Longtime server rivals Hewlett Packard Enterprise and Dell will show off high-powered systems at this week's Supercomputing 2022 conference, hoping to attract enterprise users implementing modern workloads driven by AI.
Each company will take a different tack to reach its respective target audiences, however. HPE will take a top-down approach, debuting lower-priced versions of its supercomputers with the capacity to carry data-intensive workloads. Dell will take a bottom-up approach, unveiling an HPC system containing eight Nvidia GPUs, giving it what company officials believe is supercomputer-class performance.
The HPE Cray EX2500 supercomputer, which has the same architecture as the higher-end HPE Cray EX4000 supercomputer, features a form factor 24% smaller than the EX4000. The size lets it fit more comfortably in an enterprise data center. The system is direct-liquid cooled, which not only leaves a smaller footprint but also a lower carbon footprint.
The smaller footprint should appeal to scientific researchers and commercial IT shops alike. It gives them an opportunity to place it in their local work environments.
"Everyone wants smaller so they can move [supercomputers, HPC systems] out of massive data centers and place them in remote locations," said Jack Gold, president and principal analyst at J. Gold Associates. "But given the power [of the HPE and Dell] systems, they wouldn't be for edge computing -- more for distributed computing. There are many edge locations where they don't make sense."
The new system supports both the 4th Gen AMD EPYC and fourth-generation Intel Xeon Scalable processors.
Gerald Kleyn, general manager of HPE's HPC business, said, "This is taking technologies in our [higher-end] supercomputers meant for scientific applications in new packaging down to where it's more accessible for more companies. Many HPC buys are based on what IT has available for budgets. And this [system] brings water-cooled technologies down to lower price points."
Peter RuttenResearch vice president, IDC
Accompanying the EX2500, HPE also rolled out the HPE Cray XD2000 and XD6500 supercomputers, two systems that integrate the HPE and Cray portfolios also aimed at cost-conscious IT professionals. Like the EX2500, they are targeting cost-conscious users implementing more complex workloads involving AI, modeling and simulation.
Both XD machines support the latest AMD and Intel chips and, for the first time, the systems contain Nvidia's H100 Hopper GPUs.
Coming from the other direction, Dell rolled out its first high-performance system that uses either eight Nvidia H100 Tensor Core GPUs or eight Nvidia A100 GPUs. Like the HPE system, the new air-cooled PowerEdge XE9680 works with the latest Intel Xeon Scalable processor and is meant for users working on AI-driven digital transformation projects.
"Users are looking to upgrade IT environments with more advanced computing capabilities," said Rajesh Pohani, vice president in charge of managing Dell's PowerEdge portfolio. "These systems give enterprise IT [users] access to the same level of technology only large research institutions had, making AI and HPC projects easier to tackle."
In addition to the XE9680, Dell also debuted the XE9640, a direct-liquid-cooled system that contains both Intel's Xeon chips and Intel's Data Center GPUs. Company officials said it can reduce energy costs by up to 3.1 times.
The third system, the PowerEdge XE8640, is an air-cooled system with four Nvidia H100 Tensor Core GPUs with Nvidia's NVLink. It is designed for developers creating machine learning models that accelerate analysis.
What's ahead for supercomputers and HPC
With scientific researchers and traditional commercial IT shops each wanting the same higher-performing hardware and AI-driven workloads for their respective workloads, the line between supercomputers and HPC systems is becoming blurred. There is no better example than the simultaneous arrival of the HPE and Dell systems.
"The difficulty now is how do you define what is a supercomputer is; there's not a good definition for it," Gold said. "Not only is there HPE and Dell out there now, but Nvidia says they are doing [supercomputers] and Intel has announced Aurora, which has exascale-class performance."
Peter Rutten, research vice president for IDC's Worldwide Infrastructure practice, said he believes that, given the inevitable scaling up and down of technologies, supercomputing and HPC vendors will collide -- although maybe not for another couple of years. He added that while HPE and Dell are competing in this middle space, each company runs the risk of losing focus on its users.
"Dell will continue to spend much of its time on less skilled users, whereas HPE will remain very much focused on supercomputer users," Rutten said. "They will start running into each other at common clients' offices, but HPE will have to come a little further up and Dell a little further down. It will be a sweet spot for others as well."
The first indications that HPC systems, AI and supercomputers were converging came to light via academic papers four years ago at the International Supercomputing Conference in Frankfurt.
"Now we see AI sitting in the loop of a simulation, adding additional processing using an AI model," Rutten said. "Businesses are figuring out they can run AI on HPC or supercomputing infrastructure. So why spend extra dollars buying infrastructure for each?"
The HPE Cray EX2500 supercomputers with the fourth-generation Xeon Scalable processors will be broadly available sometime in the first quarter of 2023. The version with AMD's EPYC chips will be broadly available in the second quarter.
The HPE Cray EX2500 supercomputers, featuring 4th Gen AMD EPYC processors, are available to order now for select customers and will be broadly available in the second quarter of 2023. The HPE Cray XD2000, with the AMD chips with air cooling, is available to order, while the version that is direct-liquid cooled can be ordered starting Dec. 5.
The Dell PowerEdge XE9680, XE8640 and XE9640 have planned global availability in the first half of 2023.
As Editor At Large with TechTarget Editorial's News Group, Ed Scannell is responsible for writing and reporting breaking news, news analysis and features focused on technology issues and trends affecting corporate IT professionals.