Nvidia has introduced its new Blackwell chip, designed to address the increasing energy demands of artificial intelligence (AI) applications. Dion Harris, Nvidia’s head of data center product marketing, discusses the chip’s energy efficiency and performance capabilities. Harris explains that the Blackwell processor, coupled with the Grace CPU, offers twice the energy efficiency of standard x86 CPUs.
Nvidia’s proprietary communication protocol, NVLink, allows GPUs to connect at high speeds, eliminating bottlenecks and driving efficiency. The entire system is engineered to deliver 25 times more energy efficiency than Nvidia’s previous generation. In the short term, Blackwell chips provide a significant productivity boost for customers who may not have the resources to quickly build efficient data centers.
For the long term, Nvidia is collaborating with partners and energy providers to ensure the construction of efficient data centers and resilient grid infrastructure. Harris emphasizes the critical role AI will play in enabling the next wave of infrastructure, not only in grids but also in manufacturing, aeronautics, and other heavy industries. Vertiv has introduced the Vertiv MegaMod CoolChip, a prefabricated modular data center equipped with liquid cooling technology for efficient AI computing.
The MegaMod CoolChip can accelerate the deployment of critical AI digital infrastructure by up to 50 percent. The modular data center integrates Vertiv CoolChip technology to support direct-to-chip liquid cooling, high-efficiency power protection, and distribution.
Nvidia’s Blackwell boosts AI efficiency
It can be used as a modular retrofit for existing facilities or as a standalone data center, supporting hundreds of kilowatts per row. Key features of the MegaMod CoolChip include compute capabilities, liquid cooling with high-efficiency power protection, modular assembly, and single-source accountability. The system employs direct-to-chip cooling for microprocessors and GPUs, ensuring efficient heat removal from components not serviced by the direct-to-chip cold plates.
The power protection and distribution setup includes busway, switchgear, and compatibility with Vertiv’s Trinergy uninterruptible power supply and PowerNexus modular data center. The MegaMod CoolChip arrives in prefabricated units, including the building enclosure and all necessary systems, and is also available in skid-mounted units. ZutaCore and Munters have partnered to deliver a two-phase, waterless liquid cooling solution for AI factories.
The new solution is designed to handle hundreds of megawatts of AI workloads, utilizing Munters’ SyCool (N+1) system to reject heat outside the data center and ZutaCore’s HyperCool to remove heat from AI accelerators. This highly efficient, closed-loop system promises optimized power usage effectiveness (PUE) and water usage effectiveness (WUE), offering a sustainable, cost-efficient, and high-performance data center solution. The partnership addresses the challenges faced by AI factories, which are 5-10 times denser than traditional data centers.
ZutaCore’s HyperCool has demonstrated its ability to cool processors with thermal outputs as high as 2,800 watts at PUE levels as low as 1.05-1.07, significantly boosting compute density. The system increases energy efficiency by 10-20%, operates with smaller pumps, and avoids performance degradation over time. The integrated waterless two-phase solution allows for higher server densification necessary for AI factories, using up to 50% less space than air-assisted liquid-cooled data centers and up to 75% less space than immersion cooling.
Rashan is a seasoned technology journalist and visionary leader serving as the Editor-in-Chief of DevX.com, a leading online publication focused on software development, programming languages, and emerging technologies. With his deep expertise in the tech industry and her passion for empowering developers, Rashan has transformed DevX.com into a vibrant hub of knowledge and innovation. Reach out to Rashan at [email protected]
