From VFX studios to geological imaging facilities, users place increasingly heavy demands on todayâ€™s workstations. As applications for high-end graphics proliferate, vendors scramble to produce faster, more capable, and more power-efficient machines. Along with this trend, graphics processing I/O is being put to the test.
PCI Express (PCIe) technology has proven to be a flexible, scalable, and, above all, capable interconnect for graphics performance. PCIe replaced the AGP8X (Accelerated Graphics Port) in all new systems when it launched in 2004, and AGP is nearly gone now from the retail systems upgrade market.
Even so, bandwidth needs in the world of imaging will only increase, and the capabilities of graphics controllers continue their dramatic improvements. To meet these burgeoning requirements, the PCI Special Interest Group (PCI-SIG) has released a new PCIe specificationâ€”known as PCIe 2.0â€”which provides double the bandwidth of the previous PCIe generation while maintaining backward compatibility with previous PCI and PCIe versions. Although bandwidth doubling is the key advantage of PCIe 2.0, the new specification also sports a number of other improvements to PCIe that, if somewhat esoteric in details, will positively affect workstation users in every field.
Workstations in Flux
At one time, the term â€œworkstationsâ€ referred to a fairly circumscribed subset of computing solutions, produced by highly specialized manufacturers and serving a fairly small market segment of researchers, engineers, and cutting-edge gaming, film, and television production companies.
Today, the situation is different. The continued exponential growth in the processing power of computer chips has blurred the edges of what was once a well-defined marketplace, though, generally, a â€œworkstationâ€ will offer higher reliability (for example, using ECC memory) and faster and more storage than a standard PC client. The current generation of general-purpose computers significantly outperforms yesterdayâ€™s workstations, leading them to gain significant share in many once-rarefied markets.
Simultaneously, however, the power and performance needs of graphics professionals have continued to grow, pushing the performance envelope even further. This creates a situation in which workstations continue to sell briskly. Jon Peddie Research reports that the market for workstations grew by 22 percent in 2006 in terms of unit sales, with 706,000 shipped, representing $6.2 billion in sales.
The result is a market full of choices. Not only can workstations be configured with a wide variety and quantity of single- and multi-core processors, but the increasingly specialized uses to which they are put create a strong market for cutting-edge graphics processors, hardware accelerators, high-bandwidth data storage, and high-speed networking. None of this progress would be possible without efficient ways to connect this cornucopia of specialized components to the main processors.
In recent years, PCIe has become the dominant means of making those connections. Its origins stem from the original PCI specification developed in 1993, which represented tremendous performance and flexibility gains over competing bus architectures of the time. As graphics applications pushed the limits of PCI technology, the AGP was developed and became the standard way of attaching graphics to workstations, desktops, and laptops. However, early in the new millennium, the needs of graphics professionals outstripped the capabilities of parallel interconnects like AGP, leading to todayâ€™s rapid adoption of PCIe.
Even going back to the first generation of PCI before PCIe, the technology was a driver of change in the workstation market, as it was an industry-standard I/O attach point from which many companies could innovate. This differed from the proprietary interconnects of supercomputers and workstations past. With the open specifications, a company could develop high-performance computing components and simply incorporate a leading-edge, universally available interconnect in its designs. The result was greater openness, choice, and competition in the workstation market, driving workstation hardware innovation up and prices down.
While PCIe is serving the graphics and other high-performance computing markets well, it is only a matter of time before it, too, needs to evolve to meet the new requirements of multi-core processors and increasingly powerful graphics controllers. Anticipating that requirement, PCI-SIG has released a new version of the PCIe specification. This breakthrough development for workstations and for computing in general is PCI Express 2.0 (PCIe2).
The major difference between PCIe1 and PCIe2 is an increase in bandwidth. This is good news for workstation users, for whom itâ€™s all about speedâ€”whether for real-time memory access when manipulating complex objects, calculating images from physical data, or rendering complex VFX.
The numbers speak for themselves: PCIe2 provides 5.0 GT/s compared to 2.5 GT/s for PCIe1. Looking at graphics applications in general, the benefit of increased throughput is obvious. More bandwidth simply means less time spent sending data back and forth between the graphics controller and the CPU host chipset. For todayâ€™s critical medical, geological, and scientific imaging needs, in which vast datasets are the norm, PCIe2 has the potential to enable entirely new applications of graphics technology. As PCI-SIG states on its Web site, the inspiration for PCI from its initiation to the present day comes from three areas, the first of which is graphics processing.
