But graphics performance has actually been exceeding this 40-year-old trend. For example, one company, Nvidia Corp., says its FX 5600 provides double the performance and 50% more framebuffer memory than its predecessor, the FX 5500, introduced in 2006.
The FX 5600 is based on the G80 graphics processing unit (GPU) architecture first introduced in the company’s GeForce 8800 GTS and GTX graphics cards. But while those boards were intended for gamers, the new card targets users that need interactive displays of large, complex models, such as scientific simulations, oil and gas-exploration visualizations, and aerospace and automotive designs.
As for size, the FX 5600 is a full-length board measuring 13.25-in. long. The board’s 1.5-in. thickness means it takes up as much space as two expansion cards, so the adjacent expansion slot in your PC must be empty to accommodate the card. The board plugs into a PCI Express x16 slot, but due to its maximum power consumption of 171 W, it requires two additional auxiliary connections to the computer- system power supply. And to exhaust the heat the card generates, it has a large cooling fan and plastic cowl.
On the output side, the card provides two DVI-I display connectors, each with dual-link capabilities. This lets the board power two monitors. Analog resolution tops at 2,048 1,536 pixels. The board can even power higher-resolution digital displays (for example, those beautiful 30-in. LCD panels such as the Apple Cinema HD) at up to 2,560 1,600 resolution and those with 3,940 2,400 resolutions at 24/41-Hz refresh rates. There’s also a VESA stereo connector and the board supports Nvidia’s Quadro G-Sync and SDI option cards, for frame lock and broadcast-quality video capabilities. The FX 5600 is also one of the first graphics cards to support High-Bandwidth Digital Content Protection (HDCP), which plays commercial HD DVD and Blu- Ray movies on a PC.
The card also supports Scalable Link Interfacewhat (SLI). This lets users combine the power of two Quadro FX PCI Express graphics boards to improve the performance of a single application on one display. Users can also drive four displays from one workstation, or run two applications simultaneously on two displays, with each application having its own GPU.
The new card has a 384-bit memory width, compared to 256 bits in the older model. The FX5600 also has 681 million transistors, compared to 278 million in the FX 5500, thereby, exceeding Moore’s Law, and it comes with 1.5 Gbyte of onboard Graphics Double Data Rate 3 (GDDR3 memory), the most on any available graphics accelerator. This produces a 76.8-Gbyte/sec memory bandwidth. The company says the FX 5600 delivers 300 million triangles/sec and 19.2 billion texels/ sec. It also features a 128-bit memory interface to maintain high accuracy and 12-bit subpixel precision. The board supports Open GL 2.1 and DirectX 10 with its new Shader Model 4.0 for complete hardware acceleration of even the latest professional 3D graphics software, including that running on Microsoft Vista.
Perhaps best of all, the FX 5600 features the company’s new G80 GPU unified core architecture. Previous graphics-card generations had multiple pipelines separate banks of vertex and pixel shaders that could only be used for accelerating vertex calculations and shading. This meant that portions of the GPU often sat idle while others were maxed out.
For example, games often use models with relatively few polygons with lots of textures, while CAD models often have many polygons with few or no textures. The new G80 architecture dynamically allocates the GPU’s power between vertex and pixel shading. This improves performance for all types of users. In fact, the GPU has an array of 128 parallel, 1.35GHz processor cores that harness massive floating-point computing power.
To check out the card’s performance, I tested it and the FX 5500 in the same computer, using the SPEC viewperf 9.0 benchmark. While the newer FX 5600 was clearly faster than its predecessor, it wasn’t twice as fast, at least not on the benchmark. But since the benchmark is a synthetic analysis and none of its datasets really tax the board’s capabilities, I’d expect to see a more significant performance improvement in real-world applications, particularly when dealing with large models.
Nvidia Corp., 2701 San Tomas Expressway, Santa Clara, CA 95050, (408) 486-2000 (www.nvidia.com), does not normally sell its products directly to end users. Instead, boards come installed in new PCs, or can be purchased separately from resellers and VARs, such as Parsippany, N.J.- based PNY Technologies Inc., www2.pny.com. But the FX 5600 is an exception, one of a few products now available through the company’s new online store. The Quadro FX 5600 has a suggested retail price of $2,999.