Not content to cede the high-performance computing market to IBM, Hewlett-Packard has addressed a key design flaw...
of its Scalable System server platform with the new SL6500 enclosure, plus several ProLiant server models to slide into it.
High-performance computing (HPC) constitutes a niche, but prestigious, market segment that is important for server vendors. HPC shops are often the first to try out leading-edge technologies.
In the prior iteration of SL6000, if a server went down, administrators could service it only by also taking down the adjacent server. But there is no such problem with the half-width servers in the new SL6500 chassis; they can be individually serviced.
“Customers told us they liked the half-width blades because they were extremely power efficient, but they didn’t like that you had to take down both servers to service the other,” said Ed Turkel, Hewlett-Packard Co.’s manager for worldwide HPC marketing.
The SL6500 is a 4U chassis that provides power and cooling to front-loading full and half-width ProLiant SL blades. Compared with HP BladeSystem chassis, “the [SL6500] chassis is very simple; it’s just some bent sheet metal with power and cooling,” Turkel said. The new chassis swells to 4U, up from 2U, and can house a mix of 1U and 2U half-width or full-width servers.
New ProLiant servers
HP also unveiled new ProLiant servers for the SL6500 chassis. Like the SL6000’s SL170z,the new SL170s G6 is a 1U, front-loading, half-width system with dual-socket Intel Xeon 5600 processorss, 16 dual-inline memory modules slots, and support for HP Lights Out 100i, for remote-server management based on the Intelligent Platform Management Interface (IPMI) 2.0 standard.
Also new from HP is the 2U SL390s G7, another two-socket system that features three Nvidia Tesla M2050 or M2070 general-purpose graphical processing units (GPGPUs); on-board Mellanox Connnect-X 2 sillicon with support for 40 Gb InfiniBand and 10 Gigabit Ethernet for high bandwidth, and HP’s proprietary ILO 3 for remote management.
It’s a leadership thing.
Cal Braunstein, CEO and executive director, Robert Frances Group
The GPGPUs are an increasingly popular addition to scale-out servers, because they can be programmed to accelerate parts of an HPC workload, Turkel said.
HP said that the SL6500 is the foundation of Tokyo Institute of Technology’s new 2.4 petaflop Tsumabe supercomputer, a 1,440 node cluster based on SL390s G7 servers. In addition, HP has ramped up production of its POD data center in a shipping container with the goal of delivering fully configured PODs within six weeks of a customer placing an order.
Take that, IBM iDataPlex
HP’s Scalable System competes with the IBM iDataPlex, which launched in 2008 targeting both scale-out Internet-facing workloads and HPC.
While not large, the HPC market is critical for large server vendors, said Cal Braunstein, the CEO and executive director for research at Robert Frances Group. It’s important not just because customers buy a lot of units (although they do); but it is often the first showcase of innovative technology.
HPC is “where a lot of leading-edge stuff occurs, and companies like people to know that they’re participating in this space,” Braunstein said. “It’s a leadership thing.”
HP also has a strong incentive to protect is server turf, he said. “The last thing they want is for their customers to be buying from IBM; you just want to keep them off the floor.”