Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizen OC_Formula for the clue delivery! CTOnews.com, January 15 (Xinhua)-- if self-driving cars are widely adopted, another uncalculated source of carbon emissions, the computer brain that provides them with computing power, may exceed the emissions of current data centers around the world, according to a new study by researchers at the Massachusetts Institute of Technology (MIT).

Unsplash of course, this is based on the ability to achieve L4 or L5 autonomous self-driving cars, but three MIT researchers say the framework they have established to simulate the carbon emissions of computers in self-driving cars (AV) should draw attention to hidden carbon costs and help the auto industry plan for a greener future.

Researchers at the Massachusetts Institute of Technology say that if self-driving cars can gain as much as 95 per cent of the market by 2050, assuming that the world's 1 billion self-driving cars drive an average of one hour a day and use an 840W computer, one year will generate the total carbon emissions of global data centers in 2018.

According to the International Energy Agency, global data centers accounted for 0.3% of global carbon emissions in 2018, roughly equivalent to the emissions of Argentina as a whole, according to CTOnews.com.

The researchers also found that in more than 90% of simulation scenarios, in order for self-driving cars to emit no more than current data centers, each car must use an on-board computer with less than 1.2 kilowatts, which will require more efficient hardware. If 95% of the world's vehicles are self-driving in 2050 and the computational workload doubles every three years, the world continues to decarbonize at its current rate. the study found that hardware efficiency needs to double at least every 1.1 years to keep emissions below these levels.

The researchers have set up a framework to explore the running emissions of computers on the global fleet of electric vehicles, which are completely autonomous, which means they do not need a backup human driver.

It is worth mentioning that each of the variables in the research model contains a lot of uncertainties. For example, some studies have shown that the driving time of self-driving cars may increase because people can deal with other things while driving. Young and old people can drive longer. But other studies have shown that driving time may be reduced because the algorithm can find the best way for people to get to their destination faster.

In addition to considering these uncertainties, researchers also need to model advanced computing hardware and software that do not exist at present. To achieve this goal, they modeled the workload of a popular algorithm for autonomous vehicles, which is called multi-task depth neural network because it can perform many tasks at the same time. They explored how much energy would be consumed if the depth neural network simultaneously processed high-frame-rate, high-resolution input from many cameras.

When they used probabilistic models to explore different scenarios, the researchers were surprised that the workload of the algorithm increased so quickly. For example, if a self-driving car has 10 deep neural networks that process images from 10 cameras, and the vehicle drives for an hour a day, it will make 2160 million inferences (Inference) per day, and 1 billion cars will make 216,000,000 (quadrillion) inferences. From this perspective, Facebook makes trillions (trillion) of inferences every day in all data centers around the world (1 quadrillion equals 1000 trillion).

In addition, their model only takes into account computers and does not take into account the energy consumed by vehicle sensors or emissions generated during the manufacturing process.

One way to improve computational efficiency may be to use more specialized hardware designed to run specific driving algorithms. In addition, future researchers can also make the algorithm more efficient and require less computing power.

The researchers hope that automakers will consider emissions and carbon efficiency as important indicators in their designs.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.