Physics processing unit

A physics processing assemblage (PPU) is a committed dent advised to handle the calculations of physics, abnormally in the physics agent of video games. Examples of calculations involving a PPU ability accommodate adamant anatomy dynamics, bendable anatomy dynamics, blow detection, aqueous dynamics, beard and accouterment simulation, bound aspect analysis, and fracturing of objects. The abstraction is that specialized processors offload time arresting tasks from a computer's CPU, abundant like how a GPU performs cartoon operations in the capital CPU's place.

The aboriginal PPUs were the SPARTA and HELLAS.

The appellation was coined by Ageia's business to call their PhysX dent to consumers. Several added technologies in the CPU-GPU spectrum accept some appearance in accepted with it, although Ageia's band-aid is the alone complete one designed, marketed, supported, and placed aural a arrangement alone as a PPU.

AGEIA PhysX

The aboriginal processor to be advertised as a PPU was alleged the PhysX chip, alien by a fabless semiconductor aggregation alleged AGEIA. Games adulatory to booty advantage of the PhysX PPU charge use AGEIA's PhysX SDK, (formerly accepted as the NovodeX SDK).

It consists of a accepted purpose RISC amount authoritative an arrangement of custom SIMD amphibian point VLIW processors alive in bounded banked memories, with a switch-fabric to administer transfers amid them. There is no cache-hierarchy as in a CPU or GPU.

The PhysX was accessible from three companies affiliated to the way video cards are manufactured. ASUS, BFG Technologies1, and ELSA Technologies were the primary manufacturers. PCs with the cards already installed were accessible from arrangement builders such as Alienware, Dell, and Falcon Northwest.2

In February 2008 afterwards Nvidia bought Ageia Technologies and eventually cut off the adeptness to action PhysX on the AGEIA PPU and NVIDIA GPUs in systems with alive ATi/AMD GPUs, it seemed that PhysX went 100% to Nvidia. But in March 2008, Nvidia appear that it will accomplish PhysX an accessible accepted for everyone,3 so the capital graphic-processor manufacturers will accept PhysX abutment in the abutting bearing cartoon cards. Nvidia appear that PhysX will additionally be accessible for some of their appear cartoon cards aloof by downloading some fresh drivers.

Havok FX

The Havok SDK is a above adversary to the PhysX SDK. Used in added than 150 games, including above titles like Half-Life 2, Halo 3 and Dead Rising.6

To attempt with the PhysX PPU, an copy accepted as Havok FX was to booty advantage of multi-GPU technology from ATI (CrossFire) and NVIDIA (SLI) application absolute cards to advance assertive physics calculations.7

Havok's band-aid divides the physics simulation into aftereffect and gameplay physics, with aftereffect physics actuality offloaded (if possible) to the GPU as Shader Model 3.0 instructions and gameplay physics actuality candy on the CPU as normal. The important acumen amid the two is that aftereffect physics do not affect gameplay (dust or baby bits from an explosion, for example); the all-inclusive majority of physics operations are still performed in software. This access differs decidedly from the PhysX SDK, which moves all calculations to the PhysX agenda if it is present.

Since Havok's accretion by Intel, Havok FX appears to accept been shelved or cancelled

GPUs vs PPUs

The drive against GPGPU is authoritative GPUs added and added acceptable for the job of a PPU; DX10 adds accumulation abstracts types, unified shader architecture, and a geometry shader date which allows a broader ambit of algorithms to be implemented; NVidia CUDA provides a little added in the way of inter-thread advice and scratchpad-style workspace associated with the threads.

Nonetheless GPUs are congenital about a beyond cardinal of best latency, slower threads, and advised about arrangement & framebuffer abstracts paths; this distinguishes them from PPU's & the Cell as actuality beneath able-bodied optimized for demography over bold apple simulation tasks.

The Codeplay Sieve compiler supports the PPU, advertence that the Ageia physX dent would be acceptable for GPGPU blazon tasks. However Ageia assume absurd to accompany this market

Intel Larrabee & AMD Fusion

It is speculatedby whom? that Intel's Larrabee (a throughput-optimized many-core accomplishing of the x86 architecture) will be adapted to the role of a PPU; like the Cell, it sits amid the CPU and the GPU in the spectrum of accepted purpose processing against specialized high-performance back-end processing. Intel has accepted that Larrabee's anamnesis architectonics will not use scratchpads like the Cell or Ageia PPU, and will instead be afterpiece to a accepted CPU accumulation hierarchy. However, it will accept extensions to accredit high-throughput accretion (most acceptable a abounding accompaniment of cache-control instructions).