Issue No.08 - August (2006 vol.39)
Published by the IEEE Computer Society
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/MC.2006.284
Graphics-engine vendors are looking or ways to improve the processing of the calculations necessary to create more realistic effects and interactions between players and game objects.
The trend in computer gaming is always to provide higher performance and more realistic effects. Realism is important, said Rob Enderle, principal analyst with the Enderle Group, a market research firm, "whether it's allowing objects like grass and water to behave properly or cause-and-effect sequences to look real."
Gamers want large and small details to be more realistic, whether it's characters' clothes moving naturally with the wind, or explosions causing appropriate amounts of damage or debris. This greatly enhances the gaming experience, Enderle explained.
However, even the best of today's games don't look all that realistic.
With this in mind, graphics-engine vendors are looking for ways to improve the processing of the calculations necessary to create more realistic effects and interactions between players and game objects.
These calculations simulate real-world physical events and utilize the same type of mathematics used in areas such as classical physics, fluid mechanics, and ballistics physics.
Until recently, CPUs running graphics-processing-engine software have performed the limited physics processing needed for computer games. Graphics-processing units (GPUs) handle the work necessary to actually render the images themselves.
However, Enderle noted, today's complex game-physics processing entails many real-time calculations and thus is too resource intensive for CPUs.
Vendors are, therefore, turning to new approaches. Ageia has developed a stand-alone physics processing unit (PPU), while GPU market leaders ATI and Nvidia are using new physics engines that run on their chips.
These approaches use massively parallel architectures, thereby improving physics-processing performance.
However, they also add cost to gaming systems and face other barriers to adoption.
Let's Get Physical
"Nvidia released the first GPU in April 2000," said Jon Peddie, president of Jon Peddie Research, a market analysis firm.
Companies developed GPUs because graphics processing differs from standard data handling, particularly because the volume of information that must be handled quickly by today's applications requires a massively parallel approach that CPUs can't offer, explained Raja Koduri, ATI's director of engineering. For example, he said, ATI's X1900 GPU has 48 arithmetic logic units dedicated to parallel pixel processing.
Gaming has become so popular that GPU sales have risen steadily in recent years. Between 2004 and 2008, the number of shipments of GPUs and integrated chipsets that include GPUs will increase from 240.1 million to 334 million, as Figure 1 shows, with the resulting revenue growing from $5.2 billion to $8.1 billion, predicted analyst Dean McCarron with Mercury Research, a market analysis firm.
The current approach in which CPUs handle physics limits the realism of effects in today's complex games, as well as the ability of players to interact with multiple objects in a scene. Thus, vendors are turning to physics schemes that implement parallelism.
In addition to adding a physics engine, the new approaches change the nature of communications within graphics systems. The traditional GPU-based scheme requires a simple data exchange in which the CPU tells the GPU what to render, said Jason Gripp, physics-engine programmer for S2 Games.
Physics processing, on the other hand, requires more complex communications. In this case, a physics engine calculates the necessary realistic response to an event in a game, such as an explosion, and sends the information to the CPU. The CPU then communicates with the GPU, which renders the scene appropriately.
Ageia—a vendor of physics-acceleration hardware—makes the PhysX PPU. "It is a true dedicated hardware accelerator for physics problems. It offloads the work from the CPU, which enhances performance," noted Peddie.
According to Ageia chair and CEO Manju Hegde, his company's multichip approach works as follows: "The CPU thinks and orchestrates, the GPU renders and displays, and the PPU moves and interacts." In other words, the PPU performs the physics processing necessary to determine how to move the objects that will be subsequently rendered by the GPU.
The PhysX uses parallel processing across 48 pipes—about four times as many as a GPU—and several dozen cores within the chip. (Ageia declined to say exactly how many cores the chip has.) Each core is dedicated to a different type of physics calculation, such as ballistics physics for gunfire or fluid mechanics for liquid movement.
The PhysX has several high-speed buses that provide about 2 Tbits per second of bandwidth. This lets the cores exchange calculation results quickly. The cores must share information because in game physics, Hegde noted, the input of one equation frequently requires the output of another.
For example, if a player tries to use a cannon to damage a dam so that water will inundate an enemy village, the system must use the calculations of the cannon shot to figure out the shell's trajectory, which would affect the resulting damage to the dam. This will determine how the escaping water would flow.
In GPUs, Hegde noted, vendors didn't design the pipes to communicate with one another, as basic graphics processing doesn't require this.
