Editing Graphics processing unit (section)

==={{Anchor|SP-GPGPU}}Stream processing and general purpose GPUs (GPGPU)===
{{Main|GPGPU|Stream processing}}

It is common to use a [[GPGPU|general purpose graphics processing unit (GPGPU)]] as a modified form of [[stream processing|stream processor]] (or a [[vector processor]]), running [[compute kernel]]s. This turns the massive computational power of a modern graphics accelerator's shader pipeline into general-purpose computing power. In certain applications requiring massive vector operations, this can yield several orders of magnitude higher performance than a conventional CPU. The two largest discrete (see "[[#Dedicated graphics processing unit|Dedicated graphics processing unit]]" above) GPU designers, [[AMD]] and [[Nvidia]], are pursuing this approach with an array of applications. Both Nvidia and AMD teamed with [[Stanford University]] to create a GPU-based client for the [[Folding@home]] distributed computing project for protein folding calculations. In certain circumstances, the GPU calculates forty times faster than the CPUs traditionally used by such applications.<ref>{{cite web |author=Murph |first=Darren |title=Stanford University tailors Folding@home to GPUs |date=29 September 2006 |url=https://www.engadget.com/2006/09/29/stanford-university-tailors-folding-home-to-gpus/ |url-status=live |archive-url=https://web.archive.org/web/20071012000648/https://www.engadget.com/2006/09/29/stanford-university-tailors-folding-home-to-gpus/ |archive-date=2007-10-12 |access-date=2007-10-04}}</ref><ref>{{cite web |author=Houston |first=Mike |title=Folding@Home – GPGPU |url=https://graphics.stanford.edu/~mhouston/ |url-status=live |archive-url=https://web.archive.org/web/20071027130116/https://graphics.stanford.edu/~mhouston/ |archive-date=2007-10-27 |access-date=2007-10-04}}</ref>

GPGPUs can be used for many types of [[embarrassingly parallel]] tasks including [[ray tracing (graphics)|ray tracing]]. They are generally suited to high-throughput computations that exhibit [[data-parallelism]] to exploit the wide vector width [[SIMD]] architecture of the GPU.

GPU-based high performance computers play a significant role in large-scale modelling. Three of the ten most powerful supercomputers in the world take advantage of GPU acceleration.<ref>{{cite web |title=Top500 List – June 2012 &#124; TOP500 Supercomputer Sites |url=https://www.top500.org/list/2012/06/100 |url-status=dead |archive-url=https://web.archive.org/web/20140113044747/https://www.top500.org/list/2012/06/100/ |archive-date=2014-01-13 |access-date=2014-01-21 |publisher=Top500.org}}</ref>

GPUs support API extensions to the [[C (programming language)|C]] programming language such as [[OpenCL]] and [[OpenMP]]. Furthermore, each GPU vendor introduced its own API which only works with their cards: [[AMD APP SDK]] from AMD, and [[CUDA]] from Nvidia. These allow functions called [[compute kernel]]s to run on the GPU's stream processors. This makes it possible for C programs to take advantage of a GPU's ability to operate on large buffers in parallel, while still using the CPU when appropriate. CUDA was the first API to allow CPU-based applications to directly access the resources of a GPU for more general purpose computing without the limitations of using a graphics API.{{Citation needed|date=August 2017|reason=CUDA as first API}}

Since 2005 there has been interest in using the performance offered by GPUs for [[evolutionary computation]] in general, and for accelerating the [[Fitness (genetic algorithm)|fitness]] evaluation in [[genetic programming]] in particular. Most approaches compile [[linear genetic programming|linear]] or [[genetic programming|tree programs]] on the host PC and transfer the executable to the GPU to be run. Typically a performance advantage is only obtained by running the single active program simultaneously on many example problems in parallel, using the GPU's [[SIMD]] architecture.<ref>{{cite web |author=Nickolls |first=John |title=Stanford Lecture: Scalable Parallel Programming with CUDA on Manycore GPUs |url=https://www.youtube.com/watch?v=nlGnKPpOpbE |url-status=live |archive-url=https://web.archive.org/web/20161011195103/https://www.youtube.com/watch?v=nlGnKPpOpbE |archive-date=2016-10-11 |website=[[YouTube]]|date=July 2008 }}

* {{cite web |author=Harding |first1=S. |last2=Banzhaf |first2=W. |title=Fast genetic programming on GPUs |url=https://www.cs.bham.ac.uk/~wbl/biblio/gp-html/eurogp07_harding.html |url-status=live |archive-url=https://web.archive.org/web/20080609231021/https://www.cs.bham.ac.uk/~wbl/biblio/gp-html/eurogp07_harding.html |archive-date=2008-06-09 |access-date=2008-05-01}}</ref> However, substantial acceleration can also be obtained by not compiling the programs, and instead transferring them to the GPU, to be interpreted there.<ref>{{cite web |author=Langdon |first1=W. |last2=Banzhaf |first2=W. |title=A SIMD interpreter for Genetic Programming on GPU Graphics Cards |url=https://www.cs.bham.ac.uk/~wbl/biblio/gp-html/langdon_2008_eurogp.html |url-status=live |archive-url=https://web.archive.org/web/20080609231026/https://www.cs.bham.ac.uk/~wbl/biblio/gp-html/langdon_2008_eurogp.html |archive-date=2008-06-09 |access-date=2008-05-01}}

* V. Garcia and E. Debreuve and M. Barlaud. [[arxiv:0804.1448|Fast k nearest neighbor search using GPU]]. In Proceedings of the CVPR Workshop on Computer Vision on GPU, Anchorage, Alaska, USA, June 2008.</ref> Acceleration can then be obtained by either interpreting multiple programs simultaneously, simultaneously running multiple example problems, or combinations of both. A modern GPU can simultaneously interpret hundreds of thousands of very small programs.