HP Discover 2014: Using advanced techniques HP is squeezing more CPU power per rack
LAS VEGAS –Cloud computing may be on every CIOs lips but Hewlett-Packard Co. believes there’s lots of life left in data centres.
To prove it, the company announced a new supercomputing system called Apollo aimed at enterprises, as well as new storage, networking solutions to kick off its annual Discover 2014 infrastructure conference here on Monday.
The top of the line Apollo 8000 uses liquid cooling to more densely pack HP x86 ProLiant blades into its considerable structure –it stands 94 inches high. If the ceiling in your data centre isn’t tall enough, HP has a special crane that can take off the top half of the compute and assemble it separately.
HP officials wouldn’t put a price tag on an 8000 system, saying each one is customized. In fact, they aren’t saying an 8000 is necessarily less expensive up front than comparable supercomputers. However, they insist that there are considerable savings from the energy and space the system saves.
“For enterprises to outcompete they need to outcompute,” Antonio Neir, senior vice-president of HP’s server and networking divisions, told technology reporters as he unveiled the system.
Generally supercomputers – or systems for high performance computing –are bought by governments and academic institutions. HP hopes to sell model 8000s and its smaller brother, the air-cooled 6000, to financial institutions and others that need predictive analysis workloads and simulations. Among its North American competitors are IBM and Cray.
“We’re not known for supercomputing,” said Jim Ganthier, VP, global marketing. “This is our foray and we’re going in strong.”
Because the 8000 can hold up to 144 servers per rack, the system can offer four times the teraflops per rack compared to air-cooled designs, says HP. The warm water used to cool the system can be recycled as a heat source for facilities, or, as one early customer did, to melt outside snow.
An early customer is the U.S. National Renewable Energy Labs (NREL), which expect to save US$800,000 in operating expenses per year over a traditional HPC system. Steve Hammond, director of NREL’s Computational Science Center, said in a statement that because we are capturing and using waste heat, it will save another US$200,000 in heating costs.
NREL bought the 8000 in a competitive bid when it was looking for a new system.
“When people are looking for that kind of supercomputer, it’s how can I get that kind of performance very densely packed so I can scale it out to larger levels,” John Gromala, senior director of product management for hyper scale servers, said in an interview.
“We integrate the networking, we integrate the cooling, the management. We’ll have support for next-generation Intel Xeon E5 processors and next-generation graphic processors as well.”
“I believe that the most aggressive enterprises will, in fact, consider using an Apollo 8000 system,” Patrick Moorhead, president of Moor Insights and Strategy, said in an email. “Water cooling is exotic, but HP’s implementation has greatly reduced the risk as it is in the rack, not near processors, memory, or networking. It’s also a closed, warm water system that doesn’t require expensive cold water. If enterprises want the impressive density and throughput for heavy-duty computing, they will consider it.”
In an interview Gartner vice-president of research Neil MacDonald said the announcement shows that innovation is still alive at HP as it uses industry-standard Intel processors to get into HPC. Just as important, he added, is that HP can now offer server solutions from the low to high end for any organization’s needs.
There is an increasing need for high performance computers in the enterprise, he said, and HP wants to be a part of it.
An Apollo 6000 System can pack up to 160 single-socket servers per rack. When compared to competitive blade solutions, HP says these servers can deliver greater performance and efficiency in less space while using up to 46 percent less in energy, and lowering total operational expenses.
Pricing for a full system with 160 servers works out to US$1,850 per node, HP says.
Apollo systems can be bought direct from HP and through its system integrator partners.