Physics processing unit

A physics accelerator ( short for English PPU: Physics Processing Unit) is a special processor for independent calculation primarily physical effects. Analogous to the use of a graphics processor (GPU) to relieve the use of a PPU the main processor (CPU ) of a computer.

Physics engines play an important role to convey a realistic environment in modern computer games. It therefore makes sense to outsource frequently occurring, a similar structure calculations, such as the collision of two bodies to a processor optimized for this purpose. By means of a physics accelerator, it is possible to calculate more accurate physical effects, extensive, and above all faster than a main processor. This offloads the main processor, then more processing time available for other tasks, eg a calculation of Artificial Intelligence.

The first commercially available physics engine called PhysX Ageia was presented in March 2005 at the Game Developers Conference in San Jose. This supports the simulation of clothes, hair, solids and liquids, but can also be used for collision detection and calculation.

Software

In order to use the functionality of an accelerator physics, the physics engine used in a computer game must support this. The physics engine thus forms the interface between the application (eg, computer games ) and hardware by providing the application, the interfaces of the PPU available. In the case of PhysX specifically for this purpose the PhysX API (formerly Novodex ) was developed. A physics engine is running even without physics accelerators, however, benefited from the presence of such.

PPUs also be mounted in brackets: as Sony and Microsoft have respectively the integration of the PhysX API in the PlayStation 3. Xbox 360 agreements with Ageia taken. Offer or terminate support for PhysX also provider of 3D software such as Autodesk, Softimage and Futuremark.

Products

Currently supports PhysX both the PCI and the PCI Express interface and will thus probably offered based on an expansion card. For the future, the integration is planned into the motherboard. Even the implementation on a graphics card is possible.

The first expansion cards with PCI interface were expected by the end of 2005. The launch postponed due to the small number at this time PhysX support software. The graphics card manufacturer BFG and Asus sell for around mid-May 2006 their expansion cards with PhysX PPU. The expensive between 180 and 200 € products, keeping on the reference design of Ageia and are sold under the name " PhysX P1 Graw Edition" ( Asus ) and " PhysX Accelerator " ( BFG ). Ageia was bought by NVIDIA. From the 8000 series PhysX is on the card included. In the complete systems from Alienware, Dell and Falcon Northwest, however, been rather the physics accelerator of BFG were integrated.

Graphics card as a physics accelerator

Alternatives to a separate PPU are motivated by the relatively high cost of currently at least 80 € (as of 2008 ).

In more general than GPGPU designated ( for General Purpose Computation on GPU) procedure suitable GPUs for compute-intensive tasks beyond the actual graphics algorithms are used. Especially for physics acceleration, the two major graphics chip maker ATI and Nvidia have already developed systems and corresponding interfaces and presented.

• NVIDIA is based on the PhysX engine. Prerequisite for the use of PhysX technology is called a graphics card of 8800 series or higher. In addition to the calculations for the realistic representation then run simulations for example, smoke and fluids on the GPU. SLI systems are to receive support from a further mode of the 90 - series of the ForceWare driver. In SLI Physics one of the two graphics cards installed, assumes the physics calculations. Different calculations for image synthesis With the introduction of the chipset NVIDIA nForce 680i offers the possibility to use a third graphics card for physics calculations.
• ATI announced in June 2006 Crossfire Physics, which uses up to three GPUs discrete graphics cards.

Here, take one or two GPUs the image, an additional GPU physics calculations. The mentioned at NVIDIA SLI Physics mode supports a system with two CrossFire GPUs also.

The great advantage of GPUs over CPUs is the high achievable parallelization of the calculations. Although it has been done in recent years by architecture optimization and instruction set extensions such as SSE in the main processors much - the power of modern GPUs to achieve this is not by far.

Furthermore, the architecture of a GPU already includes vorsortierende and thus throughput gentle elements (such as Z- buffer) that will help in reducing the effort required. These same elements are reused in a similar structure calculations (eg collision detection ).

Physics computations on GPUs use a process called Debris Primitive rigid body element, which can be efficiently implemented using the shader model 3.0. These units are either already existing parts of the scene (eg, static wall) or generated dynamically when needed ( eg, flying projectiles ). So, the two primitives wall and projectile, for example, analyzed in the case of collision detection.

In the summer of 2008 was issued by NVIDIA ForceWare drivers for the first time a version that made ​​it possible to let render the PhysX effects of the GPU. Prerequisite for this card with Shader Model 4 that is from the Geforce 8000 series. You can PhysX on a separate card in SLI systems, on the second GPU in dual GPU cards or on a single card with artwork, in which case only free shaders are used so as not to greatly affect the graphics performance, can be calculated.