Wednesday, July 17, 2024

From Glass to Air: How Hollow Core Fiber is Accelerating AI

Microsoft Hollow Core Fiber

This blog is a part of the series “Infrastructure for the era of AI,” which addresses new developments in large-scale computing and related technologies. This article delves deeper into hollow core fibre (HCF), one of Azure’s newest technologies.

What is hollow core fiber?

In comparison to conventional solid-core fibres, hollow core fibres (HCF) are specialised optical fibres with a hollow centre that guide light largely through the air, decreasing transmission losses and dispersion.

People are thinking more and more about AI, and breakthroughs are coming at a breakneck pace. However, in order for businesses to maintain the current rate of AI innovation, the compute-intensive AI workloads they are attempting to run require the appropriate infrastructure. Microsoft has promised its clients this “purpose-built infrastructure” for artificial intelligence. This commitment extends beyond Microsoft merely putting partner-developed hardware in its datacenters; the company is committed to collaborating with partners and sometimes working independently to build cutting-edge technology that will enable breakthroughs in science and artificial intelligence.

Hollow Core Optical Fiber

Hollow core fibre (HCF), an inventive optical fibre that is expected to optimise Microsoft Azure’s global cloud architecture by providing better network quality, reduced latency, and secure data transmission, was one of these technologies that was showcased at Microsoft Ignite in November.

Airborne transmission

Hollow Core Fiber technology was created to increase global latency and connection while handling the demanding workloads of AI. Its unique construction, which propagates light in an air core instead of a solid glass core like standard fibre, has several advantages. The nested tubes in the Hollow Core Fiber structure are an intriguing feature that aid in preventing undesired light leakage and maintaining a straight path for light into the core.

 hollow core fiber
Image credit to Microsoft Azure

Hollow Core Fiber latency

Hollow Core Fiber is 47% quicker than regular silica glass because light moves through air more quickly than through glass, increasing overall speed and reducing latency. It also has more bandwidth per fibre, but what makes bandwidth, latency, and speed different from one another? Network latency is the duration of time it takes for data to move between two end points across a network, whereas speed is the rate at which data moves across the fibre medium. The response time increases with decreasing latency.

Furthermore, the quantity of data transmitted and received within a network is referred to as bandwidth. Consider the scenario when two cars are leaving at the same time to go from point A to point B. A car (representing single mode fibre, or SMF) is the first vehicle, and a van (HCF) is the second. The data shows that both cars are carrying passengers; the van can accommodate 16 people while the automobile can only accommodate 4. The van can go at a higher pace than the car because of its various top speeds. As a result, the van will get at its destination sooner and experience lower latency during its shorter trip time to point B.

single mode fibre vs hollow core fibre
Image credit to Microsoft Azure

The industry has been committed to steadily but modestly advancing silica fibre technology for more than 50 years. The constraints of silica loss have resulted in modest increases notwithstanding the advancements made. Early in 2024, Hollow Core Fiber technology achieved a major milestone when it achieved the lowest optical fibre loss (attenuation) ever measured at a wavelength of 1550 nm even lower than pure silica core single mode fibre (SMF). In comparison to SMF, Hollow Core Fiber provides lower attenuation, more launch power handling, a wider spectral bandwidth, and enhanced signal integrity and data security.

The necessity of speed

Consider yourself engaged in an internet video game. The game demands fast thinking and snap decisions. Your in-game actions will be promptly transferred to the game server and your buddies if you have a fast, low-latency connection. This will enable you to respond in real time and have a fluid gaming experience. If you have a sluggish connection with high latency, the fast-paced gameplay will be difficult to follow.

There will be a delay between your actions and game events. Lagging ruins the gaming experience and is really frustrating, whether you’re missing important action sequences or are just trailing other players. Similar to this, high-speed connections and reduced latency can improve the efficiency of AI models by enabling them to analyse data and make judgements more quickly.

Lowering the latency of AI tasks

How therefore might Hollow Core Fiber improve AI infrastructure performance? Tasks involving the use of neural networks and machine learning algorithms to process vast volumes of data are referred to as AI workloads. These assignments may include speech synthesis, computer vision, picture recognition, natural language processing, and more. Because AI workloads frequently entail several stages of data processing, including data ingestion, preprocessing, training, inference, and assessment, they necessitate fast networking and low latency. Sending and receiving data from several sources, such as cloud servers, edge devices, or other nodes in a distributed system, may be required at each stage.

