|The Leading Source for Global News and Information Covering the Ecosystem of High Productivity Computing / November 13, 2007|
If you were looking for a handle to attach to HP, you could safely think of them as the quiet company of HPC. They've held the revenue pole position for the past four and a half years by IDC numbers. On the June 2007 list they claimed over 40 percent of all systems on the TOP500, but until yesterday they hadn't had a machine in the top 10 since June 2004. To put HP's TOP500 position in perspective, the number two system holder on the June list was IBM with just over 38 percent; together, IBM and HP built nearly 80 percent of all systems in the TOP500. No one else is even in double digits.
In the lastest edition of the TOP500 announced yesterday, HP captured two top spots: the number four position with a 117.9 teraflop deployment at the Computational Research Laboratories in India, and the number five spot with a 102.8 teraflop machine for a Swedish govermment agency. Both were Cluster Platform 3000BL blade systems.
But HP doesn't seem particularly focused on securing top positions on the list. So what is this company's strategy? According to Ed Turkel, "HP isn't avoiding the high end, we're just focused on building systems that meet customer needs, in a way that's consistent with our overall market strategy."
HP is concerned about leadership at the high end of HPC -- the top of the TOP500 list -- but not just for the sake of being on top. "Creating systems at the extreme end of the performance spectrum provides one stimulus for innovation in software and system design, and allows us to create new technology that we can push down into smaller systems," says Turkel. "Of course, there are many other innovation drivers as well, that we find as we create systems that meet different needs. For example, the requirement for high system reliability in an enterprise setting."
In case you're wondering, I asked, and there are no proprietary systems in HP's foreseeable future for high end customers. IBM is pursuing a strategy similar to HP's in the mid-market, but has two special-purpose systems for the high end (Blue Gene and the Cell-accelerated systems still in development). This strategy isn't for HP. According to Turkel, HP's focus is on innovating on top of standard components and software, and taking advantage of industry standard economies.
HP's big push at SC07 is to launch products and services to capture that market that many HPC vendors are salivating over right now: the small and mid-sized business (SMB) market.
The Cluster Platform Workgroup System: taking aim at SMBs
HP sees its biggest growth in the bottom half of the HPC market, with companies and institutions that are not currently HPC customers. HP's Cluster Platform Workgroup System -- launched this week -- is aimed squarely at those customers.
The Cluster Platform Workgroup System can be configured with up to 8 nodes in HP BladeSystem c3000 enclosures, each with 2 quad core processors, for a total of 64 cores per system. Systems will come stocked with AMD or Intel processors, and loaded with Windows CCS 2003 or Linux. In the Linux case, the system comes loaded with the XC stack that includes SLURM, LSF integration, HP-MPI, and just about everything you need to get a system up and running. The CCS version includes the 64 bit Windows Server, the CCS layer, resource management, and everything else you'll need to get a Windows cluster going. You can configure the system with either InfiniBand or GigE, and they don't need special cooling or power: starter configurations from a single 110V wall socket.
The system is designed to be a supercomputer in a box, configured directly by users taking advantage of HP's mass system sales expertise with tools like their web configurators, or sold directly to customers through software channel partners.
That last sales avenue is interesting. HP recognizes that many of the customers they're after in the SMB market don't know about HPC, or if they do, they don't think it applies to them. But many of these customers are using sophisticated software analysis tools on workstations (like FLUENT or Accelrys Materials Studio). HP is educating scientific and engineering ISVs about the value of the Cluster Platform Workgroup System for these customers, in the hope that they'll get a leg up on this valuable new market. In fact, software resellers will be able to offer configurations with their software pre-installed, and tuned to support their applications.
HP plans to start shipping the Cluster Platform Workgroup System shortly after SC. They weren't specific on price, but did mention that they are targeting a starting configuration priced well under $50,000 US.
The Multi-Core Optimization Program
HP started talking about the Multicore Optimization Program (MCOP) last year, when it was just a framework for things to come. The program is envisioned as a way to bring together technology from HP and its partners to provide open, non-proprietary solutions that help users deploy applications more effectively for multicore processors. The program officially launched in June, and at the beginning of November HP announced several new partners: Allinea, Interactive Supercomputing, The Portland Group, RapidMind, the Stanford Pervasive Parallelism Lab, and Visual Numerics.
According to HP, the MCOP is a "set of products, reference material, HP best-practice research, white papers, and links" that will help users port, tune, and optimize applications for effective use of multicore processors. You can find more information on the Web at www.hp.com/go/multi-coretoolkit.
What's HP's role in all of this? They're making sure that the program partners get good requirements from HP's customers, and they're actually helping ISVs with software ports to HP platforms. HP is also using those porting experiences to drive the creation of best practice materials that are posted on the web site. According to Ed Turkel the company is taking things a step further by funding new technology initiatives through MCOP and internally to the company, but he declined to get more specific about these efforts.
The HPC Accelerator Program
HP is also talking about its HPC Accelerator Program at SC this year. The new program, also announced at the start of November, provides a vehicle for 3rd parties to qualify their accelerators with HP ProLiant and BladeSystem servers. Current partners include ClearSpeed and Celoxica on the hardware front, with NVIDIA, AMD, RapidMind, and Mitrionics contributing innovation on the programming front.
Oh, and new hardware, too
HP has also launched new HPC hardware. This week they've announced new products built around quadcore offerings from both Intel and AMD.
The DL160G5 updates the ProLiant GL140G3 with a 1U server targeted at the scale out customer. The system is built around the Stoakley platform -- a combination of the Harpertown 45nm quad core Xeon 5400 processor and the Seaburg chipset. HP is showing these on the show floor this week, and according to Turkel they anticipate shipping them in Q1 2008.
HP is also doing technology previews of an AMD Barcelona-based system, but they aren't projecting a ship date on these yet.
John West is a freelance technology writer, a frequent contributor to HPCwire, and the publisher of insideHPC.com. He's been involved in supercomputing for the past 15 years.