Editing Distributed.net (section)

==Development of GPU-enabled clients==
[[Image:Average RC5-72 production.png|thumb|350px|Average daily RC5-72 production by platform for 26 May 2024 &ndash; 4 January 2025<ref name="currentwork">Calculated by subtracting the completed work units as of 4 January 2025 from the totals on 26 May 2024, creating a 223-day average.</ref>]]

In recent years, most of the work on the RC5-72 project has been submitted by clients that run on the [[Graphics processing unit|GPU]] of modern [[video card|graphics cards]]. Although the project had already been underway for almost 6 years when the first GPUs began submitting results, as of January 2025, GPUs represent 88% of all completed work units,<ref name="platformlist">{{cite web |url=http://stats.distributed.net/misc/platformlist.php?project_id=8&view=tco |title=RC5-72 / CPU Participation |publisher=distributed.net |access-date=27 May 2023}}</ref> and complete more than 95% of all work units each day.<ref name="currentwork" />

*NVIDIA
:In late 2007, work began on the implementation of new RC5-72 cores designed to run on [[NVIDIA]] [[CUDA]]-enabled hardware, with the first completed work units reported in November 2008. On high-end NVIDIA video cards at the time, upwards of 600 million keys/second was observed<ref>{{cite web|url=http://cgi.distributed.net/speed/query.php?cputype=all&arch=4&contest=rc572&multi=3|title=Client Speeds Database (GPU RC5-72 search)|publisher=distributed.net|access-date=27 February 2014}}</ref> For comparison, a 2008-era high-end single [[Central processing unit|CPU]] working on RC5-72 achieved about 50 million keys/second, representing a very significant advancement for RC5-72. As of January 2025, CUDA clients have completed almost 11% of all work on the RC5-72 project, and perform about 9% of the work each day.<ref name="platformlist" />

*AMD / ATI
:Similarly, near the end of 2008, work began on the implementation of new RC5-72 cores designed to run on [[AMD FireStream]]-enabled hardware. Some of the products in the Radeon HD 5000 and 6000 series provided key rates in excess of 1.8 billion keys/second.<ref>{{cite web |url=https://arstechnica.com/civis/viewtopic.php?p=19780660&sid=19c2e0958c0248dfd2cae221f5e737c1#p19780660 |title=Benchmark results for Radeon HD 5870 |publisher=MrJackson2000 |date=April 1, 2010}}</ref> As of January 2025, FireStream clients have completed over 21% of all work on the RC5-72 project.<ref name="platformlist" /> Daily production from FireStream clients has dropped below 0.5% as the majority of AMD GPU contributors now use the OpenCL client.<ref name="platformlist" />

*OpenCL
:An [[OpenCL]] client entered beta testing in late 2012 and was released in 2013. As of January 2025, OpenCL clients have completed more than 56% of all work on the RC5-72 project, and now perform almost 86% of the work each day. No breakdown of OpenCL production by GPU manufacturer exists, as AMD, NVIDIA, and Intel GPUs all support OpenCL.<ref>{{Cite web |date=2010-01-16 |title=[TEST] GPU Computing - GeForce and Radeon OpenCL Test (Part 3) {{!}} Geeks3D |url=https://www.geeks3d.com/20100116/test-gpu-computing-geforce-and-radeon-opencl-test-part-3/ |access-date=2023-09-05 |language=en-US}}</ref>