RSS Feeds
Nehalem Fires up Desktop Workstations
Intel's Xeon 5500 processor ups the performance standards for desktop workstations, delivering new ways to deploy computationally intensive design tools
Beth Stackpole, Contributing Editor -- Design News, August 17, 2009
Among the many projects underway at the Engines and Energy Conversion Lab. at Colorado State University is a joint effort with Envirofit to develop clean cook stoves for developing nations. While the cook stove appears simple, the actual design is fairly complex. The lab's engineering team spends days on end creating 3-D CAD models, as well as performing cutting-edge Computational Fluid Dynamics (CFD) simulations to get a read on everything from the stove's performance to unpredictable combustion patterns.
Because of the massive processing demands, the CFD simulations and design work has traditionally been done on the lab's server farm, chewing up to a full week of round-the-clock processing and leaving the team with huge gaps of downtime before seeing results. Last spring, the lab brought in Hewlett-Packard's Z600 workstations, based on Intel's Xeon 5500 multi-core processor series and everything changed, according to Morgan DeFoort, co-director of the EECL. The computationally intensive rendering that used to take a full week on the server farm was cut back to one day, DeFoort says. As a result of that significant time savings, the lab is able to improve the combustion model, while simultaneously test driving a bigger portfolio of potential designs.
“There's not a lot of work you can do while the computer sits and chugs on a model — the Z600 has really reduced our downtime while waiting for results,” DeFoort says, adding the lab is seeing a 30 percent increase in processing time compared to what was possible with the server farm. “With all the computer power we now have built into these workstations, it's much simpler for us to have our own workstation on the desktop that we maintain. We don't have to worry about firewalls or permissions and we have enough horsepower of our own to chug through these models.”
Colorado University's EECL is among the many engineering organizations riding a dramatic new performance wave brought on by the Intel Xeon 5500 chip platform, which was introduced last spring. A sea of companies like HP, Dell, Fujitsu Siemens Computers, Lenovo and Core Microsystems are rolling out new desktop workstations based on that platform, while many design tool vendors are modifying their programs to exploit the new multi-core and multi-threaded architectures. As a result, engineering teams are starting to see signs of dramatic performance increases that will push the boundaries of modeling and visualization tools going forward. Thanks to what Intel is calling “new intelligent” technologies in the areas of performance, memory management and multi-tasking, dual-core workstations based on this new processor platform can deliver up to 2.8x performance increases over previous generation Intel Xeon processors and over what many engineers are accustomed to getting on their desktop, according to Intel officials. The platform supports up to eight computational cores, up to 192 Gbyte of memory and more than 100 GFlops of compute performance.
For CAD and CAE users, there's more to this new breed of desktop workstations than just the need for speed. The multi-core architecture and other intelligent technologies help create a digital workbench that Intel officials say will enable new engineering workflows. This will allow users to design models and verify those designs with simulation concurrently, early on in the development process and on the same compute platform. Proponents say the digital workbench allows engineers to design and test a greater number of ideas more efficiently, without the bottlenecks and time delays associated with design and simulation.
“This is the first time in a long time we've had this big a jump in performance and it's a jump for several reasons,” says Tom Salomone, marketing manager for the engineering workstation segment at HP, citing the new Intel Xeon chip, intelligent processing technologies and advances in GPUs such as the NVIDIA Tesla, which off-loads graphics processing from the core processor. Typically, engineers do an analysis right after they do a design, but with quad-core technology, they're able to do design and analysis in parallel, Salomone says. “That's a big deal because before they put anything into final analysis, they're pretty confident their designs are going to pass,” he explains.
Parallel Play
There are several “intelligent technologies” that help the Xeon 5500 usher in these changes. On the memory management front, there is the QuickPath Technology, a scalable, shared memory architecture that creates direct, point-to-point connections between each processor and dedicated memory. Taking a page from AMD with its Athlon and Opteron processors, Intel has moved away from its Front Side Bus architecture to QuickPath, aiming to reduce latency and speed performance — both critical for graphics-intensive 3-D applications.
“This is where analysis applications that are floating-point-intensive benefit because those applications are designed to scale over multiple cores or processors,” says Don Maynard, product marketing manager for dual socket workstations for Dell, which is now offering the Dell Precision line of workstations with the Xeon 5500 platform.
Hyper-threading technology is another springboard for performance gains, but only for applications that have been rewritten to take advantage of executing on multiple cores. In addition to the four cores delivered in the Xeon 5500 chip platform, hyper-threading technology simulates additional cores to boost the performance capacity. This is where some experts say the performance gains promised by the new processor platform can be more theoretical than real. Because most current 3-D programs don't yet take advantage of hyper-threading technology, they can't benefit from the possible performance gains.
“In the past, getting performance gains was in the hands of Intel, with a little help from compilers to increase performance with the same binaries or maybe with a recompile and you could squeeze more performance from a single thread of execution,” says Alex Herrera, senior analyst with Jon Peddie Research. “Software companies didn't have to do anything.” Not so with this new architecture, which requires the design tool vendors to make wholesale changes to their programs. “Fundamentally multi-core is the way to go — the problem is it's not in the hands of the processor vendor to get the additional performance,” he says.
There is a bright spot in the Intel architecture for single-threaded design tools that don't yet tap into hyper-threading capabilities, Herrera says. The Xeon 5500's Turbo Mode automatically adjusts the speed of the cores dynamically, switching off those that aren't in play and allocating that unused power to the cores that are in use. “For those applications that don't deal in a multi-core environment well, you don't necessarily need the 4x improvements,” he says.
Gallery »
Nehalem Platform Fuels Desktop Workstation Power
