- Data Center Digest
- Posts
- Nvidia: PUE’s gotta go! and Dell's new AI servers
Nvidia: PUE’s gotta go! and Dell's new AI servers
Plus, Dells new servers, Hawk x CBRE and lots more
Here’s what you should know today:
COOLING, TECH, AND POWER: Nvidia: PUE’s gotta go! and Dell announces new AI servers, Meta’s solar ambitions
CHECK THIS OUT: Great conversation: Data Center Hawk x CBRE
BIG DEALS: Musk shelling out billions for servers, plus Involta borrows a new name and JLL doubles down
- Cooling, tech, and power -
The PUE Debate - Nvidia says “It’s gotta go!”
Nvidia believes it's time to move beyond Power Usage Effectiveness (PUE) as the primary measure of data center efficiency, advocating for new metrics to better reflect the performance of modern, energy-intensive applications.
Run its course?
PUE, introduced by the Green Grid in 2007, compares the total energy used by a data center to the energy used by its IT systems. While the average PUE was 2.2 in 2007, advanced data centers can now achieve PUE scores as low as 1.2.
Despite PUE's historical importance, Nvidia argues that it doesn't account for the useful output of a data center, akin to measuring a car's fuel consumption without considering the distance traveled.
Nvidia suggests that modern metrics should focus on energy usage relative to the work performed, such as kilowatt-hours or joules.
“Over the last 17 years, PUE has driven the most efficient operators closer to an ideal where almost no energy is wasted on processes like power conversion and cooling,” the tech giant said.
“PUE served data centers well during the rise of cloud computing, and it will continue to be useful. But it’s insufficient in today’s generative AI era when workloads and the systems running them have changed dramatically.”
Chris Belady, the engineer who invented PUE, agrees. “PUE improved data center efficiency when things were bad, but two decades later, they’re better, and we need to focus on other metrics more relevant to today’s problems,” Belady said.
Not so fast…
However, some industry experts are cautious. Vladimir Galabov from Omdia agrees that PUE doesn’t address issues like server overprovisioning and underutilization. But he also points out that PUE's simplicity and widespread acceptance make it a resilient standard despite its limitations.
Daniel Bizo of Uptime Institute thinks both sides are right. He notes that while PUE effectively measures power distribution and thermal management efficiency, it doesn't account for IT infrastructure efficiency. According to Daniel, establishing a standard benchmark for IT efficiency across different architectures is complex and often impractical due to varying workload types.
“Ultimately, customers need to understand which compute platform works best for a given set of their applications, a balancing act between performance, energy and cost. Trying to make generic comparisons between vastly different platforms that have very different workload purposes has little value,” he said.
This conversation on the efficacy of the PUE measurement has been going on for several years. If you’re interested in a potential alternative to PUE, read this paper by Lawrence Berkeley National Laboratory.
_____________________________________
Dell unveils new AI servers as stock soars
If you blinked, you may have missed the massive run-up in Dell Technologies’ stock price. Part of that has been in response to the growth in their AI business.
In furtherance of this growth, last week Dell introduced new PowerEdge servers featuring designs from the Open Compute Project and optimized for cold-aisle deployment.
The PowerEdge R670 and R770 CSP Edition servers are tailored for high-density cloud workloads like virtualization and analytics, offering improved performance and a standardized architecture for easy deployment and servicing.
For smaller customers, the PowerEdge T160 and R260 servers double the performance of previous models in a reduced footprint.
The new servers incorporate Smart Cooling technology with options for air, direct liquid, and immersion cooling.
The R670 and R770 are designed for large data centers, featuring compact form factors with front-facing ports for cold aisle configurations. They utilize Intel Xeon 6 processors with efficiency cores, promising up to 2.3 times more performance per rack compared to previous models.
These servers include Dell Open Server Manager for remote management, built on the OpenBMC project, and debut the Data Center – Modular Hardware System (DC-MHS) architecture for modular, scalable, and flexible hardware deployments.
The PowerEdge T160 and R260 are available globally this month, while the R670 and R770 CSP Editions will be available in July.
_____________________________________
More in Cooling, Tech, and Power
1. DigitalBridge CEO: “We’re kind of running out of power.” Marc Ganzi of digital infra developer DigitalBridge predicts that, on our current trajectory, we will be out of power in 18-24 months. With power infrastructure taking over three years to develop (permits alone in Dallas can take over two years), LightEdge suggests we may have a problem…
2. Meta inks new deal to bring 200MW of solar to Indiana. Meta’s energy agreement with Spanish solar operator Solarpack will provide renewable energy to Meta’s coming Indiana data centers. Indiana currently generates only 10% of its power from renewables.
Touring Ericcson’s 5G Plant
Data Center Adjacemt: LightReading was invited to walk through and document Erricson’s new $150 million 5G manufacturing facility. Featuring robotics, lasers, and radios, this is a cool look into a next-gen factory.
- Deals and Developments -
Musk’s xAI to spend $10 billion on AI development
Elon Musk’s AI venture, xAI, is in talks with Oracle to purchase around $10 billion in cloud servers.
Elon’s vision
The AI firm was founded last March after his breakup with OpenAI, aiming to "understand the true nature of the universe.”
As of December they had raised $134.7 million in outside funding, and are reportedly finishing up a $6 billion funding round now.
One of xAI’s first goals is to create an AI capable of advanced mathematical reasoning, something not available in other AI models. So far the company has released a Beta version of a chatbot, called Grok, that will be available to X’s Premium+ subscribers.
The Oracle deal
To compete with Anthropic and OpenAI which have both struck major cloud deals with Amazon and Microsoft, respectively, xAI needs a lot more computing power than they’re currently running.
According to The Information, the firm is already renting 15,000 H100s from Oracle. (An amount Lary Ellison describes as far smaller than Musk had wanted.)
This new deal would be a multi-year agreement that would make xAI one of Oracle’s biggest customers.
For Oracle’s part, they’re using this and other deals to jumpstart their AI and digital infrastructure business. In March CEO Larry Ellison stated the company has more data centers in development, including “the world largest” facility.
Oracle’s stock jumped over 5% on news of the xAI deal.
_____________________________________
More Big Deals
Involta rebrands to Ark Data Centers: Not to be confused with Ark Data Centres… The edge, colocation, and cloud services stalwart says, “The decision to relaunch as Ark was driven by the insatiable growth of infrastructure requirements for space, power, and compute.” Founded in 2007, Involta was acquired by the Carlyle Group in 2021.
JLL is doubling down on DC strategy with latest acquisition: The real estate and investment management firm (whose Global Data Center Outlook has become the industry standard) just acquired SKAE Power Solutions. SKAE provides data center technical and project management services. This is another acquisition in the industry-specific “technical expertise” niche that JLL has dominated in the data center space over the last decade.
Latest in PE - King Street scoops up Colovore: Seen as a pioneer in the liquid cooling industry, Colorvore is a Silicon Valley-based high-density colocation firm specializing in AI and high-performance computing.
For King Street the move is strategic. The firm “aims to leverage its real estate platform to support Colovore’s growth in response to increased demand for liquid-cooled data center solutions.
Thank you for reading.
You can support this newsletter by telling me how we did or sharing it with a friend or colleague.
- Taylor