Artificial intelligence is challenging rack power density and battery backup

Artificial Intelligence (AI) takes an awful lot of processing power. That’s putting real pressure on data center rack power density and rack-based battery backup.

 

The application of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) is business as the new usual. The rapid adoption of these applications has created an explosion in data center workloads. OpenAI—an artificial intelligence research laboratory—has released an analysis showing that since 2012, the amount of compute used in the largest AI training runs has grown by more than 300,000 times[1].

 

In other words, the computer resources consumed by AI has doubled every 100 days.

 

The AI age is skyrocketing on rack-based servers built around multiple CPUs/GPUs (with hundreds of cores integrated with terabytes of memory, multiple high-speed communication channels, DDR5 RAM memory, 100G+ Ethernet and other advances that all need increasing power.

 

What does this have to do with battery backup?

 

Powerful Racks Rely on Integrated Battery Backup

The massively increasing compute workload is gravitating to companies focused on vast cloud businesses. There are already over 500 hyperscale data centers operated by these firms[2]. They, in turn, are opting for power infrastructures with battery backup distributed out to individual racks, in architectures such as those defined by the Open Compute Project (OCP).

 

Racks need more backup power, and the backup power is moving into the racks.

 

As a result, power density is becoming the topmost criteria for battery backup. While lead-acid battery technology has been the workhorse for decades, newer technologies can deliver more power density in the rack. Nickel-zinc (NiZn) technology, in particular, has specific advantages over lead-acid solutions – and lithium-ion chemistry as well – in terms of performance, safety, reliability, cost, and sustainability.

 

ZincFive NiZn battery backup solutions offer dramatically higher power density than lead-acid batteries when measured by either weight (Watt hours per kilogram) or by volume (Watt hours per liter). NiZn batteries have two times the power density and half the weight of lead-acid batteries. So a NiZn battery is about half as big of a comparable lead-acid battery—saving room in the rack for AI-churning servers.

 

More Power and Higher Safety With NiZn

While power density is critical, safety is the other side of the coin. Placing battery backup in the rack, instead of in a separate UPS facility, heightens the need for entirely safe operation to protect employees and equipment.

 

Data center operators concerned about the possibility of thermal runaway in lithium-ion batteries should note that NiZn batteries have been rigorously tested to the UL 9540A test method at cell level, and they did not exhibit thermal runaway in any of the five arduous and destructive test types in that test method. Comparable lithium-ion battery systems require a Battery Management System (BMS) to manage safe battery operation during UL 9540A testing.

 

Cell-level safety is a clear advantage when you put a battery next to expensive rack servers.

 

There’s more to the story, in terms of reliability, maintenance and sustainability. Get all the details in our new article, High Performance Computing Density Drives High Demands on Rack-level Battery Backup.

 

The high-density batteries for high density compute

As rack-level compute density drives power density levels continuously upward, backup battery requirements rise accordingly. The high-power density of NiZn batteries makes them the ideal choice, along with with smaller size and weight to allow easier integration into server racks. They also can reduce operating expenses by virtue of their superior safety, high reliability, wider operating temperature range, long life and simpler maintenance requirements. By simplifying the challenge of high density battery backup, NiZn batteries are the better choice for the hyperscale data center operators dealing with the boom in AI.

 

Read more about how to meet the high performance computing power density challenge.

 

[1] Programmer Info – ‘An Exponential Law For AI Compute’

[2] ZincFive White Paper – Optimizing Data Center Operations with NiZn Backup Technology