VISUALIZATION

By Steve Fisher, Editor In Chief -- Last week was a busy one for SGI. It literally flooded the news wires with six significant press releases in two days. The reasons for this flurry of activity were four new products and a new concept in advanced visualization that the company is calling Visual Area Networking. The products include OpenGL Vizserver 2.0 software, the SGI Onyx 3000 visualization system series with new InfinitePerformance graphics subsystem, the Onyx 300 visualization system, and the Fuel visual workstation. The Visual Area Networking Vision OpenGL Vizserver 2.0, as the driving force behind Visual Area Networking, allows remote or mobile users to interact with visualization supercomputers or to collaborate graphically with globally dispersed teams using any client device. So, although there are of course variables in the industries that would use technology such as this, one could say that the basic model is pretty much the same. There’s some guy out there with a massive amount of data, but the experts that need to look at the data aren’t out there at the same spot he is. They need a way to collaborate without having to send all that data over a network or the Internet. You don’t want to send 20 Terabytes of data about an oil field to somebody that’s out on an oil rig somewhere in the North Sea. Even if you did do that, they probably don’t have the technology they need to view the data. If by chance they were sitting out there on the rig with some sort of burly computer system and could view the data while you did, you now have two different copies that need to be synchronized and you’ve now got 40 TBs on your hands. Not great. With OpenGL Vizserver, one person can in effect “drive” an application and the data therein while the other person sees the exact same thing the “driver” does. Much nicer. Especially when one considers that you don’t need that burly computer system to view the data. A laptop or even a PDA would do just fine since the device doesn’t even have to support OpenGL. The server does. The application itself has to run on the Onyx, but the client can use anything because all you’re doing is sending packets back to actually drive the manipulation of the object. You also don’t bog down the network with the transfer or storage systems with the 20 TB duplicate that’s produced. What’s really exciting is the next step in the evolution of this technology. The next step SGI is working on is one where people teaming together on a project are each looking at the same data set, but viewing it from multiple eye points. Estes explained it like this, “Imagine we were working on building a city together, doing urban planning. I might be building a hotel over in this part of the city and you’re building a skyscraper over there. I could look off in the distance and I could see your skyscraper going up and you could do the same with mine. So we’re literally working on the same data but we both have different eye points. We have simultaneous control.” InfinitePerformance Graphics on the Onyx 3000 Series InfinitePerformance is a SGI’s new scalable graphics system and is not replacing the company’s well known InfiniteReality offering. InfinitePerformance reportedly delivers the world's fastest interactive graphics performance with high image quality, flexibility and affordability. It is built on the shared memory SGI NUMAflex architecture of the SGI Onyx 3000 series systems. In performance terms it offers up to 283 million triangles per second of sustained performance and 7.7 billion pixels per second. InfinitePerformance graphics are built using multiple low-cost, high-performance graphics pipes and optional SGI scalable graphics compositor technology. From a technical standpoint what SGI has done is taken the engine from its V-Pro graphics, its top of the line desktop graphics family, and put multiple (currently four) V-Pro graphics engines into a system. Then with the help of something called a scalable graphics compositor on the back end, they are able to channel the multiple graphics into a single display. According to the company the performance is basically linear and it’s given them a significant boost in polygon performance. “The core thing here is, two basic benefits come out of this. The first one is that you can increase the number of polygons that you look at. So if you’re a manufacturer and you’re trying to look at a very complex car or crash analysis elements or whatever you’re doing, then this is going to be a godsend,” Estes said. “It’s very cost effective by the way to get into, and that leads me to the second way people will use this and that is…as four keyboards and four mice off a single shared memory system. So, four users but with one set of CPUs, disks and memory.” The Onyx 300 The Onyx 300 leverages the NUMAflex modular computing architecture and provides denser rack configurations and lower price points for InfiniteReality3 graphics capabilities. Its relationship with the Onyx 3000 is similar to the relationship between the Origin 300 and the Origin 3000. The product is clearly meant to expand the reach of SGI’s visualization products to those folks that traditionally weren’t able to afford them. Basically SGI has lopped about 37% off the price tag to get into an Onyx system. Despite that fact though, the 300 is no shrinking violet. It can compactly scale up to 32 MIPS R14000 500 or 600 MHz processors and two graphics pipes in a single rack configuration. The system also offers Full-scene anti-aliasing; DPLEX multipipe rendering; Multiple high-resolution, large-screen display support; and binary compatibility with Onyx2 and Onyx 3000 series systems. Estes commented, “Nobody ever looks at an Onyx and says, ‘yeah love the price but if it just had a few more features or more power.’ They go ‘that’s the best thing I’ve ever seen. I sure wish it cost less.’ So we’re really trying to drive that cost down by leveraging the fact that we have a ten X lead in terms of the amount of power draw that the MIPS processors take versus everything else. We’re going to do a better job exploiting that and what you’re seeing now is that we’ve got 4 CPUs in a 2 rackmount unit and we can get 32 processors just in that single rack footprint with two graphics pipelines so we’re going to start driving the density a lot better than we have been.” Fuel Visual Workstation Your basic Fuel workstation comes with either a single 500MHz R14000A MIPS processor with 2MB L2 cache or 600 MHz with 4MB L2 cache; a 200 MHz front side bus; V-Pro V10 or V12 graphics with up to 128 MB configurable graphics memory, 104MB texture memory and 48-bit RGBA with 16-bit Z buffer capability; and the fifth-generation 64-bit IRIX operating system. “What we did was, we have taken the basic infrastructure in terms of at the chip level and the front side bus and the structure of how quickly we can get to memory and get the information to the processor, we’ve taken that from the high end system and married that with the V-Pro graphics engine that we use on our Octane class machines and a single MIPS processor at 600 MHz. So it’s our fastest processor, our fastest desktop graphics and an architecture that’s borrowed from the high-end,” Estes stated. After mentioning the fact that Fuel is SGI’s first red computer in eight years and chuckling a little, Estes said that Fuel does not replace either the O2 or the Octane. However, he does believe that due to its superior price/performance it may cannibalize some or even all of SGI’s single-processor Octane business. “You put the faster processor, the faster bus, the same graphics and a lower price point and you do the math. It’ll provide a lot better value and help us stay competitive in the Unix workstation space.” So there you have it. All in all, a pretty interesting week for SGI. Of particular interest I thought was the fact that the Visual Area Networking “vision” or “philosophy” seemed to overshadow the product announcements a bit. Philosophies don’t often win the battle for attention with hardware and software in this industry. Kind of a nice change.