Tesla reportedly places massive order of next-gen self-driving chips with TSMC
Tesla has reportedly placed a massive order of chips for its next-gen Full Self-Driving (FSD) computer with Taiwan's TSMC.
electrek.co
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
electrek.co
AlanSubie4Life
Efficiency Obsessed Member
“the automaker said that it is not required to achieve Full Self-Driving – though it would improve performance.”Tesla reportedly places massive order of next-gen self-driving chips with TSMC
Tesla has reportedly placed a massive order of chips for its next-gen Full Self-Driving (FSD) computer with Taiwan's TSMC.
Not sure when they said this exactly, but perhaps someone can find the quote.
Not even sure what to make of that statement. I don’t believe it. I guess it depends on the definition of FSD though.
In what way does it improve performance? Does it get in fewer accidents? But the number of accidents with HW 3.0 will be acceptable, so it is not required?
Maybe they’ll re-evaluate this balancing act after they go wide with FSD. Just see what happens!
Although I also cannot find a previous reference by Tesla to HW4 improved performance, the most likely way to do that is in milliseconds and velocity.
Currently FSD beta is limited by how fast it can react to oncoming objects. If you increase the computing power, especially if balanced with an increase in resolution, the reaction time at a given resolution will go down and the speeds allowed should go up.
Sounds like 4x resolution and 3x computing power is that trade off of constraints.
I’t will be interesting to see if HW4 at 4-5nm even ships on the legacy air cooled (pre 2021 liquid cooled) format or if the heat produced requires some additional die shrinkage at even smaller gate sizes to get down to thermal levels allowed in our older cars.
Last edited:
Tam
Well-Known Member
It's like saying a 3 year old can walk but an 8 month old can also walk.
The difference is performance, stability, and no longer falling with each few steps.
Plain FSD cannot turn left at an intersection but FSD beta can.
However, that is a task that a $15,000 FSD is expected to do: not just goes straight only but also make turns too.
But its current performance is poor: If you don't monitor it, it can get into an accident.
Yes, performance as in accident probability.
NanaimoMY
Member
Translated:
Tesla's strongest autopilot chip is exposed, 5nm process capacity has tripled, mass production next year Smart Car Reference ·2022-11-21 20:02 focus on TSMC foundry, computing power above 400TOPS Tesla's next-generation "brain" for autonomous driving -- The latest self-developed self-driving chip, which is the core of HW 4.0 revealed by Musk in 2020, has the latest progress. Unsurprisingly, the new product is a huge improvement over the current FSD chip, and TSMC is a foundry. Unexpectedly, after NVIDIA Huang Renxun threw out the self-driving chip "Wang Bang", Musk changed his original plan and chose to follow up. Tesla's new brain Tesla's latest self-driving chip has at least completed design verification. Relevant news exposure shows that TSMC has undertaken a huge order for Tesla's self-driving chips. Judging from the conventional process of chip production, such news shows that the new generation of FSD chips is likely to have been successfully taped out. There is other hidden information in this message. As we all know, the current 14nm process Tesla FSD chip has been produced by Samsung. However, the latest self-driving chip order fell to TSMC, and Samsung's name did not appear. Most importantly, the latest Tesla order undertaken by TSMC uses a 5nm process. Two points are important. First of all, it can be basically 100% sure that this batch of orders must be Tesla's latest autopilot chip. Because Samsung also has 4-5nm process mass production capacity, but the yield rate is lower than TSMC, it is reasonable to be abandoned by Tesla. Second, the capabilities of Tesla's new self-driving chip may have improved beyond expectations. Because when HW 3.0 was released in 2019, Musk revealed that the next generation of chips will use the 7nm process, and the current situation is that Tesla directly adopts a more advanced process. why? Don't you see Lao Huang just dropped the self-driving nuclear bomb - the 2000TOPS DRIVE Thor, which uses at least 5nm technology. From any point of view, Tesla, the leader in autonomous driving, cannot be left behind. Tesla's new chip, what level? It can be divided into two dimensions, horizontal and vertical. In the vertical dimension, the current FSD chip has a computing power of 144TOPS. It is manufactured with Samsung's 14nm process technology, and includes 3 quad-core Cortex-A72 clusters, a total of 12 2.2 GHz CPUs, a 1 GHz Mali G71 MP12 GPU, and 2 2 GHz neural processing unit, and various other hardware accelerators. FSD supports up to 128-bit LPDDR4-4266 memory. According to the news, the performance of the new self-driving chip will be about three times that of the current self-driving chip. The performance here may refer to the comprehensive energy consumption/computing power parameters, but it does not rule out that it refers to the single-chip computing power. If so then the new chip is likely to reach 400-500TOPS. In addition, for the task characteristics of autonomous driving, the new FSD chip will also be optimized for AI computing. In a horizontal comparison, if the latest autonomous driving chips start mass production in 2023 and start mass production in 2024, they will still be at the leading level in the world. Nvidia's 2000TOPS nuclear bomb will not start mass production until 2025 at the earliest. At this stage, Orin's single-chip computing power is 256TOPS. For OEMs that require large computing power for autonomous driving, they generally use multiple chips. Qualcomm's latest Snapdragon Ride has been launched on the Great Wall Wei brand, with a computing power of 360TOPS. Not long ago, Qualcomm released the Snapdragon Ride Flex series with a computing power of 2000TOPS, but the mass production time was not