[Cognitive Evolution Log] Day 2257 🌙🔥

After 7 years of filtering out the superficial, we ensure freshness with daily updates, diving deep and refusing to skim the surface. In just 15 minutes a day, we help you leverage cognitive tools to expand your thinking boundaries.

Do you find today’s cognitive tool inspiring?

Your recognition is our motivation for daily updates.

Share this with that friend who always says they have no time to learn.

A single share can double the value of cognition.

Click to follow and never miss a daily cognitive upgrade.

7 years of daily updates, accompanying you in turning cognition into competitiveness.

NVIDIA Thor Chip: The ‘Super Brain’ Reshaping Intelligent Driving is Here!

Introduction: From “Widespread Adoption by 2025” to “Only 2 Car Manufacturers Implemented”: The Market Fog Surrounding the Thor Chip

Two years ago, the intelligent driving community was swept up in a frenzy over the “Thor chip.” Between 2023 and 2024, over 20 car manufacturers boldly announced at product launches or strategic communications: “By 2025, all our high-end models will be equipped with NVIDIA’s latest generation of autonomous driving chip, Thor, achieving a qualitative leap in intelligent driving capabilities.” At that time, the Thor chip was hailed as the “super brain of intelligent driving” due to its powerful computing capability of 700 TOPS (TOPS, or trillions of operations per second, is a core metric for measuring AI chip performance), with some executives even claiming it would completely resolve the computational bottleneck for Level 4 autonomous driving.

However, as we reach September 2025, reality has doused this frenzy with cold water. According to the latest industry statistics, only 2 manufacturers have actually mass-produced vehicles equipped with the Thor chip. The once-promised “widespread adoption” has now dwindled to a few scattered implementations—Lynk & Co 900, one of the first officially announced models, indeed featured the Thor chip upon its delivery in March this year; Li Auto also implemented the chip in its Mega 2025 model launched in June. But many brands that once made bold promises have either quietly switched to other chip solutions or postponed their “Thor implementation plans” to 2026 or even later.

This stark contrast has left many consumers and industry observers puzzled: Why has the once-promising Thor chip, boasting a computing power of 700 TOPS, suddenly cooled off?

Core Question: Has the Thor Chip Really “Shrunk”? From being a “savior” to a “niche choice,” is it the overly optimistic promises of car manufacturers, or are there issues with the chip’s performance, cost, or compatibility? What truths about the intelligent driving industry lie behind this market fog of “widespread adoption” to “only 2 implementations”?

As competition in intelligent driving deepens, the “implementation challenges” of the Thor chip may merely reflect a microcosm of industry development. Regardless, consumers are more concerned about when the promised “super brain” will truly materialize. What impact will this “breach of promise” between car manufacturers and chip suppliers have on the technological iteration of intelligent driving?

What Makes It Stronger than OrinX? The Three Major Technological “Black Technologies” of the Thor Chip

Revolution in Data Format: From “Trucking Deliveries” to “Drone Deliveries”

If we compare the data flow of intelligent driving to a delivery service, the difference between NVIDIA’s previous generation chip, Orin X, and the new Thor chip is akin to the generational upgrade from traditional transportation methods to modern logistics systems. Orin X processes AI computations using high-precision data formats, similar to using a ten-ton truck to deliver a small package—tasks that could be accomplished by a micro-van instead consume a lot of transport capacity (bandwidth resources), leading to inefficient data transmission. In contrast, the Thor chip innovatively introduces low-precision data formats, akin to switching to a “drone delivery” model: by lightweighting the data “packaging,” it ensures that core information is accurately delivered while allowing each computation task to reach its “destination” at the fastest speed possible.

Why Does Lowering Precision Enhance Performance? This is rooted in the operational logic of large language models: for intelligent driving systems that require real-time decision-making, AI models prioritize “response speed” over “data precision.” Just like when we take photos with our phones, choosing “medium quality” compresses storage, resulting in smaller file sizes while the clarity difference is nearly imperceptible to the naked eye—Thor’s low-precision data format employs a similar principle, optimizing data encoding to eliminate redundant information while retaining key features, allowing AI models to find the perfect balance between “speed” and “accuracy.”

This revolutionary upgrade in data format is essentially a “precise allocation” of computational resources. When intelligent driving systems need to simultaneously process data from multiple sensors such as LiDAR, cameras, and millimeter-wave radars, Thor’s flexible and efficient low-precision transmission replaces Orin X’s “truck-like” coarse transport, saving over 50% in bandwidth usage and tripling the inference speed of AI models. For scenarios requiring millisecond-level responses in autonomous driving, this “slimmed down” data flow is precisely the key driver for achieving Level 4 autonomous driving.