Early adopters of PCIe2 need not fear that they will have to buy new hardware. PCIe2 is fully backward-compatible with PCIe1. PCIe2 uses the same form factor, bus widths, connector, pin-outs, and protocols as PCIe1. PCIe2 interconnect is physically different from the original parallel bus PCI, but it is, in fact, software-compatible all the way back to this 1993 technologyâ€”meaning it is supported by all production operating systems. This is a significant benefit in the kaleidoscopic universe of hardware currently available. Hardware vendors and end users can implement PCIe2 in preparation for the future without getting rid of the technology they already have.
While the speed jump is the most important new feature, PCIe2 has some additional functionality that makes it even more reliable and robust than PCIe1, as well as functionality to facilitate improved power management. When software detects that not all the bandwidth of PCIe2 is currently being used, dynamic link speed management allows the bus to fall back to the slower 2.5 GT/s rate, thereby conserving energy. Similarly, software can dynamically narrow the PCIe2 bus from 16 lanes (x16) as physically used for graphics controllers, to eight (x8) or four (x4) electrically, also reducing energy consumption based on the workstationâ€™s current load.
These new features are all the more important because of the increased power consumption now allowed in PCIe2 for the graphics controller itself. Using a new 2x4 pin connector, GPUs will be able to draw 225W or 300W from PCIe2 buses.
Only the most data-intensive graphics applications are pushing the limits of PCIe1 today, but the future is clear. PCIe2 will enable a new generation of imaging technology. The financial and speed benefits of virtual prototyping have made it central to modern manufacturing, particularly in high-tech sectors like automotive and aerospace. PCIe2, along with next-generation multi-core chips, will allow engineers to work with high-definition digital mock-ups more flexibly and rapidly than ever before. Immersive engineering applications that allow three-dimensional visualizationâ€”in effect, virtual-reality digital prototypingâ€”may become a standard way of working.
This Intel S5000XVN workstation board, in production today, will soon be replaced with Intel workstation boards supporting dual x16 PCI Express 2.0 graphics slots.
The petroleum industry has always pushed the limits of data-driven visualizationâ€”and bandwidth is one of those limits. New quad-core workstations with PCIe2 adapters will enable industry workstation users to process larger datasets with more attributes in less time. The discovery and analysis of petroleum reserves will become less costly and less time-consuming. With next-generation, high-power graphics adapters enabled by PCIe2, users will be able to visualize these images with greater precision in new, more meaningful ways.
In the entertainment field, faster data transfer and better computers will lead to more transparent technology. Quad-core workstations and PCIe2 graphics cards will accelerate digital storyboarding. Artists will be able to interact with ultra-high-resolution imagery in full floating-point detailâ€”and in real time. Without bandwidth bottlenecks, their imaginations will be free to roam across the digital landscape and produce results more quickly and with less effort. As they say in the movie business, â€œItâ€™s coming soon to a workstation near you.â€
Itâ€™s not far-fetched to say that enhanced graphics processing could save human lives. Medical imaging is another bandwidth-hungry field. PCIe2 will let the applications of the future stretch their wings. Doctors and nurses already have a number of ways to visualize patientsâ€™ bodies in 3D and 4D, inside and out. These â€œvirtual bodiesâ€ allow more precise diagnosis and more effective procedures. The larger and more detailed the underlying data, the more accurate the images. PCIe2 will enable a world of portable, high-performance medical workstations that allow a physician to image body systems in real time.
It wonâ€™t be long before computers with up to eight computational cores, 80 gigaflops of performance, and PCIe2 graphics will usher in the era of the workstation supercomputer. Users will have the processing power and bandwidth to create their most innovative work, experimenting, discovering, and creating more quickly than ever before. PCIe2 opens the door to the graphics of the future.
Nvidia just announced the Quadro FX 570, one of several new professional graphics boards that are PCIe2 compliant.
Graphics today help diagnose disease, discover resources, entertain mass audiences, and unveil the secrets of the universe. If I/O bandwidth is limited, those applications are limited as well. In historical perspective, PCIe2 represents yet another increase in computersâ€™ ability to help human beings visualize what they cannot see on their own. As these innovations continue to affect our lives in amazing ways, the PCIe interconnect will quietly stay one step ahead of computer graphics innovators everywhere.
AMD and Nvidia are relatively silent about plans for a PCIe 2.0-capable chipset. Nvidia will not comment on new products except to say it will have new AMD and Intel platforms available later this year. Intel, though, has stated it will soon ship boards supporting PCIe 2.0 graphics slots.
David Fair, working at Intel, manages industry evangelism and enabling for I/O technologies in enterprise server and workstation applications, including PCI Express and InfiniBand, as well as 64-bit and virtualization insofar as they affect I/O.
Wes Shimanek is the workstation strategic marketing manager for Intelâ€™s Digital Enterprise Group. He is responsible for product marketing, where he defines products, platforms, and initiatives for Intelâ€™s workstation. He has also led teams in the development and marketing of high-end computer graphic solutions used in medical imaging, immersive visualization, and simulators.