▲Water-cooled A100 PCIe GPU (Image - NVIDIA)

Water-cooled GPUs for mainstream servers unveiled for building high-performance eco-friendly data centers

Regulations establishing energy efficiency standards are pending in Asia, Europe, and the United States. This is incentivizing banks and other large data center operators to evaluate water cooling. Furthermore, the need for water cooling is increasing not only in data centers but also for cooling high-performance systems embedded in confined spaces, such as in automobiles and other systems.

NVIDIA announced on the 24th that it is launching the first liquid-cooled A100 PCIe GPU in its mainstream server GPU family to meet market demands for high-performance, carbon-neutral data centers.

Equinix, a global service provider managing more than 240 data centers, issued $4.9 billion in green bonds as of last April. These are investment-grade instruments to be applied to reduce environmental impact by optimizing Power Usage Effectiveness (PUE), an industry metric that measures the amount of energy used directly for computing tasks in Equinix data centers.

Data center operators are striving to lower this ratio to a level close to the ideal 1.0 PUE. The average PUE of current Equinix facilities is 1.48, and the best new data centers are known to be less than 1.2.

▲ Equinix is steadily advancing in data center energy efficiency, measured by PUE (inset). (Photo - NVIDIA)

Going a step further, Equinix opened a dedicated facility this January to advance energy efficiency. As part of this effort, the focus was placed on liquid cooling. Liquid cooling, which originated in the mainframe era, is maturing further in the AI era. It is now being implemented in a modern form called direct-chip cooling and is widely used in the world's fastest supercomputers.

■ Efficiency through acceleration

Globally, switching all CPU-only servers running AI and high-performance computing to GPU-accelerated systems could save a massive 11 trillion watt-hours of energy annually. This is equivalent to the amount of energy consumed by approximately 1.45 million households in a year.

NVIDIA is currently enhancing sustainability by launching the first data center PCIe GPU using direct chip cooling. Equinix is certifying the A100 80GB PCIe liquid-cooled GPU for use in data centers as part of a comprehensive approach to sustainable cooling and heat capture. The GPU is currently being sampled and is scheduled for general availability this summer.

■ Cost-effective water- and power-saving liquid cooling

Cooling systems evaporate millions of gallons of water annually to cool the air inside data centers, and data center operators aim to eliminate these cooling systems. Accordingly, water cooling is attracting attention as it enables systems that recycle small amounts of fluid in closed systems focused on major hot spots.

In separate tests, Equinix and NVIDIA found that data centers using liquid cooling can save approximately 30% of energy while running the same workloads as air-cooled facilities. NVIDIA estimates that liquid-cooled data centers can reach a PUE of 1.15, which is an improvement over the 1.6 PUE of air cooling.

Water-cooled data centers can provide twice the computing performance in the same space. This is because the A100 GPU uses only one PCIe slot, whereas the air-cooled A100 GPU requires two slots.

▲ Rack and power cost reduction

Nvidia announced that at least 12 system manufacturers plan to integrate these GPUs into their products by the end of this year. This includes ASUS, ASRock Rack, Foxconn Industrial Internet, Gigabyte, H3C, Inspur, Inventec, Natrix, QCT, Supermicro, Wewin, and Xfusion.

Jack Smith, Head of Edge Infrastructure at Equinix, noted, "This is the first liquid-cooled GPU introduced into our lab, and it is very exciting because customers are craving sustainable ways to utilize AI," adding, "We will turn waste into an asset."