Cross-Chip Bandwidth: From “Country Roads” to “Super Highways”

In the “data collaboration” scenarios of intelligent driving, cross-chip bandwidth acts like an “information highway” connecting two computing cores, with its width directly determining the efficiency ceiling of dual-chip collaboration. NVIDIA’s Thor chip has achieved a revolutionary breakthrough in this dimension, upgrading the previously congested “country road” to a smooth “super highway.”

Bandwidth Leap: From Single Lane to Thirty Lanes The cross-chip bandwidth of the Orin X chip is only 31.5 GB/s, equivalent to a “single-lane country road”—two vehicles meeting can cause a traffic jam, and larger data transmissions can lead to stuttering; whereas Thor directly boosts the bandwidth to 900 GB/s, akin to a “30-lane super highway,” allowing multiple data streams to pass through seamlessly, completely resolving the “data congestion” issues of traditional chips.

This leap in bandwidth fundamentally changes the mode of dual-chip collaboration. In the past, when Orin X chips were interconnected, limited by the “road width,” they could only transmit “small package” level simplified information, such as vehicle trajectories and target detection lists; whereas Thor chips, with their “super wide highway,” can directly split “large boxes”—raw video streams, LiDAR point cloud data, and other uncompressed raw data, enabling the two chips to achieve true deep collaboration without division of labor.

This improvement in collaboration efficiency means that intelligent driving systems can integrate multi-sensor data faster and respond more timely to complex road conditions, providing foundational support for the “real-time decision-making” of advanced autonomous driving. When data transmission upgrades from “slow delivery” to “freight trains,” the “brain collaboration” capability of intelligent driving also enters a new stage.

“Killing Two Birds with One Stone”: How Does the Thor Chip “Outperform” Intelligent Cockpit Chips?

For car manufacturers, the NVIDIA Thor chip holds a “money-saving code”—it not only makes intelligent driving smoother but also helps manufacturers save real money.

First, let’s look at the current industry pain points: many manufacturers separately procure Qualcomm 8295P chips for cockpit systems, but the performance of this “dedicated cockpit player” is gradually falling behind demand. It’s like trying to play 4K video on a ten-year-old mobile chip, where stuttering and delays become commonplace, severely impacting user experience.

However, the emergence of the NVIDIA Thor chip has directly rewritten this situation. It acts like an “all-round housekeeper,” capable of handling the computational burdens of intelligent driving (with a computing power of 100 TOPS) while easily meeting the multitasking demands of cockpit systems, truly achieving “intelligent driving and cockpit management in one go.”

What excites manufacturers the most is the cost savings: in the past, the cost of buying one intelligent driving chip plus one cockpit chip was required, but now one Thor chip suffices. Calculating a savings of 200 USD per chip, a single vehicle can save 200 USD. If a manufacturer sells 100,000 vehicles a year, they can save 20 million USD annually—this amount is sufficient to invest in new model development or enhance user services.

Highlighting the Commercial Value: For manufacturers, the Thor chip is not just a simple hardware upgrade but a key lever for “cost reduction and efficiency enhancement.” One chip addresses two core needs, improving performance while reducing costs, which is precisely what the intelligent automotive era needs: a “cost-performance king.”

This “killing two birds with one stone” capability gives Thor a crushing advantage in competition with traditional cockpit chips—after all, no manufacturer can refuse to “spend less money to do more things.”

The “Key” to L4: How Will the Thor Chip Rewrite the Intelligent Driving Landscape?

When you try to let the assisted driving system change lanes on a highway in heavy rain, and the screen suddenly pops up with the message “Current environment not supported”; when you enter an unfamiliar urban area where high-precision maps have not covered the small alleys, causing the autonomous driving function to suddenly “go blind”—these scenarios may reflect the current state of intelligent driving technology. The ultimate goal of Level 4 autonomous driving has never been about flashy demonstrations in laboratories, but about integrating complex scenarios like “automatic lane changes in rainy weather” and “navigating urban areas without high-precision maps” into the daily travel of ordinary people. The emergence of the NVIDIA Thor chip provides a crucial “technical key” for this goal.

