Friday, September 29, 2023

AI Impact on Data Centers

The use of internet mechanisms for content apps created a need for content delivery networks, so it is likely that artificial intelligence, perhaps a form of high-performance computing, will shape requirements for data centers. 

 

Area

Expected impact

Electricity consumption

AI applications are typically more energy-intensive than traditional IT applications. This is because AI applications often require the use of high-performance computing (HPC) resources, such as GPUs and FPGAs. 

Computing cycles

AI applications typically require more computing cycles than traditional IT applications. This is because AI applications often involve complex mathematical operations, such as matrix multiplication and convolution.

Storage

AI applications typically require more storage space than traditional IT applications since databases must be accessed to make inferences. 

Data center design

Data centers--in large part--are increasingly designed to support high-performance computing and ability to support AI training and inference operations. Additionally, data centers are being designed to be more energy-efficient and to provide better cooling for HPC resources. 

Edge computing

Generative AI and AI applications are increasingly being deployed at the “edge” of the network, closer to where the data is generated, as well as “on the device,” in large part because inference operations often require lower latency than many other types of apps. 

Cloud computing

Generative AI and AI applications are also driving the growth of cloud computing. Cloud providers offer a variety of AI-specific services, such as pre-trained AI models and AI development tools. This makes it easier for organizations to develop and deploy AI applications without having to invest in their own computing infrastructure. 

Chips

FPGAs and neuromorphic chips are being developed specifically for AI applications. These technologies can provide significant performance and energy efficiency improvements for AI applications.

Content delivery networks

CDNs can be used to distribute AI workloads (load balancing) or storage. 


No comments:

Have LLMs Hit an Improvement Wall, or Not?

Some might argue it is way too early to worry about a slowdown in large language model performance improvement rates . But some already voic...