Dell has positioned its latest PowerEdge C-series platform to meet the needs of both traditional HPC and the hyperscale market. The recently hatched PowerEdge C6320 is outfitted with the latest generation Intel Xeon E5-2600 v3 processors, providing up to 18 cores per socket (144 cores per 2U chassis), up to 512GB of DDR4 memory and up to 72TB of flexible local storage.

HPCwire spoke with Brian Payne, executive director of Dell Server Solutions, to explore how the new PowerEdge C6320 fits in with Dell’s broader portfolio and approach to the widening HPC space.

With two Intel Xeon E5- 2699 processors, the new server offers a 2x performance improvement on the Linpack benchmark, delivering 999 gigaflops compared with 498 gigaflops from the previous generation PowerEdge C6220 (outfitted with Xeon E5-2697 CPUs). The C6320 also achieved a 45 percent improvement on the SPECint_rate benchmark and up to 28 percent better power efficiency on the Spec_Power benchmark.

The PowerEdge C6320 employs a “4in2U” design, meaning it has four independent server nodes in a 2U chassis, which offers a density that exceeds that of traditional rack servers, and is twice as dense as a 1U server, according to the company. “It also provides an interesting and unique balance of memory and storage and connectivity options,” Payne noted.

In the HPC sphere, Dells sees the C4130 as addressing pain points like scarce datacenter space, delivering double the density from a traditional rack server, allowing customers to scale more compute nodes per rack. Many of the datacenters in the HPC space have been engineered to take advantage of that density, meaning that they have the requisite power and cooling infrastructure in place, said Payne.

Beyond addressing the density, Dell recognizes that the HPC space is changing to become more heterogeneous, and there is burgeoning demand for acceleration technology coming from an ever-widening user group that includes technical computing, scientific research, financial services, oil and gas exploration, and medical imaging.

Customers with problems from these and other domains that lend themselves to being solved more efficiently by GPGPUs and Xeon Phi have the option to pair the PowerEdge C6320 with the accelerator-optimized PowerEdge C4130. Introduced back in Q4 of 2014, the PowerEdge C4130 is a 1U, 2-socket server capable of supporting up to four full-powered GPUs or Xeon Phis.

Dell says its PowerEdge C4130 offers 33 percent better GPU/accelerator density than its closest competitors and 400 percent more PCIe GPU/accelerators per processor per rack than a comparable HP system. A single 1U server delivers 7.2 teraflops and has a performance/watt ratio of up to 4.17 gigaflops per watt.

Dell works closely with the major coprocessor suppliers to align roadmaps and ensure that future developments can be deployed in a timely manner. Currently, the C4130 supports NVIDIA’s Tesla K40 and K80 parts; Intel Phi 7120P, 5110P and 3120P SKUs; and AMD’s Firepro line, including the S9150 and S9100 graphics cards.

Advanced seismic data processing is one of the segments benefiting from accelerator technology. Dell has already scored a win in this market by delivering a combination of the 4in2U form factor and the C4130 server to a customer in the undersea oil and gas space. The unnamed business was able to double compute capacity with 50 percent fewer servers, supporting new proprietary analytics, according to Dell.

Dell’s marquis customer in the academic space is the University of California San Diego, which relied on the new PowerEdge C-series for its Comet cluster. The new petascale supercomputer has been described as “supercomputing for the 99 percent” because it will serve the large number of researchers who don’t have the resources to build their own cluster. Deployed by the San Diego Supercomputer Center (SDSC), Comet leverages 27 racks of PowerEdge C6320, totaling 1,944 nodes or 46,656 cores, a five-fold increase in compute capacity compared with SDSC’s previous system.

Payne noted that SDSC was able to get this cluster powered up and starting to run test workloads in under two weeks, months ahead of Dell’s general availability, which begins next month. Payne pointed to the packaging of the platform as a key enabler. “Instead of racking up four discrete rack servers, having those in a single chassis simplifies that process and can help with the speed of deployment,” he said.

“Our goal is to democratize technology and help the [HPC] industry move forward to drive innovations and [discovery],” he stated. “The way we can do that is by driving standardization and by bringing down the marginal cost of compute – to increase their productivity and also engage with them to understand the nuances and challenges that they have and adapt to those. In the case of San Diego Supercomputing Center, they had a timeline that didn’t necessarily line up with our product general release timeline and we found a way to adapt and respond to their timing needs to fulfill the demand for this latest platform.”

Payne added that Dell is opening up market opportunities beyond high-performance computing. The PowerEdge C6320 along with its embedded management software will be used as a host platform for hyper-converged systems such as Dell Engineered Solutions for VMware EVO: RAIL and Dell’s XC Series of Web-scale Converged Appliances.

By targeting the hyper-converged market, Dell was able to design in a new capability in this product class, a management capability called iDRAC8 with Lifecycle Controller. The tool allows customers to rapidly deploy, monitor and update their infrastructure layer. Larger high-performance cluster users may have the means to build their own tools and capabilities. For everyone else, Dell is making this technology available in its PowerEdge C-Series line. Prior to that it had only been available in the mainstream PowerEdge lineup. For those that don’t need this capability, Dell can still deliver the baseline capabilities without the added cost or complexity burden.

“We are seeing more applications of high-performance computing in mainstream industry, outside the domain of traditional national labs, traditional universities,” said Payne, addressing the symbiosis that is occurring at the interplay of HPC, enterprise and big data. “Going into R&D departments, in oil and gas and other segments that are building out big systems, you see some big data problems being treated very similarly to the way high-performance computing problems are solved.

“You have to think about the skill set and the staff in the IT department that is responsible for deploying and administering this infrastructure, and many times that staff is hosting and supporting a diverse set of workloads for the company – from email to database and now high-performance computing as well as some Web technologies. These folks were trained and accustomed to using server OEM tools to manage the infrastructure and they rely on those versus building their own, now we have extended and given them something that they are familiar with that makes it easier for them to take on a high-performance computing project.”

The new server starts at roughly $16,600 and includes the chassis and four C6320 nodes (2x Xeon E5-2603 v3 CPUs, 2x8GB DDR4 memory, 1×2.5-inch 7200rpm 250 GB SATA, iDRAC8 Express, and 3-year warranty). More details, including networking options, are available on this product page.