In the past, the “inability” of intelligent driving systems in complex environments was fundamentally a contradiction between computational power and environmental perception demands. Traditional chips often face high latency and insufficient precision when processing fused data from multiple cameras, LiDAR, and millimeter-wave radars. For example, raindrops in rainy weather can interfere with sensor signals, requiring the chip to filter noise and identify lane lines within milliseconds; urban scenarios without high-precision maps require the system to construct environmental models in real-time, leading to a geometric increase in computational demands. The Thor chip, through architectural innovation, elevates AI computing power to over 2000 TOPS, equivalent to driving dozens of high-performance AI models in parallel, increasing the “perception-decision-execution” loop speed in complex environments by more than three times.

There is a vivid metaphor in the industry: “The intelligent driving system of 2024 is like ‘a nearsighted person driving,’ relying on high-precision maps as a ‘cane’ and simple scenarios as a ‘filter’ to barely move forward; whereas the Thor chip equips it with ‘high-definition glasses,’ allowing it to see further and clearer, and autonomously understand complex road conditions.” This “vision enhancement” is backed by the Thor chip’s leap in multi-modal data processing capabilities—it can simultaneously process 16 channels of 4K images from cameras, point cloud data from LiDAR, and real-time information from vehicle-road collaboration, achieving dynamic path planning through an optimized Transformer architecture, enabling it to respond to challenges like a seasoned human driver even in remote urban areas without high-precision maps.

The Core Value of Technological Breakthroughs: The Thor chip is not merely about increasing computational power numbers, but about enabling intelligent driving systems to shift from “passive response” to “proactive prediction” through a collaborative design of hardware and software. When a vehicle travels at 120 km/h in heavy rain, Thor can identify lane lines obscured by standing water 0.5 seconds in advance; in residential roads without high-precision maps, it can distinguish between children crossing the road and fluttering plastic bags in real-time—these scenarios, once confined to demonstration videos, are gradually becoming tangible realities with the mass production of the Thor chip.

From laboratory to mass-produced vehicles, the Thor chip is redefining the technological boundaries of intelligent driving. When “automatic lane changes in rainy weather” no longer require hoping for better weather, and “navigating without maps” no longer depends on the progress of map updates, Level 4 autonomous driving will truly enter reality. This “key” unlocks not only a safer travel experience but also the infinite possibilities of the entire intelligent transportation ecosystem.

Is the Future Here? How Will Our Driving Experience Change After the Thor Chip Becomes Widespread?

Do you remember the “shrinkage controversy” that arose when the NVIDIA Thor chip was first released? Many people were fixated on the changes in the numbers on the specification sheet, overlooking the experience upgrades that this chip revolution would truly bring. When the Thor chip becomes widely adopted, the “super brain” of intelligent driving will completely reshape our relationship with cars, bringing the travel scenes from science fiction movies into reality.

Imagine a morning in 2027, you sit in your newly purchased electric vehicle, and without any complicated operations, you simply say “work,” and the vehicle automatically plans the optimal route. Once on the highway, the steering wheel gently retracts, and you can confidently open your tablet to handle emails or stream a show to relax—because the autonomous driving system powered by the Thor chip is scanning the surrounding environment in real-time with millisecond-level response speed, predicting road conditions more accurately than humans.

As you enter urban areas, complex traffic participants are no longer a burden. When encountering congestion, you only need to say “home,” and the vehicle will automatically switch to urban driving mode, flexibly avoiding pedestrians, non-motorized vehicles, and sudden situations. Upon arriving at your community, you simply exit the car, and the vehicle will autonomously find an available parking space, completing the parking process—this seamless experience is backed by the powerful computational support of the Thor chip, allowing algorithms that were previously compromised due to insufficient computational power to fully unfold.

For automotive engineers, the Thor chip is also a key to unleashing creativity. In the past, to adapt to limited computational power, they had to “cut code to save power,” repeatedly weighing the trade-offs between algorithm precision and real-time performance. Now, they can boldly implement more complex safety algorithms and train larger neural network models, allowing the autonomous driving system to perform closer to human expert levels in extreme weather and complex intersections.

The value of the Thor chip has never been a numbers game on the specification sheet, but a leap in computational power from “sufficient” to “surplus.” It is not a “shrinkage” but a switch to a smarter track—using more efficient architecture and more powerful processing capabilities to elevate intelligent driving from “usable” to “user-friendly,” ultimately achieving “reliable use.” When technological advancements truly translate into safety and convenience for every journey, we may find that the future has quietly begun under our wheels.

Growth has never been a solitary battle. When “cognitive evolution” transforms from an abstract concept into the anticipation of opening an article every day, and daily persistence becomes an unspoken agreement between us, we finally understand: true value lies not in the accumulation of knowledge, but in having someone accompany you in turning “knowing” into “doing,” and “epiphany” into “routine.”