Handling Increasing A.I. Workloads

Blog home
October 5, 2020

Over time, it has been made clear that AI requires a significant increase in processing power. This means that the demand for advanced processor density, more efficient cooling methods, and greater power requirements also increases.

Data centers are already limited in power output. Consider this: The average power draw for a data center running organization applications is ~7 kW. As for AI apps, most use more than 30 kW alone, as some GPUs draw double the amount of power as a CPU.

Cooling requirements have their own issues as well. AI-based servers require processor density to be higher, and are known to run at high temperatures. So you have more chips installed into the same area, running at a higher processing power, which leads to a significant increase in temperature. Then, of course, as demand for more advanced cooling methods increases, so does the power requirements to support such a system.

As an organization, you need to take advantage of new technologies as they become available. However, you may find that your company's infrastructure is limited by its power capacity. In this instance, there are only a couple solutions:

1. Processing data where it resides

The act of moving large amounts of data around requires a high energy output. Relocating data, as opposed to processing in place, can use over one hundred times more energy. Ideally, the goal should be to move data as little as possible, over as small a distance as possible. It simply doesn't make sense to move huge amounts of information to a third-party cloud. If you can put data closer to where it's processed, you can cut costs by a measurable margin.

2. Consider leasing data center space

The majority of enterprises that are doing this are also attempting to utilize AI. Some of the data center administrations are not able to handle the increased power demands of AI, but there are those that have made the necessary transition as well. It may take some effort to locate those that can handle the higher demand, but more and more data center operators are making changes.

Schedule time with us to ensure that you have the power capacity and infrastructure necessary to make the most of your A.I. needs, or whatever your latest project is.