known Member
My source is from a friend in Tesla management. I will honor his request to remain nameless, but this is a person that has been on-stage with Elon.
It's fact, if they transitioned the Model 3 over to a cast setup, there would be substantial savings in both space, manpower, and $$$.
Also, bear in mind, for Fremont, half or more of the Model 3 robots are all fully depreciated, as that line has been churning out cars since late 2017 (5 year point - which is often the depreciation schedule for equipment like this).
EDIT - Elon himself even said with an interview with Sandy Munroe he would love to transition the Model 3 over to a cast setup for cost savings and improved throughput, but it wasn't practical because of the downtime involved when they have so many orders to fulfill.
It's fact, if they transitioned the Model 3 over to a cast setup, there would be substantial savings in both space, manpower, and $$$.
Also, bear in mind, for Fremont, half or more of the Model 3 robots are all fully depreciated, as that line has been churning out cars since late 2017 (5 year point - which is often the depreciation schedule for equipment like this).
EDIT - Elon himself even said with an interview with Sandy Munroe he would love to transition the Model 3 over to a cast setup for cost savings and improved throughput, but it wasn't practical because of the downtime involved when they have so many orders to fulfill.
Yeah, I can totally see it for a new line, it was conversion of an existing depreciated setup that I had trouble swallowing. Similar to Elon's comments, it's inefficient, but it works.
bkp_duke
Well-Known Member
If memory serves, Tesla has 2 lines for the Model 3 in Fremont.
I would be possible to take one down at time and completely retrofit for casting usage (although, they would need to find room for casting machines - Fremont is a trainwreck in terms of layout and available space).
In practice, I would love to see a new NA giga announced soon, and that giga get a "second gen" Model 3 line that has all the trimmings, like the Model Y lines in Texas and Berlin.
I think this is the issue.
Making Model Y with front and rear castings at Fremont is the priority.
I still wonder if Berlin will make Model 3 with front and rear castings and a structural 4680 battery pack. If Berlin was going to make Model 3 this seems like the logical place to to do it. But it would probably be after 4680 cell production was ramped in Berlin.
It 'could', but would it work well?
The Tesla FSD TRIP chip is built around 8 bit ints. SD is float32 or float16.
As to efficiently/ natively:
Tesla FSD chip TRIP is based around 8x8 int multiplies with a 32 bit accumulator. Stable Diffusion uses float32 (or float16 for memory limited systems) weights.
Reminds me of running Autocad on a 486SX25 with math co-processor emulation.
Yeah... but the question was "run on", not "display the results from"... i.e. web browser.
At least the refresh Model S&X have a discrete GPU for gaming that would probably provide decent computing power...
Oh yeah, there are lots of CPUs and GPUs running around. TRIP could probably do it too by breaking up the math like on an standard 8 bit processor.
Does no one downshift any more?
Doesn't 1st gear go up to ~30mph (at redline) on most cars? On some twisty roads, that's still too fast.
The Limiting Factor Patreon has a new video showing how Tesla 4680 side cooling is actually better than bottom plate cooling that Rivian and Lucid use.
Similar threads
- Marketplace listing
