Pumping New Life into HPC Clusters, the Case for Liquid Cooling

By Scott Tease

July 10, 2018

High Performance Computing (HPC) faces some daunting challenges in the coming years as traditional, industry-standard systems push the boundaries of data center space, heat and power limitations. As the industry pushes to achieve exascale computing, overcoming performance barriers such as heat and power use will be the first steps in ushering in the next generation of HPC innovations.

A Bit of History

In the last few years, liquid has evolved from novelty to mainstream in HPC. There are multiple reasons why: support for higher bin CPUs for greater computational muscle, extreme density, reduced noise, and, not least of all, lower electric bills. As HPC progresses toward exascale, these factors will continue to impact cluster decision making.

Lenovo’s Scott Tease. Photo by HPCwire.

So, how did we get here? In 2012, Leibniz-Rechenzentrum (LRZ) in Munich, Germany, a supercomputing center supporting a diverse group of researchers from around the world, gave the HPC vendor community a unique challenge: LRZ wanted to dramatically cut the electricity it consumed without sacrificing compute power. The IBM System x team delivered a server that featured Warm Water Direct Water Cooling, piping unchilled water directly to the CPU, memory and other high power consuming components. Thus, was born the era of warm water cooled supercomputers.

Chillers had been a staple of water cooling going back to the old mainframe days. Instead, at LRZ a controlled loop of unchilled water, up to 45°C, was used. In addition to the energy efficiency and data center-level cost savings, several additional benefits emerged. Since the CPUs were kept much cooler by the ultra-efficient direct water cooling, there was less energy loss within the processor, saving as much as five percent more than a comparable air-cooled processor. If desired, the Intel CPUs could run in “turbo mode” constantly, boosting performance up to an additional 10-15 percent. Because the systems had no fans – except small ones on the power supplies – operations were nearly silent. And, the hot water produced from the data center was piped into the building as a heat source. With additional software support savings from SUSE, total savings at LRZ was nearly 40 percent.

And Today…

Several years have passed, and most, if not all, of the major vendors of x86 systems have jumped into water cooling in some manner. These offerings run the gamut from water-cooled rear-door heat exchangers, which act like a car’s radiator and absorb the heat expelled by air-cooled systems to systems literally submerged in a tank full of special dielectrically compliant coolant – something akin to a massive chicken fryer with servers acting as the heating elements. Direct water-cooled systems have evolved too. Advances in thermals and materials now allow intake water up to 50°C. This makes water-cooling a viable option almost anywhere in the world without using chillers. Also, the number of components cooled by water has expanded. In addition to the CPU and memory, the IO and voltage regulation devices are now water-cooled, driving the percentage of heat transferred from the system to water to more than 90 percent.

Unfortunately, not everything in the data center can be water-cooled, so LRZ and Lenovo in partnership with Intel, are in the process of expanding alternative cooling by converting the hot water “waste” into cold water that can be reused to cool the rest of the data center. This process utilizes “adsorption chillers”, which take the hot water from 100 racks with direct-to-node liquid coolin compute nodes and passes it over sheets of a special silica gel that evaporate the water, cooling it. From there, the evaporated water is condensed back into a liquid, which is then either piped back into the compute racks, or into a rear-door heat exchanger for racks of storage and networking gear, which aren’t water-cooled. This process is able to deliver more cold water than the data center can actually consume. This approach to data center design is made possible because the water delivered to the chillers is hot enough to make the process run efficiently. The tight connection and interdependence between the server gear and the data center infrastructure has strong potential.

