1 MegaWatt Of GPU Deployed In Contiguous Cabinets

Customer Profile

Publicly-traded Fortune 500 enterprise based in Silicon Valley

Industry

High-performance computing products and platforms

Challenges

A large global technology provider planned a large-scale HPC computing cluster driven by powerful GPU servers. The company utilizes multiple colocation providers globally but understood that the power and cooling requirements for 1,800 servers housing more than 5,000 GPU cards required a differentiated colocation solution as it could not “spread the load” due to performance and efficiency considerations.

In aggregate, the company required more than 1 megawatt of power across fully-packed, contiguous cabinets—averaging 23 kW of critical load in every cabinet.

Solution 

Colovore has engineered its Santa Clara facility specifically to address the needs of HPC and dense GPU platforms. With modern liquid cooling and power and cooling densities of 35 kW per rack, wall-to-wall, Colovore provided the customer with the smallest, most compact server and cabinet footprint. No customized infrastructure was required.

Benefits 

  • Increased IT operating efficiency driven by fewer racks, all in a contiguous footprint
  • Decreased costs as a result of less networking and cabling infrastructure
  • Optimized application performance due to shorter Infiniband/cabling runs within and between cabinets
  • Easy scalability from day 1 with additional power available in the existing racks to support future growth
  • Improved reliability with a modern liquid cooling system built specifically to handle HPC heat loads

Questions?

To learn more, please contact us at (408) 393-9290 or info@colovore.com.