What will Performance Edition R2 LE equivalent cost with LIDAR version

[DuoRivian]

If you go to the following video and specifically to the section about vision versus LIDAR (29:45) RJ Scaringe answers most of these questions.



Here are some quotes from RJ Scaringe:

"a high performance long range LIDAR you can buy for a couple hundred bucks"

"The most expensive part in a self-driving system is the brain. ... It's the inference platform plus the entire compute platform necessary support that. So, all the associated memory ... and all that consolidated onto a big PCBA, the cooling systems to support that that's far more expensive than the perception stack."

"radars are tens of dollars, cameras are tens of dollars, lidars are hundreds of dollars."

"[LIDAR is] incredibly useful for training your cameras."

"you just have to go El Camino or page and you'll see a bunch of Telsas drive by ... with LIDARS mounted on them and it's part of their ... ground truth fleet for training their models"

"in fact any of the incremental costs that would have been there on its own is offset by the fact that we brought inference in house and reduced the cost of our inference platform so dramatically from [the R1S that] uses an Nvidia inference platform."

"I say all this because in the infinite long-term ... you could make the case that once the models are very very robust you could have less cameras or you may be able to get away with less radar. It's not clear yet if that's the case for covering all these corner cases. I'd say it's unknown but ... in our case it's very clearly accelerates the rate of progress for training the model and it very clearly allows us to deliver level four features"

My takeaways are:

The cameras, radars and LIDAR are for training the model; it is unknown what will be needed for level four self-driving (the inference stack as opposed to the training wheel for the model). I would assume radar would be needed for poor visibility but LIDAR may not be needed.

The RAP1 chip is four times as powerful as the Nvidia chip (800 TOPS versus 200 TOPS) so there might be some reason to wait for it but I do not see any reason to wait for LIDAR which may not even be needed for level four driving (which is still along ways off). But I think the difference would just be the speed at which the inference stack responds to the sensors.

The inference stack with the RAP1 chip is going to cost less than the Nvidia inference stack even with the additional LIDAR sensor so it will save Rivian some money. I do not think they will charge more for the LIDAR and RAP1 chip.

If you are worried about obsolescence I would worry about solid state batteries. I think they are far enough off that they are not a concern but they may be widely available by around 2030. So if you wait for LIDAR then you may then find yourself waiting for solid state batteries.

Good summary and it makes perfect sense to do their own chip and cut out the middle man (Nvidia who have a 70% profit margin). Simplifies supply chain too as they can deal direct with the chip manufacturer.

Sponsored

 

[River Ghost]

Perhaps I should explain a little for those that are not familiar with AI. AI consists of two parts: training and inference. Training uses a huge amount of data and requires a massive amount of computation that requires GPUs (such as Nvidia makes) to accelerate the computation. The training step produces a model. Then the model is used by the inference step to produce the output (i.e. self-driving). The model is put on each car and requires much less computation than the training step. Over the air updates periodically push out a new improved model from the data flywheel (training step).

This model RJ Scaringe calls a Large Driving Model (LDM) as opposed to a Large Language Model (LLM). While it makes sense to cut out Nvidia to save costs there is probably a more important reason for the RAP1. It is optimized for the operations required to run a LDM. So while the RAP1 may be four times as powerful as the Nvidia chip that may be understating the performance because it is specifically designed to handle the types of computation required by the LDM.

 

[AirplaneDr]

Starting with the current Eskar Silver/Black Crater Signature Performance at $59,485 and adding Autonomy+ ($2500), Towing ($1000 est) and Key fob ($250) brings the total to $63,235.

My estimate is that LiDAR will add $3k to the price. So $66,235. I consider both the towing and LiDAR pricing estimate to be on the high side with the hope of being pleasantly surprised.

 

[JasonK]

I suspect it will be more for sure. But I would expect only about 2k to 4K more.

 

[DuoRivian]

I suspect it will be more for sure. But I would expect only about 2k to 4K more.

Plus the $2.5k for Autonomy plus subscription otherwise you won’t see the benefit.

 

[DJFrerichs]

I could of swore I heard this MSRP price stays after launch edition is done when LIDAR & RAP1 are integrated. The cost of Gen 3 hardware is already baked in. Keep in mind, LIDAR is sooooo much more cheaper than years ago ($250 a car I think I heard in past interviews with RJ) and Rivian is making their own Chips (RAP1) that is costing them peanuts compared to buying someone else's chip that is in the Launch Edition R2. Now if you want to add what is being provided in the Launch package after the fact, then yes that will cost you more as "add options". Same goes for if you want the light interior over the black interior or different rims than the 21's included in the base price.

 

[Luxus]

As far as the $45k base model goes, I don't see them tacking on more money for lidar. At least not at first. They have gone out of their way to keep their promise of a $45k starting price. I can't see them throwing that out the window. Not everybody operates like Elon.

 

[JasonK]

I believe the performance will start arround 56 or 57k with no added options. Then you add options to get up arround 60k.

 

[Mark_AZR1T]

$59,900 for the Performance, Lidar, 1-year free Autonomy + and a free key-fob (tow package extra) and the choice of 3 exterior colors. This is the likely promotional deal according to me

 

[therealcmj]

Rivian is making their own Chips (RAP1) that is costing them peanuts compared to buying someone else's chip

The incremental cost of the 2nd through 100,000th chip is low-ish compared to just buying from NVIDIA. But the IP they're licensing from ARM and the "setup" cost for the first chip off TSMC's line is very very expensive. It may work out cheaper overall than buying from NVIDIA, but only because they're going to order enough for every R2 (and probably R1) they make in 2026 and 2027 in one go. And how much cheaper it actually is is still very much TBD and definitely a closely guarded secret by Rivian, ARM, TSMC, and by NVIDIA.

It could wind up being more expensive. And it could still be a smart business decision. Because there are other reasons to bring it in house than just price.

 

[Guuma]

Good summary and it makes perfect sense to do their own chip and cut out the middle man (Nvidia who have a 70% profit margin). Simplifies supply chain too as they can deal direct with the chip manufacturer.

I agree it was nice summary on their Lidar strategy/chip development. From this video, it definitely sounds like the Lidar and new chip set will not increase the overall cost of the R2 given the stated savings from developing their own chip vs buying the NVIDIA (which was indicated to be significant and more than offsets the cost of Lidar).

Lidar discuss starts around 34 mins in to the interview and 34:40 is when he talks about the costs.

Sponsored