I'll admit it: I have a soft spot for SGI. As an undergraduate, and in graduate school, I had the good fortune to work in the NSF's Engineering Research Center for Computational Field Simulation at Mississippi State University under the leadership of Professor Joe Thompson. A lot of my work at that time was in virtual reality and scientific visualization, and I had something that most other students didn't have: my very own SGI Crimson, a VPL headset, a DataGlove, and a lot of free time.

At that time SGI was the undisputed graphics king of the known universe. It wasn't until I started my first full time job in a supercomputing center that we acquired SGI's first HPC product, the Power Challenge Array. The company eventually left the graphics business altogether (more or less) to focus on HPC, and has been through many changes over the years. In some respects, SGI hopes the future bears at least some resemblance to the past.

This is SGI's silver anniversary, and the theme for the booth this year is "25 years of changing the world." As conference goers talk to SGI this week they'll be hearing about how SGI's solutions enable customers to solve their problems. These stories will unfold against a backdrop of SGI's technology portfolio, and there is a lot of gear to talk about this week.

SGI's compute offerings center around Altix and its various flavors, the Altix XE and Altix ICE. This week SGI is announcing new versions of the Itanium-based 4700 big boy and the mid-range 450 based on Intel's new Montvale chip. SGI will also have a functional ICE system on the floor, and a non-functional one with showing off the water-cooled rear doors.

Attached to the ICE system will be part of SGI's newly announced NAS product line, NEXIS. This week SGI is launching the new NEXIS 9000 product, an InfiniBand-based RDMA storage device specifically designed for the HPC market.

SGI is also showing off changes to its XE lineup this week. The Altix XE240 2U server series and the XE310 1U high density server series are both being revved to support the Stoakley platform -- a combination of the Harpertown 45nm quad core Xeon 5400 processor and the Seaburg chipset.

SGI is also talking about advances in its acceleration technology this week. SGI's Reconfigurable Application Specific Computing (RASC) platform in the RC100 form factor has been available for the Itanium-based Altix line for a while. Now SGI has launched the RC200 in order to bring FPGA acceleration to the Xeon-based Altix XE and ICE systems. SGI is demonstrating the power of its RASC solution on the show floor this week in a variety of running applications.

So where is SGI headed with all this new kit? Dave Parry describes SGI's customer base in three segments: "A huge government segment, a strong but smaller industrial high performance technical computing segment, and the high performance business segment that we see as a key growth area for us." This last segment is the area that so many other vendors are also targeting this year. For SGI this is a small but growing business, and their ccNUMA technology lines offer real advantages for customers with gigantic databases and lightning fast search requirements. SGI is targeting Altix as a key product in this space, a move that makes sense given that the Itanium is well-positioned to compete against the installed based of RISC-based Unix systems.

Systems deployed in this target market are likely to be small -- from 8 to 32 cores -- and volume will be key to making market this a financial success for the company. Unlike Linux Networx, however, SGI sees this as a "push" market, and they intend to engage this market directly with a reorganized sales team that is focusing exclusively on winning this business. The company intends to go after customers directly, and it sees partnerships with commercial systems integrators as a key component of its sales strategy. "Our goal is to use commercial systems integrators as torque converters to transform SGI's portfolio-based product offering into an expertise-based sale using the knowledge and relationships of the integrators," says Parry. Another key aspect of their strategy in this market is the company's Industrial Strength Linux Environment (ISLE) initiative, aimed at creating a software environment that meets the expectations of high performance business computing customers.

Long-time SGI customers have no doubt been intrigued by the hints SGI's new CEO has been dropping about the company's return to the visualization business, and this is sure to be a big topic of conversation in the booth this week.

SGI is developing a new strategy around visualization that it's calling "visual supercomputing," in which visualization nodes are placed into a supercomputing environment. The nodes will use commodity NVIDIA gear for painting polygons, but the rendering isn't the focus for SGI this time around. According to Dave Parry, "the interesting problems in visualization today are moving from being graphics problems to visualization modeling problems." For SGI painting pixels quickly is a solved problem, but finding ways to deal with the enormous volumes of data that users are generating is a problem that SGI sees as worth trying to solve.

My graduate and early professional work benefitted enormously from SGI's work on GL and Performer, so I was naturally curious about whether SGI has a new paradigm-changing library in the works to help solve this visualization modeling problem. No such luck, according to Parry. While the company does still have a lot of intellectual property in visualization, particularly in data fusion and decision support, Parry says that "the industry and open source tools have come a long way, and we believe that most of the software in our visual supercomputing stack will be open or 3rd party."

SGI is demo'ing a visual supercomputing solution on the show floor, and has announced that it will be making visualization nodes available for the Altix ICE and XE products.

-----

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.