The compute power for the metaverse is unlike anything we’ve seen and will require a huge amount of support from the physical world to support its offering of an unconstrained digital universe. While this will undoubtedly generate a huge demand for more power and edge computing sites, it will also require billions of gallons of water in the coming years – which quite frankly, we do not have to spare. For context – Google reported that their sites alone consumed more than 4.3 billion gallons of water in 2021.
As consumer expectations for the metaverse include little to no “lag” (aka latency), enterprises looking to capitalize on the opportunities of the metaverse will be forced to establish an edge data center presence across the nation and update any existing hardware to meet this demand. These smaller edge sites will be full of more advanced CPUs and GPUs with a higher thermal power density. Cooling these smaller, hotter spaces sustainably will be challenging. This will be exacerbated for enterprises looking to establish edge sites in regions that are susceptible to drought as they will need to perform a balancing act between maintaining uptime and remaining compliant with local water usage regulations.
When looking for cooling methods, “air” cooling is a bit of misnomer as it actually requires more water than liquid cooling. This is not to say that all liquid cooling methods are completely water friendly, however the benefits of adopting liquid cooling along with software that can provide insights into usage and efficiency are a great way to reduce a data center’s water usage.
In regards to liquid cooling, there are two categories and five approaches to those categories that all data center operators should be aware of:
- Direct to chip (single-phase)
- Direct to chip (two-phase)
- Immersive, chassis (single-phase)
- Immersive, tub (single-phase)
- Immersive, tub (two-phase)
When considering these approaches, one must also take into account the various types of liquid that each use – of which there are three main types: water based, hydrocarbon-based oils and engineered fluids. The latter two utilize dielectric fluids which are the best options for data centers looking to reduce their water usage. Dielectric fluids can be in direct contact with operating electrical components and provide excellent heat removal capability. As an example, oil has been used in electrical transformers for many years and allows for effective heat transfer from the internal coil and core to the outer casing.
Of the five approaches, the direct to chip two-phase and the immersive cooling approaches are the best options for those looking to reduce their water usage.
That said, there are additional considerations that data center operators must take into account when deciding between hydrocarbon-based oils and engineered fluids, including material compatibility, cost, life, safety, and environmental. For example:
- Oils are much cheaper than engineered fluids but tend to have a shorter life span.
- Direct to chip methods require more filtration as the micro channels on cold plates can be clogged due to poor water quality.
- Immersive methods have seen issues with unexpected failures from things like paper labels separating, and plasticizers leeching out of cabling into the oils and engineered fluids.
For new sites and those in harsh environments, immersive cooling is a better option as it captures all of the heat and isolates the equipment from the surrounding air. For existing sites, the rack-based solutions enable an easier transition.
Traditionally edge sites and data centers have been predominantly air cooled. The metaverse, however, requires more power density, therefore making liquid cooling the priority for those looking to capitalize on all that the metaverse will bring.
Regardless of whether you feel the metaverse is critical to enterprise operations or not, there are leaders across all industries who most certainly view it as such, and it is coming. Whether it be gaming, healthcare, defense or communications, making the metaverse a reality for enterprises all starts with data center professionals and we all must have a deep understanding of how our physical world supports the digital world and how it can do so without draining the finite resources we have here in the “real” world.