The PhysX is implemented as part of a PCI card and will work with any graphics-chip manufacturer's hardware, communicating with the CPU via the PCI bus.
The company has released a PhysX software development kit for game designers.
PPUs are good for complex games whose effects and interactivity entail a lot of physics. However, their operations are so complex, they may take considerable time to process games that don't require extensive physics calculations, noted Enderle.
The PhysX's principal disadvantages include the $300 cost, which might be too much for non-enthusiasts, and the current lack of supporting content, according to Enderle.
Current titles supporting the PhysX include CellFactor: Combat Training from Artificial Studios, Tom Clancy's Ghost Recon Advanced Warfighter from Ubisoft Entertainment, Rise of Nations: Rise of Legends from Microsoft Game Studios, and City of Villains from NCsoft. However, 65 game publishers are currently working on 105 games that will support the PhysX.
The PPU is now in some Dell desktop computers and will soon be in gaming machines by Alienware and Falcon Northwest.
Nvidia's GPU-based approach
Nvidia uses Havok's FX physics engine to process physics on one or more Nvidia GPUs that are all on the same add-in board, said Chris Seitz, the company's developer account manager. With multiple GPUs, one chip can handle physics processing, while the rest perform graphics processing, or all the chips can share graphics and physics processing.
The system works with Nvidia's SLI technology, which runs on its own add-in card or cards and enables multiple GPUs to work in tandem to perform a task, such as rendering or physics processing.
Havok's FX engine handles effects physics, according to Jeff Yates, the company's vice president of product management. Effects physics creates a richer, more immersive environment but doesn't currently affect game-playing activities.
In the system, the CPU does only the processing necessary to determine which game objects are going to move or hit one another. It then passes this information to the GPU, which does the rendering.
The system has been optimized to create complex, highly detailed shapes efficiently and quickly, according to Nvidia spokesperson Brian Burke.
According to Peddie, adding physics engines to existing GPUs, which are not designed for physics processing, is less expensive but offers less performance than using a dedicated PPU.
Thus, Nvidia's approach is best for users who will need less-complex physics processing, said Enderle.
Games that support Nvidia's system include Eden Games' Alone in the Dark and Flagship Studios' Hellgate: London.
ATI's GPU-based approach
ATI's approach works with Havok's FX engine and an add-in card with either one ATI GPU, a card with multiple GPUs, or multiple cards that each has more than one GPU.
ATI uses the massively parallel pixel-processing portion of its GPUs to run the Havok FX engine, according to Raja Koduri, the company's director of engineering.
The GPUs have a dedicated branch execution unit and use PCI Express to provide high-bandwidth connections to the CPU and other graphics cards.
ATI's technology enables up to three add-in cards with GPUs to split up physics and rendering duties.
The company offers an API that programmers can use to write physics code for their games.
Games that support ATI's system include Alone in the Dark and Hellgate: London.
In the Marketplace
Ageia's PhysX uses a PCI slot, rather than the PCI Express slot that Nvidia's and ATI's approaches use, according to Enderle. PCs generally have more PCI slots available, so this gives PhysX a competitive advantage, he said.
In addition, he noted, PhysX offers higher performance than GPU-which based approaches and works with any graphics chip manufacturer's hardware.
According to Enderle, Ageia could dominate the market if it can get motherboard makers to build PhysX into their products. PCI buses, which the PhysX uses, tend to get saturated and cause data-transmission delays. Therefore, placing the PhysX on the motherboard would enable other ways to link the chips into the graphics system and thereby reduce the latency, he explained.
However, Enderle added, ATI and Nvidia may have an advantage over Ageia because they have more money to spend on development and marketing, which can be critical to commercial success.
Several other factors might affect how the new physics approaches fare with consumers.
For example, Havok's Yates said, many consumers might not want to spend money on a physics-only approach such as Ageia's when they could use less-expensive schemes that handle physics, rendering, and other graphics functions.
The new physics-processing approaches add cost to gaming systems, frequently necessitate the upgrade of other hardware, and tax the CPU and GPU because of the additional data exchange and processing they entail. This could delay their widespread adoption.
And, said Enderle, "Until there are games that use the new physics approaches, demand will be limited. This kind of thing could gain ground as compelling games come to market."
David Geer is a freelance technology writer based in Ashtabula, Ohio. Contact him at firstname.lastname@example.org.