It’s Not the Humidity, It’s the Heat

The driving force behind processor innovation for the last 50 years has been Moore’s Law, which states that the number of transistors in an integrated circuit will double approximately every two years. Moore’s company, Intel, condensed to double CPU performance every 18 months while costs come down 50 percent. After half a century however, delivering on that prediction has become increasingly more difficult. To stay on the Moore’s law curve, Intel has to add more processing cores to the CPU, which draws more power, and in turn, produces more heat. Look at how the power draw in Intel processors has grown over the last dozen years:

 

Release date Code Processor Core/chip TDP(W) Spec FP Spec FP/W
06/26/06 Woodcrest Intel Xeon 5160 2 80 17.7 0.22
11/12/07 Harpertown Intel Xeon x5460 4 120 25.4 0.21
03/30/09 Nehalem Intel Xeon x5570 4 95 43.8 0.46
03/16/10 Westmere-EP Intel Xeon x5690 6 130 63.7 0.49
05/01/12 Sandy Bridge Intel Xeon E5-2690 8 135 94.8 0.70
01/09/14 Ivy Bridge Intel Xeon E5-2697v2 12 130 104 0.80
09/09/14 Haswell Intel Xeon E5-2699v3 18 145 116 0.80
03/09/15 Broadwell Intel Xeon E5-2699v4 22 145 128 0.88
07/11/17 Skylake Intel Xeon Platinum 8180 28 205 155 0.76

 

To deal with that heat, processors in an air-cooled environment will need larger (taller) heat sinks, which require systems with taller or bigger chassis. ASHRAE has estimated the increased heat load in a standard rack:

Could those 2U4N server “sleds” that have been the staple of HPC clusters over the last several years end up being an endangered species? Probably not, but they face some challenges in order to survive. Simply put, customers will face a difficult trade-off: system density (the number of servers their IT people can cram into a rack) vs. CPU capability (fewer cores). Customers wanting to run higher core CPUs will have to give up space in the rack, meaning more racks in the data center, meaning higher OPEX in real estate, electric and air conditioning costs.

If the customer does not want to (or cannot) go to full direct to node water cooling, but needs density, and computational power, they are in a jam.  This is where technologies like our Thermal Transfer Module (TTM) will come into play. The TTM is an advanced CPU heat sink for air-cooled dense systems, which utilizes liquid to transfer heat away from the processor to a remote area of the system where air cooling is more effective. This would allow them to maintain their profile without compromising on performance.

Going Green

A second factor at work depends on your geography. “Green” data center initiatives have been in place in Europe for a decade. They are what spurred LRZ and others like them to seek alternatives to air cooling. There is even a “Green 500” list of the most energy efficient data centers on the TOP500.org site. As more of these installations are completed, and promoted, other customers may see 50 percent savings on electricity and 15 percent performance improvement as substantial enough justification to take the plunge to water cooling.

Roughly, 55 percent of the world’s electricity is produced from burning fossil fuels, including coal. Data centers consume almost 3 percent of the world’s electricity and can no longer “fly under the radar” when it comes to energy consumption. Other governments may see these results and put regulations in place to reduce data center power consumption.

Saving Green

“Going green” in the data center is not simply an altruistic endeavor. It is, in many places, a matter of necessity. Electricity prices in some parts of the world can exceed $0.20 per kilowatt-hour, in areas it can be double that amount. This accelerates the decision on alternative cooling to today. The largest hurdle is the up-front costs for plumbing infrastructure and the small premium (in most cases less than 10 percent) for water cooled systems over comparable air-cooled.

Finance departments always ask, “How long until a system like this will pay for itself?” and in some cases, it may be one-year. Of course, TCO and ROI are dependent upon the solution itself and the installation costs, but most OEMs have TCO calculators to assist customers in determining payback on a direct water-cooled system.

Summary

Liquid-cooled HPC is now an established alternative to traditional air-cooled systems. In many cases, it is a question of “when” not “if” to make the move.  As core counts, heat sinks and power consumption all continue to grow, circumstances will dictate the shift. Going to liquid cooling technologies can save space, power, money and still deliver the computational muscle needed to run the most demanding HPC workloads.

Scott Tease is executive director of high performance computing and artificial intelligence within Lenovo’s data center group.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire