Advanced technologies, such as artificial intelligence and machine learning have explosive potential, however they also introduce unprecedented obstacles for data centers. These issues lead to under-performing ITE, inefficient power use, increased costs, and more. Conventional air cooling is unable to sufficiently remove the heat. Other cooling options, such as liquid to chip, only cool the processors. Plus, it still relies on air cooling to remove the balance of the heat and plumbing each chip complicates deployments and technology refreshes.
There’s only one cooling technology that delivers optimal cooling and performance results for these operations…single-phase liquid immersion cooling.
Obstacles AI & ML Data Centers Face
  • Advanced CPUs and GPUs Exceeding 400 W TDP
  • Processor Throttling Due to Insufficient Cooling
  • Rack Densities That Don’t Support Peak Performance
  • Significant Increase in Energy Use and Costs

 Immersion Overcomes the Challenges of AI and Machine Learning Operations

High-Density Cooling

Our immersion cooling systems reliably cool up to 368 kW/rack, enabling optimal heat removal from the high-performance servers in high-density proximity required for artificial intelligence and machine learning applications.

Energy Efficiency

Compared with conventional, air-cooled data centers, our immersion cooling systems reduce server power draw by 11% and overall power use for ITE cooling by as much as 90%. This savings frees up energy that would otherwise be dedicated to conventional air-cooling infrastructure, reducing operating costs, and making more power available for compute.

Anywhere Installation

ICEraQ and ICEtank immersion systems are simple, flexible, and resilient, meaning advanced computing operations can be deployed where they’re needed the most. Place compute in locations ranging from an existing air-cooled data center to a greenfield, repurposed warehouse or from an urban parking lot to an arid desert—the boundaries of high-performance compute are far reaching.

Space Optimization

Compared with conventional air cooling, immersion cooling’s ability to support high rack-densities allows for increased compute in a smaller footprint. Our range of cooling system models, flexible configurations, and minimal infrastructure requirements allow for optimal footprints of any space.

CapEx Savings

In greenfield builds, our immersion cooling solutions eliminate the need for investing in costly air-cooling infrastructure such as chillers, air handlers, humidity controls and even raised floors, significantly reducing capital expense. Additionally, compute capacity can be added quickly and easily as demand grows, resulting in capital expenditures can be incurred on an as-needed basis.

Rapid Deployment

The minimal infrastructure requirements and modular design of our immersion cooling systems enable a nearly build-on-demand model, typically delivering expanded compute capacity within just three months. This approach also allows for simplified capacity planning—saving time, money, and eliminating the cost of missed or erroneous projections.

Enterprise AI Deployment: Example

One ICEraQ S10-Duo with Two 42U racks housing:

  • Three Dell PowerEdge R660 servers for Base Command Manager Essentials and Kubernetes Control Plane
  • Thirteen Dell PowerEdge XE9680 8-GPU (NVIDIA H100) servers for Compute Worker Nodes
  • Two Dell PowerSwitch S5232 switches for K8S and Storage Network
  • One Dell PowerSwitch N3248 switch for PXE and OOB Management
  • ~112kW total power draw

Power Up & Cool Down

Bring the Most Efficient Cooling Technology to Your Data Center
Need more information? Call +1 512 692-8003 or Complete the Form Below and a Representative Will Contact You