A physics accelerator ( PPU for English : Physics Processing Unit ) was a special type of coprocessor for the independent calculation of primarily physical effects. Similar to the use of a graphics processor (GPU), the use of a PPU should relieve the main processor (CPU) of a computer.
Physics engines play an important role in modern computer games for simulating a realistic environment. It is therefore obvious to outsource frequently occurring, similarly structured calculations such as the collision of two bodies to an optimized coprocessor. With the help of a physics accelerator, it was possible to calculate physical effects faster than on a main processor or at least to outsource them, which allowed more precise and extensive calculations and / or relieved the main processor, so that it had more computing time for other tasks, for example for calculating artificial intelligence , was available.
The first commercially available physics accelerator called PhysX from Ageia was presented in March 2005 at the Game Developers Conference in San Jose. This supported the simulation of clothing, hair, solids and liquids, but could also be used for collision detection and calculation. Meanwhile, all common physics engines are either GPU-based or pure software solutions (computing on the main processor).
In order to be able to use the functionality of a physics accelerator, the physics engine used in a computer game had to support it. The physics engine thus formed the interface between the application (e.g. computer game) and hardware by providing the application with the interfaces of the PPU. A physics engine usually ran without a physics accelerator, but benefited from the presence of one.
The first expansion cards with a PCI interface were sold in 2006. Ageia was the manufacturer of PPUs . Dedicated chips or integrated IP cores were sometimes sold in graphics cards from BFG and Asus from May 2006. They stuck to Ageia's reference design. Ageia was acquired by NVIDIA in 2008 . From the 8000 series, PhysX is integrated on their graphics cards. Some complete PCs were sold together with graphics cards from BFG with a physics accelerator.
Graphics cards as physics accelerators
Alternatives to an independent PPU are motivated by the relatively high acquisition costs of currently at least € 80 (as of 2008).
In the procedure, known more generally as GPGPU (for General Purpose Computation on GPU ), suitable graphics processors are used for computationally intensive tasks beyond the actual graphics algorithms. The two large graphics chip manufacturers ATI and Nvidia have already developed and presented systems and corresponding interfaces specifically for physics acceleration .
- NVIDIA is based on the PhysX engine. A graphics card of the 8800 series or higher is required to use the technology called PhysX . In addition to the calculations for the realistic representation, simulations for smoke and liquids, for example, then run on the graphics processor. From the 1990s series of the forceware driver support, SLI systems are to receive an additional mode. With SLI Physics , one of the two installed graphics cards takes over the physics calculations, the other does the calculations for image synthesis. With the introduction of the nForce 680i chipset , NVIDIA offers the option of using a third graphics card for the physics calculations.
- ATI announced CrossFire Physics in June 2006 , which uses up to three GPUs from separate graphics cards.
One or two GPUs do the image calculations, and an additional GPU does the physics calculations. The mode mentioned in NVIDIA SLI Physics is also supported by a system with two CrossFire GPUs.
The great advantage of GPUs over CPUs is the high degree of parallelism that can be achieved in the calculations. Much has been done in the past few years through architecture optimization and instruction set expansion such as SSE in the area of main processors - they are nowhere near as powerful as modern GPUs.
Furthermore, the architecture of a GPU already contains presorting and thus throughput-saving elements (e.g. Z-Buffer ), which help to reduce the required effort. These elements are reused in similarly structured calculations (e.g. collision detection).
Physics calculations on GPUs use a rigid body element known as a debris primitive , which can be efficiently implemented using the Shader Model 3.0. These units are either already existing parts of the scenery (e.g. static wall) or are generated dynamically if required (e.g. projectiles flying around). For example, in the case of collision detection, the two primitives wall and projectile are analyzed.
In the summer of 2008, NVIDIA released a forceware driver for the first time that made it possible to render the PhysX effects by the GPU. The prerequisite for this are cards with Shader Model 4, i.e. from the Geforce 8000 series. PhysX can be calculated on a separate card in SLI systems, on the second GPU in dual GPU cards or on a single card with graphics, in which case only free shaders are used so as not to influence the graphics performance too much.
- ATI CrossFire ™ Introduces 'Boundless Gaming' to the World. - ATI press release, June 6, 2006. (English)
- Asymmetric Physics Processing with ATI CrossFire. ( Memento of the original from February 18, 2008 in the Internet Archive ) Info: The archive link was automatically inserted and not yet checked. Please check the original and archive link according to the instructions and then remove this notice. Whitepaper, ati.com, 2006. (PDF, English)