“The exploding volume of data in all areas of life is constantly increasing the energy requirements of data centers and represents a significant ecological and economic burden,” the Information Technology Society (ITG) in electrical engineering and IT association VDE warned in a statement. Does.” Now it has been published. impulse paperIn particular, the currently popular generative artificial intelligence (AI) requires “extreme computing power”. To solve the problem, ITG recommends relying on optical and photonic technologies. These “can play an important role in reducing the energy requirements of data centers” while also meeting increasing demands for speed and computing power.
Advertisement
The Metaverse: Where Virtual Freedom Meets Real Jurisprudence
“Traditional data centers still rely heavily on copper cables for data transmission,” the authors write. an information sheet Of VDE. “Transmitting large amounts of data in this way is efficient only to a limited extent and is therefore energy-intensive.” Copper cables exhibited signal attenuation over long distances. Therefore, amplifiers are required to transmit signals and ensure data integrity. These devices require electrical power supplies and thus contribute to the overall energy consumption of data centers. The potential of alternative optical fiber technologies has not yet been adequately exploited.
Upgrade data channel to 400 Gbit/s
According to ITG, future fiber optic cables will require data channels supporting transmission speeds of up to 400 Gbit/s at a single laser wavelength. To process these bandwidths more energy-efficiently, attention must be paid to alternatives to traditional integrated digital circuits, such as integrated silicon photonics electronic circuits.
According to the paper, such photonic integrated circuits (PICs) have several advantages: The use of integrated photonics can reduce energy consumption in data centers: “Light modulation can be used for communications technology with low energy expenditure ” In addition, data can be transmitted in higher bandwidth in the higher-frequency spectrum of light. ITG explains that converting electrical signals into optical signals provides a number of benefits including increased energy efficiency. Additionally, photonic circuits have lower losses due to heat dissipation, resulting in higher energy efficiency. In this way, the cooling system can be significantly optimized, which in turn reduces the associated losses.
Searching with AI systems will drastically increase power consumption
“Integrated photonics have already found their way into data centers, especially in high-performance computing (HPC) environments,” the information said. “PICs are used to connect servers, storage devices, and networking equipment. Photonic interconnects for data centers are therefore expected to become the industry standard in the near future.”
Damien Dudek, Managing Director of ITG, explains that the current approach can no longer be continued. AI systems require large amounts of energy both to train them and to operate them. If a user requests ChatGPT, it is estimated that they will consume three to nine watt hours. If all nine billion daily search queries were answered by AI, power consumption would increase 30 times.
Nuclear power is not necessary for AI
By focusing on photonics, VDE also wants to counter the efforts of big cloud service providers like Amazon, Google and Microsoft, which increasingly want to satisfy their AI-driven energy appetite with nuclear power. Hyperscaler Google alone wants to buy energy from small modular nuclear reactors (SMRs) from developer Kairos Energy from 2030, the authors say. Annual production should reach 500 MW by 2035. It is still unclear whether power from the reactors will go into the grid or they will be directly connected to data centers.
In the future, ITG hopes that innovations in the field of computer hardware will contribute to increasing computing power as well as reducing energy requirements. For example, efforts are being made to replace the traditional “von Neumann computer architecture” used so far with methods such as neuromorphic computing. However, these efforts alone may not be enough to meet the power demands of AI applications in data centers. Optical and photonic approaches should therefore “make significant contributions” to information and communications technology. The ITG, among other things, mentions in its paper: Thoughts from engineer Thomas Eko in his blog Photonics Report from January,
(usz)
“NRW.Genius”: New AI assistant aims to make state administration more efficient
