Hewlett-Packard has entered the market for supercomputers with a new Apollo family of systems, including a high-end machine that has a novel water-based system to keep it cool.
Lots of supercomputers have been assembled from HP hardware, but this is the first time the company has offered a listed product specifically for that market. It’s hoping to drive the use of supercomputers beyond governments and academia and more deeply into the enterprise.
It introduced two Apollo systems Monday: the water-cooled Apollo 8000 and an air-cooled system called the Apollo 6000. Both are available for order now, said Jim Ganthier, who heads worldwide marketing for HP servers.
One of the biggest costs of owning a supercomputer is the energy required to keep it cool. Liquid is far more efficient than air at carrying away heat, Ganthier said, but fears about damaging components means liquid cooling has been slow to catch on.
A lot of work lately has focused on immersing servers in special fluids that don’t conduct electricity, like 3M’s Novec fluid. But HP is bucking the trend with a system that circulates plain old water inside the system—in a way that HP says is leak-proof and safe.
IBM also sells water-cooled supercomputers, including its Blue Gene and iDataPlex servers, which can be either water-cooled or air-cooled depending on the application.
The Apollo 8000 has two cooling systems that work in conjunction with each other. One is a water tower—or what HP calls a “water wall”—that runs down the center of the server rack from top to bottom. Water circulates in the tower and draws heat away from server blades attached to the other side of the wall.
Water also circulates in the blades themselves, through sealed copper pipes connected to the underside of each CPU via a heat sink. The copper pipes connect to the water wall, which carries the heat away and out of the rack. A vacuum inside the copper pipe prevents water from escaping in the unlikely event of a leak, Ganthier said.
HP says there are several advantages to its cooling system (you can see a video of it here) compared with systems that submerge servers in special fluids. One is that operators can remove HP’s server blades without shutting the system down. That’s a process that’s difficult or impossible in a submerged system. The 8000 is also a lot more space efficient, according to Ganthier, because it doesn’t need the vats of liquid for submersion.
The National Renewable Energy Lab became one of the first customers for the Apollo 8000, which it uses for climate modeling, carbon-footprint analysis and other workloads. Its system, known as Peregrine, has a top capacity today of 582 teraflops, or 582 trillion operations per second, but it can be expanded to 1.19 petaflops, or thousands of teraflops, HP says.
Using a water-cooled system means the Apollo 8000 can accommodate a lot of power-hungry processors in a small space. Each rack can accommodate up to 144 servers, each with dual-socket Xeon E5 processors and eight InfiniBand switches. Customers can start with a two-rack configuration, which includes a half-rack cooling unit that handles jobs like water intake and maintaining the water pressure and vacuum.
HP also launched the Apollo 6000, which it’s aiming at enterprises for tasks like design automation and financial service risk analysis. It can fit 160 servers in a rack, each with two Intel Xeon E3-1200 processors and two 10GbE FlexibleLOM cards. Future systems will target higher I/O bandwidth applications, HP said.
HP said it couldn’t give prices for the systems because the configuration options are too varied.