What's in the Cards?
Issue: Volume: 31 Issue: 3 (March 2008)

What's in the Cards?

Workstation graphics cards are always at the cutting edge of technology. Graphics professionals demand speed and power for a simple reason—time is money, and faster graphics gets the job done faster. The latest workstation graphics cards sport faster chips, more memory, and, most recently, support for PCI Express 2, a faster hardware interface that gets data to the card even faster (see “The Express Way,” September 2007).

High-end 3D graphics technology usually comes in two flavors of cards—those for gaming and those for workstations. While both share similar technology, their purposes are very different. Gaming cards are the razzle-dazzle cards, and are designed mostly to provide very high frame rates for gaming applications. Workstation cards are similarly speedy, but they focus more on accuracy and open standards, particularly supporting OpenGL. This is especially important for artists and professionals doing high-end work. Many high-end 3D applications, particularly Autodesk’s Maya and Softimage’s XSI, are built around OpenGL, so support for the API can be essential.

Workstation graphics cards these days are sold by two companies—ATI and Nvidia, both of which are extremely competitive. It seems as if these two companies are always running neck and neck in terms of performance, with the latest release from one company topping the previous release from another. Moreover, other companies, such as Matrox, sell workstation cards, but many of these are geared more toward 2D workstation tasks, such as those used in video editing. As for the main vendors, AMD sells its high-end 3D workstation cards as the ATI FireGL series, while Nvidia offers the Quadro series. Nevertheless, each company offers a wide array of cards, enabling customers to get as much power and capability as they need. 

The Express Lane

Current graphics cards are based on the PCI Express hardware interface, which replaced the venerable AGP interface a few years ago. All current motherboards support PCI Express, and, as of late last year, this interface has been updated to PCI Express 2, which doubles the bandwidth over the previous generation. Systems are now just filtering into the market with PCI Express 2 slots (which are completely compatible with older PCI Express cards).

While PCI Express 2 is great to have on a graphics card, at this point, it’s not critical. The original PCI Express standard was so fast that even current cards don’t utilize the entire bandwidth of the bus, and it will take a while for newer cards to catch up. But the big reason for PCI Express 2 is that it allows more devices to work together. As more and more interface cards, such as disk controllers and network interfaces, move to PCI Express, the extra bandwidth can be used by these cards, as well.

Nvidia sells its Quadro series cards in the high-end 3D workstation market.
Graphics cards supporting PCI Ex­press 2 are just arriving on the market, and any new cards from the major manufacturers will be fully compliant. Right now, AMD has introduced the mid-range V3600 and V5600, which will run at the faster PCI Express 2 speeds. This is true for Nvidia’s Quadro FX 1700, FX 570, and FX 370. The Quadro FX 3700, also from Nvidia, is the first to offer full support for all PCI Express 2 features.

Looking at the Cards

When looking at cards, customers need to consider the speed and power of the card. Until recently, the best indicator of a graphics card’s performance was the clock speed of the chip. These days, chips are highly parallelized, and clock speed is not the only indicator of how fast the card will run; another important factor is the number of streams a card can process. Most cards can handle hundreds of streams, enabling them to process hundreds of bits of data simultaneously.

Another indicator of how the card processes its data is called the Shader Model. This is a programming interface from the operating system to the card that determines what data needs to be processed by the application and what is processed in the card’s GPU. The most current cards support Shader Model 4.0, which contains a number of features, including the ability to create geometry shaders and the ability to unify pixel and vertex shaders, thereby simplifying processing tasks and speeding up the pipeline.

Memory bandwidth and memory size can also affect performance. More video memory will make a card run faster, simply because the card has to transfer data from the system less often. The more data that can be handled on the graphics card, the more resources that are freed up in the computer system and the faster everything will run.

Today, the amount of memory in graphics cards rivals system memory. The ATI FireGL 8650 currently tops the list with a whopping 2gb of video memory. The top-of-the-line Nvidia Quadro FX 5600, meanwhile, has 1.5gb of memory. These amounts are particularly useful for simulations, especially those used in the medical field as well as in oil and gas exploration. For those doing animation or 3D authoring, the standard 256mb or 512mb found in most cards should be plenty. In addition to memory size, the speed and width of the internal memory bus can also affect performance. Some cards have memory as wide as 512 bits.

