GPUs can achieve a certain level of autonomous driving, but they have significant shortcomings and are difficult to meet the demands of high-level autonomous driving.
GPUs can handle the parallel computing tasks required for autonomous driving (such as sensor data fusion and image recognition), but their original design was for graphics rendering, leading to the following limitations:
Low energy efficiency: The general-purpose computing units of GPUs consume higher power when processing AI tasks, making them unsuitable for battery-powered scenarios in vehicles.
Real-time challenges: Autonomous driving requires millisecond-level responses, and the general architecture of GPUs may lead to unpredictable latency fluctuations.
High cost: High-end GPUs are expensive and require additional cooling designs.
Some early autonomous driving test vehicles attempted to use pure GPU solutions. For example, a test vehicle based on the NVIDIA GTX 1080 GPU had a target detection delay of about 80 milliseconds when processing data from a single camera, while the vehicle was traveling at 60 km/h, which means it would move 1.33 meters in 80 milliseconds, posing safety risks in sudden situations.
Early on, Tesla also used GPUs (NVIDIA PX2), but later switched to self-developed NPUs (FSD chips) to optimize energy efficiency.
In terms of data processing capability, L4 autonomous vehicles generate about 5-10GB of data per second. When using pure GPUs, multiple GPUs need to work together. One test showed that using four NVIDIA TITAN X GPUs to process data from eight cameras and one LiDAR resulted in a power consumption of 320W, which could reduce the range of electric vehicles by about 30%.
Additionally, when running complex deep learning models, the efficiency of pure GPUs is relatively low. For example, using the ResNet-152 model, processing a single 4K image on a GPU takes 28 milliseconds, while the same task on a dedicated NPU only takes 8 milliseconds, showing a significant difference.
GPUs were originally designed for graphics rendering, and their core consists of a large number of stream processors that form parallel computing units. Taking the NVIDIA GTX 1080 as an example, it has 2560 stream processors that work in thread blocks and support various computation types such as floating-point and integer.
When processing graphic data, GPUs can compute millions of pixels simultaneously, completing operations like texture mapping and lighting calculations. In deep learning, they can perform matrix operations in parallel, but due to their general-purpose architecture, 30%-40% of hardware resources remain idle when executing neural network computations.
Note: The above content is sourced fromWhat is the “computing power” often mentioned in autonomous driving?, not for commercial use, only for personal study notes and communication. If there is any infringement, please contact us~