The company announced the PowerEdge C6105 rack-mount server, which can accommodate up to 48 processor cores in a 2U box. The server can create large clusters to run scientific or math applications, and can also scale performance in densely packed cloud-computing environments, company officials said.
[ Keep up on the day's tech news headlines with InfoWorld's Today's Headlines: Wrap Up newsletter. ]
The server will run on Advanced Micro Devices' Opteron 4000-series processors, and includes four two-socket motherboards, with each socket accommodating processors with up to six cores. The server is a complementary offering to Dell's existing PowerEdge C6100 scale-out rack server, which runs on Intel's Xeon processors.
The high-performance and cloud-computing environments have parallel issues, as both increasingly deal with huge amounts of data, said Donnie Bell, senior manager of HPC solutions at Dell. Both require dense computing environments that can take maximum advantage of high-bandwidth fabric and parallel processing capabilities in systems.
"Cloud is really HPC 2.0," Bell said."They're not scaling up, [but] scaling out."
Servers built on modular architectures are being offered for scale-out environments by many companies. Some servers include fewer components to save space and reduce power consumption, with performance scaling through cores and memory. Hewlett-Packard offers the Extreme Scale-Out portfolio, and IBM offers iDataPlex, which are based on x86 server processors and can build out as service providers need more computing capabilities. On the other hand, SeaMicro offers the SM10000 server, which packs in 512 low-power Intel Atom processors on miniature motherboards the size of credit cards.
Dell's customers have said that cloud computing is in its formative stages, but configurations notwithstanding, companies are looking for more efficient ways to develop computing platforms, Bell said. Even high-performance computing environments are slowly going into the cloud as the utility computing model evolves, depending on how organizations want to use them, Bell said.
Beyond hyperscaling, the new server is also about power, cooling and density, Bell said.
The C6105 server is purpose-built -- like the C6100 -- and has specific features not found in Dell's other servers. The C6105 is designed to include as few components as possible to save space and reduce power consumption.
Dell has put four motherboards in each server with shared power and cooling components, Bell said. Dell's traditional rack servers have power and cooling components for each motherboard.
"That's how you get some cost down and efficiency up," Bell said.
The company also packs in more performance by sacrificing certain components like the embedded LifeCycle Controller remote system management component, which is commonly found in Dell's low-end and mid-range offerings.
Such components add to the overhead and are less relevant in the scheme of large cluster deployments, Bell said. The servers are remotely managed using features based on the industry-standard IPMI (Intelligent Platform Management Interface) specification.
Dell also announced the PowerEdge C410x, an external PCI-Express expansion chassis that can connect the C6105 servers to up to 16 devices such as graphics processing units. Companies are looking to harness the processing power of GPUs for high-end engineering and graphics applications, and a C410x chassis with sixteen Nvidia Tesla M2050 GPUs can deliver up to 16.5 teraflops of performance, the company said.
No competitor offers an external chassis to accommodate GPUs, Bell said.
The C6105 server will come with either Windows Server 2008, Windows HPC Server, Red Hat Linux 5.5 or Novell Suse Enterprise Linux operating systems. The server will start shipping globally in the next 60 days. The company did not immediately provide pricing.