The physical size of the card can be a big factor in determining which card to configure. The top-of-the-line cards are double-width behemoths, requiring two slots on the motherboard. These cards use the extra room mostly for cooling fans venting the heat produced by the power-hungry chips on the card. Some cards consume more than 200 watts of power, so users need to be sure they have a big enough power supply in their system to feed these power-hungry cards. Extra fans in the big cards can also mean extra noise, so those wanting a super-quiet system need to take this into consideration.

Going Large

There are times when one card just isn’t enough. Many motherboards support dual PCI Express slots, allowing for two graphics cards to be configured in the same system. Nvidia calls its dual-card technology “SLI,” short for Serial Link Interface. This allows for two identical cards to operate together in any of three modes. SLI Multiview combines two cards to drive multiple displays, with each card dedicated to a separate monitor. This is great for those wanting maximum screen real estate.

SLI Frame Rendering, meanwhile, allows two cards to drive one monitor with double the rendering power—ideal for those who want maximum speed. Finally, SLI FSAA (Full Scene Anti Aliasing) combines two cards to provide much higher quality on a single monitor, for those wanting the highest level of accuracy.

Instead of the complexity of multiple monitors, a lot of people are simplifying their desktops and opting for a single, large LCD display, which now measure as much as 30 inches. These monitors run up to 2560x1600, which requires that the graphics card supports a dual-link DVI interface. For a while, Apple was just about the only company selling 30-inch monitors, but in the past year, more and more companies, such as HP and Samsung, have added these large monitors to their product lines.

Prices have also dropped on these behemoths, so dual-link support will become more important as these monitors filter onto desktops. Looking to the future, the next generation of monitors will have even higher pixel densities and will support ultra-high resolutions up to 9 megapixels (3840x2400). These monitors also will need dual-link DVI, so it’s a good idea to get a card that supports this for future growth. In fact, all five cards in AMD’s latest ATI FireGL series sport two dual link-enabled DVI outputs, enabling the use of multiple high-resolution displays.

Big Decisions

As for what type of card to buy, much of that decision will be based on the customer’s budget and the type of work that is done. Most artists in 3D animation and content creation will do just fine on a midrange or entry-level card, with those doing high-end simulations or working with very large data sets needing top-of-the-line cards.

The ultra-high-end cards can cost several thousand dollars, but for those who need the absolute fastest performance, the cost is worth it. AMD’s top card is the FireGL V8650, with 2gb of video memory, a 512-bit bus, and full 320 stream processors. Nvidia’s Quadro FX 5600 sports 1.5gb of memory, with a 384-bit bus. Both of these cards are physically large (double width) and require tremendous power. They also need a full-size motherboard and chassis.

The Quadro FX 3700 runs at PCI Express 2 speeds.
Coming out of the stratosphere, high-end cards generally fall within the $1000 range and are usually fairly close to the top-of-the-line cards in terms of speed, but they have less video memory, thus making the price more manageable. These cards should be able to handle all but the toughest jobs. The Quadro FX 4500 and ATI FireGL V7600 both have 512mb of video memory, but still are very fast and require a double-wide slot to manage the extra fans needed to keep these cards cool.

Midrange cards are usually the workhorses of the business, and most professionals will opt for a card somewhere in this range. These have an excellent price/performance ratio and should be able to tackle most jobs. At the upper end of this range is the new Nvidia Quadro FX 3700, with 512mb of memory and PCI Express 2 support. The FireGL V5600 is also configured with 512mb of memory.

Entry-level workstation cards are certainly no slouches in terms of performance, and many students or independent users will choose these cards to add a professional solution at an affordable price. The FireGL V3600 has 256mb of video memory and is ideal for the CAD, DCC, and simulation markets. The latest Nvidia Quadro entry-level graphics boards—FX 570 and 370—also have 256mb of video memory and PCI Express 2 support.

Whatever your needs, rest assured there is a workstation graphics card that can provide excellent performance. Any of these cards, even the entry-level offerings, are fast enough to blow away the highest performing cards of only a few years ago. Of course, time is money, and the fastest cards will always save you more time. 

George Maestri is a contributing editor for Computer Graphics World and president/CEO of RubberBug animation studio. He can be reached at maestri@rubberbug.com.