The accuracy and speed at which data may be processed and transferred depends on the network connection’s speed and quality. The AI workflow may have hiccups, errors, or delays as a result of a slow or unstable network. Inaccurate results, resource waste, or subpar performance may follow from this. To handle increasingly complex workloads with billions of parameters, these models frequently require enormous processing power, lightning-fast networking, and ultra-fast storage. Ultimately, low latency and high-speed networking can help accelerate model training and inference, enhance performance and accuracy, and promote AI innovation.

Supporting AI workloads globally

For AI workloads that need real-time or nearly real-time responses, such autonomous vehicles, video streaming, online gaming, or smart gadgets, fast networking and low latency are extremely crucial. These workloads cannot tolerate any network slowness or interruption since they must process data and make choices in milliseconds or less. High-speed connections and low latency help guarantee that the data is received and processed on schedule, enabling the AI models to produce precise and fast results.

The practical application of AI is best shown by autonomous cars, which use AI models to quickly recognise objects, anticipate motions, and map routes across challenging environments. Low latency and high-speed connections enable quick data processing and transmission, which speeds up decision-making and improves performance and safety almost instantly. By offering quicker, more dependable, and more secure networking for AI models and applications, Hollow Core Fiber technology can improve AI performance.

Regional ramifications

There are further ramifications beyond the specific technology that powers your AI models. Datacenter locations are costly, and Azure and the customer have different requirements when determining where to locate these datacenters. Distances between regions and the customer might also have a significant impact. Higher latency occurs when a region is situated too far away from a client since the model has to wait for the data to go to and from a more distant centre.

When considering the vehicle vs. van comparison and how it applies to networks, we can see that more data can be sent between two places in a network in two thirds of the time by combining larger bandwidth and quicker transmission speed. As an alternative, Hollow Core Fiber provides greater reach by up to 1.5 times the transmission distance in an already-existing network without affecting network performance. In the end, you can travel farther with more data while maintaining the same latency envelope as standard SMF. Azure users will be greatly impacted by this since it minimises the requirement for close proximity to datacenters without sacrificing performance or raising latency.

The framework required for the AI era

In order to better Azure’s worldwide connectivity and handle the needs of AI and future workloads, Hollow Core Fiber technology was developed. It provides end users with a number of advantages, such as enhanced security, better signal integrity, and greater bandwidth. HCF technology can facilitate quick, dependable, and secure networking in the context of AI infrastructure, which will enhance the efficiency of AI workloads.

Infrastructure technology continues to be a vital component of the jigsaw as artificial intelligence develops, guaranteeing effective and safe connectivity for the digital age. AI users are increasingly looking to take use of new technologies like Hollow Core Fiber, virtual machines like the recently introduced ND H100 v5, and silicon like Azure’s own first partner AI accelerator, Azure Maia 100, as AI breakthroughs continue to put further demand on existing infrastructure. All of these developments lead to faster data transit, more effective processing, and eventually, more potent and responsive AI systems.

Hollow core fiber price

The cost of hollow core fibres can vary considerably depending on several aspects, such as the fibre type, specifications, manufacturer, and purchasing amount.

Below are several estimated pricing brackets:

  • Standard hollow core fibres typically have a price range of $50 to $200 per metre.
  • Advanced hollow core fibres, which are specifically built for applications like ultrafast laser delivery or medical use, can cost anywhere from $300 to $1,000 per metre or even more.
  • Custom or specialty fibres, which are specifically produced with unique features, may cost over $1,000 per metre. The price depends on the intricacy of the fibre and the desired performance standards.

Engaging in bulk purchasing or entering into long-term contracts may provide certain cost reductions. To obtain accurate pricing, it is advisable to directly contact manufacturers or suppliers, such as NKT Photonics, OFS Fitel, or Lumenisity. They are capable of providing customised prices based on specific requirements and volumes.

Follow our website series to learn more about these emerging technologies, the reasons behind our current investment strategy, your personal implications, and how they make AI workloads possible.

Since June 2023, Drakshi has been writing articles of Artificial Intelligence for govindhtech. She was a postgraduate in business administration. She was an enthusiast of Artificial Intelligence.


Please enter your comment!
Please enter your name here

Recent Posts

Popular Post Would you like to receive notifications on latest updates? No Yes