disclosed. Therefore, Qualcomm’s move was also interpreted as a forced response after being pressured by Nvidia’s nuclear bomb, in order to maintain market confidence. For domestic players, Horizon Journey 5 is based on TSMC’s 16nm process, and the AI computing power can reach 128TOPS. In 2023, it is planned to launch Journey 6, with a computing power of 1000TOPS, but the time for mass production and boarding may be around 20205. Beyond the horizon, Huawei is another important player. MDC 810, with a computing power of 400TOPS, has been mass-produced. The MDC 810 is not equipped with a GPU that does not support general computing, but uses Ascend, an AI chip with a "specific domain architecture", to be responsible for computing. The products of Black Sesame Smart and Xinchi Technology are still in the stage of catching up with NVIDIA Orin. Another old autopilot player, Mobileye, is far behind in terms of paper parameters. The mass-produced products in 2025 are only planned to reach 176TOPS. The ride-on project was also snatched up by other rising stars. Therefore, Tesla's latest self-driving chip, which will start mass production in 2023, with a computing power of 400-500TOPS, will lead the world for at least two years. Tesla's strongest brain, what kind of "soul" is it suitable for The new brain will undoubtedly help FSD's capabilities take a big step. The latest FSD Beta V11 version has just been released, and there are 8 new features: 1. FSD Beta can be used in high-speed scenarios. Unifies the visual and planning stacks on highway and off-highway, replacing the traditional highway stack that has been in use for over four years. Previous highway stacks relied on several single-camera and single-frame networks and could only handle simple lane-specific manipulations. The new FSD Beta multi-camera video network and next-generation planner allow for more complex agent interactions while reducing lane dependencies, making way for adding more intelligent behavior, enabling smoother control and making better decisions. 2. The Occupancy Network has been improved to recall the data of close-range obstacles and the accuracy in bad weather conditions, the spatial resolution of the transformer has been increased by 4 times, the capacity of the image feature matcher has been increased by 20%, and the side camera has been calibrated , and added 260,000 video training clips (both real and simulated). 3. Improve vehicle merging behavior by leveraging lane shapes and lane boundaries, correlation with general map information, and better gap selection algorithms to provide a smoother and safer experience. 4. Added highway behavior to stay away from common obstacles such as blocking lanes and road debris, while also making switching between lane departures and vehicle lane changes smoother. 5. Improved speed-based lane change decisions to better avoid slowing traffic in fast lanes and reduce disruption to navigation. 6. Reduce the speed-based lane change sensitivity in CHILL mode. 7. Improved lane changing for higher jerk (jerk) maneuvers when it is necessary to stay on course or stay away from blocked lanes. 8. By using numerical techniques for more efficient computation, the delay of trajectory optimization is reduced by an average of 20% while maintaining the existing performance. In addition to the refinement of individual function optimization, the most important progress is to expand FSD to high-speed scenarios to achieve connection with urban road scenarios. This also shows that the FSDV11 version being tested at this stage already has the ability to automatically drive from P gear to P gear in theory. Musk's original words are "you can reach your destination without touching the vehicle controls." Therefore, Tesla's latest autopilot chip will definitely be able to support the mature mass-produced version of the FSD software, and truly fulfill the "full autopilot" promised by Musk N years ago. One more thing Tesla's self-developed chips: Earlier in 2014, Tesla still used the Mobileye assisted driving chip EyeQ3, with a computing power of less than 1TOPS. But since the Tesla Model S crashed into a truck in 2016, Tesla has parted ways with Mobileye and has moved into the arms of Nvidia's Drive PX 2. The computing power is 24TOPS, and it has soared. But even the Drive PX 2, known as the super car computer, failed to become the "perfect self-driving chip" in Musk's eyes. There are still three problems: high cost, high power consumption, and the computing power cannot fully meet the demand. In 2019, Tesla officially broke up with Nvidia, released its self-developed Hardware 3.0 hardware and said "this is the best chip in the world", with a computing power of 144TOPS. This is also the current Tesla-assisted computing hardware. In 2020, it was reported that Tesla is cooperating with Broadcom to develop Hardware 4.0 chips, which are expected to use TSMC’s 7nm process and be mass-produced in the fourth quarter of 2021. In 2021, Tesla confirmed that the autopilot chip will continue to be handed over to Samsung, but not HW4.0. In 2022, it is reported from the supply chain that the outsourcing of Tesla HW4.0 chips will be transferred to TSMC, which will be built using the 4nm/5nm process. This article is from the WeChat public account "Smart Car Reference" (ID: AI4Auto), author: Jia Haonan. 36氪 is authorized to publish. The opinions of this article only represent the author himself, and the 36Kr platform only provides information storage space services.
Listen
content_copy
share
NanaimoMY
Member
I believe it would be 2024 as TSMC is building their factory in Arizona since 18 months ago, and ideally the factory (3nm) will be completed in 2024. For 4nm/5nm HW4, likely 2023.
Not sure if Tesla will get HW4 from Samsung though.
Last edited:
DanCar
Active Member
NanaimoMY
Member
According to the News, it's for FSD.
Overall, more than 1000 people (TSMC, with family members) in Taiwan will join the new factory in Arizona. ~ News in Taiwan.
Arizona is gonna need a lot more boba shops and noodle restaurants.
Similar threads
