Graphics Processors (GPUs)
For years ATI and NVIDIA, known for their high performance graphics boards, have been the synonyms of fast games. Recently everybody have been talking about Sony PS3 and its ultra high performance graphics based on the NVIDIA RSX engine coupled with the Cell processor. And suddenly it looks like what was initially designed for games finds its way into high performance servers.
A few days ago Wired reported about an astrophysicist who replaced his number - crunching supercomputer with a cluster of PS3s. This trend becomes even more mainstream, as some serious companies offer serious programming tools. Terra Soft Solutions has a version of Linux tailored for the PlayStation and RapidMind offers its stream programming toolkit for the console. This year's acquisition of PeakStream (a competitor to RapidMind) by Google amplified the trend even more. All these platforms and solutions share a common paradigm - multiple processing cores and parallel execution of tasks. What seems now a "new" trend by Intel and AMD, has been mastered for years by the graphics chips designers. