GIGABYTE Deep Dive: How We Built Our Industry-leading Liquid Cooling Solution

In the artificial intelligence (AI) age, server cooling is key to getting better return on investment (ROI) when purchasing advanced processors. GIGABYTE Technology is a proven pioneer in server cooling. Our direct liquid cooling (DLC) and immersion cooling solutions are used by clients such as a world-renowned semiconductor giant based in Taiwan, the Japanese telco leader KDDI, and the German Aerospace Center (DLR). In this article, we are pleased to interview our DLC experts and discuss the three pivotal "customer-centric insights" that transformed GIGABYTE's liquid cooling offerings from a "product" into an "all-in-one solution".

Learn More:
《Dive into GIGABYTE's Advanced Cooling Solutions》
《Ready for a Smart Revolution? Visit GIGABYTE's AI Solution Page》

● Question: Why should data center operators and IT professionals consider adopting liquid cooling?

● Answer:

Here is a list of liquid cooling components that are not present in air cooling: the cold plates, which transfer thermal energy from the servers' heat-generating elements, such as the CPUs and GPUs, to the coolant; the cold loops, which carry the coolant in and out of the servers; the leak sensor boards; the rack manifolds, which circulate the coolant to the coolant distribution units (CDUs); and the CDUs themselves, which may be positioned at the top or bottom of the rack if they're "rack-based", or separate from the rack if they're "row-based" and can provide cooling for multiple racks.

Among these components, cold plates and manifolds are dependent on server models. This means they may have to be designed from scratch. They are then manufactured, assembled, and tested. While CDUs are not specific to server models, they also take time to produce and deliver. These are all important factors that must be considered.

The three primary reasons for adopting liquid cooling are: a, better performance due to higher TDP and compute density; b, a smaller carbon footprint due to improved power usage effectiveness (PUE), which helps to achieve green computing; and c, lower operating expenses (OpEx) and total cost of ownership (TCO). These advantages are so significant in the age of generative AI and high-performance computing (HPC), where companies across sectors are developing AI with accelerators like AMD's Instinct™ MI300 Series, Intel® Gaudi® 3, and NVIDIA Hopper and Blackwell GPUs, a growing number of customers are getting on board with the fact that liquid cooling will become mainstream.

● Question: What is the first "customer-centric Insight" that defines GIGABYTE's liquid cooling solution?

● Answer:

Having described the entry barriers to adopting liquid cooling, it cannot be stressed enough that the first step to creating a successful all-in-one DLC solution does not begin with GIGABYTE alone. It begins with communicating with and understanding our customers.

When a client makes an inquiry about liquid cooling, we initiate a holistic evaluation of their data center infrastructure. We cover things such as material design, data center installation, network layout, and software testing. We also discover to what extent the customer is willing to modify their existing cooling hardware. If they would prefer minimal changes to the air-cooling infrastructure, we might recommend a liquid-to-air DLC solution. We may also suggest installing Rear Door Heat Exchangers (RDHx) on existing racks so the operator may gradually transition from air to liquid cooling.

Using our findings as a basis, we propose new ways to set up the data center that's in accordance with the client's power, space, and infrastructure limitations, as well as their compute target. Then, working within the framework of the client's requirements, we provide a selection of suitable liquid-cooled servers, along with options for server placement, infrastructure power supply, and thermal management. This procedure ensures that the client can achieve the best overall performance and enjoy the most optimized PUE.

It is only after completing this process that we move on to creating a bespoke DLC solution. This is where our second "customer-centric insight" becomes key.

Learn More:
《Beyond Liquid Cooling: Break New Boundaries with GIGABYTE's Immersion Cooling》
《Tech Guide: How to Pick a Cooling Solution for Your Servers?》

● Question: What is the second "customer-centric insight" that defines GIGABYTE's liquid cooling solution?

● Answer:

As we've said, the selection of cold plates depends on which server components they are cooling. Judging by whether the component is the CPU, GPU, or memory; and based on the client's choice of processors and the layout of the chips—the cold plates may be very different. GIGABYTE not only has the capability to design and make our own cold plates, but we also work with our verified ecosystem partners to integrate their offerings.

We use this as an example of the second insight that goes into creating the ideal all-in-one DLC solution. Our solution must always follow the customer, not the other way around. We make all the necessary adjustments to perfectly meet our client's requirements.

Another example is the rack manifolds. Reliable quality and stable, balanced delivery of the coolant are just the basics. GIGABYTE's manifolds can be installed on the front or rear side of the rack. When required, they can even be installed in the servers. They boast a proven stainless-steel design and dry-break quick disconnect features for the safe maintenance of the server nodes. Leak sensor boards, which help our clients control risk, also adhere to this principle. GIGABYTE's leak sensor boards are compatible with different server models and can be installed hassle-free.

One of the new products born from our insight into our customers' actual needs is GIGABYTE’s DLC Racks. Our versatile designs support front and rear rack manifolds. They range in size from the DL90-ST0, a 22U (rack unit) half-rack design that can support systems with a power draw of up to 14.5kW, to full-rack solutions that can handle four power distribution units (PDUs) in a single rack. These products can provide our clients with L11 multi-rack solutions for liquid cooling. The GIGAPOD, GIGABYTE's cluster computing solution for superscale AI training and AI inference, also leverages liquid cooling to improve compute performance and density. It is the pinnacle of AI development and advanced cooling.

Learn More:
《Tech Guide: How to Get Your Data Center Ready for AI? Part One, Advanced Cooling》
《5 Must-Know Technical Terms Before You Choose Data Center Cooling Solutions》

● Question: What is the third "customer-centric insight" that defines GIGABYTE’s liquid cooling solution?

● Answer:

While GIGABYTE is proud of its R&D and manufacturing capabilities, when it comes to liquid cooling, it should be clear that there are as many unique requirements as there are customers. We work with various industry-leading ecosystem partners to make sure we put together the most suitable bespoke solution for our clients. By moving manufacturing and integration to GIGABYTE, our solution can be delivered with a shorter lead time—and at a better price.

For example, we work with the industry veteran Stäubli to offer quality-assured connectors in our DLC solution. We partner with global innovators in IT cooling, such as Motivair, nVent, and Delta, to incorporate their best-in-class CDUs to deliver cost-effective and high-density cooling. A single rack-based liquid-to-liquid CDU that only takes 4U of space can cool a full rack of GIGABYTE's most advanced GPU Servers. GIGABYTE can install the CDU at the bottom or top of the rack according to customer request; we can also adjust the manifolds based on the CDU's position. Installing, adjusting, and regular maintenance of the CDU are part of GIGABYTE's DLC installation service and all-in-one solution.

Combining components from different vendors—from cold plates to manifolds to CDUs—to build a total solution is a lot of work. As the branded server company that directly services our clients, it falls on us to demonstrate value chain leadership. GIGABYTE needs to define the materials used in the solution, such as the coolant and the connectors, to avoid problems with compatibility or coolant erosion. In short, we must maintain quality control over the whole solution to deliver the very best to our customers.

The team holds conferences with our value chain partners many times a month to go over the design of our solutions. Through communication and management, we ensure that the quality of every component complies with the highest industry standards, guaranteeing a smooth development, production, and installation cycle for our DLC solution.

● Question: What is the last step in completing a successful DLC project?

● Answer:

After we've designed, produced, and tested the DLC solution, it needs to be installed in the customer's data center or server room. Regular service and support are required to deliver customer satisfaction.

GIGABYTE has extensive experience in data center installation. As mentioned earlier, we create our DLC solutions to perfectly meet the users' requirements. By working with our ecosystem partners and vendors across multiple regions, we provide global installation services and support.

Once the DLC solution is up and running, if any situation arises, GIGABYTE's R&D team and field application engineers (FAEs) have a direct line to the main body of our component vendors, ensuring maximum efficiency and effective troubleshooting.

Only after the DLC solution is flawlessly installed and in operation, and only after any possible issues have been satisfactorily resolved, can we pat ourselves on the back and say: "That's the wrap on a successful DLC project!"

Thank you for reading "GIGABYTE Deep Dive: How We Built Our Industry-leading Liquid Cooling Solution". We hope this article has been helpful and informative. For further consultation on how you can incorporate advanced cooling in your data center, we welcome you to reach out to our representatives at marketing@gigacomputing.com.

Learn More:
《Tech Guide: To Harness Generative AI, You Must Learn About "Training" & "Inference"》
《Case Study: A Sophisticated Air Cooling Design for a World-renowned Automaker》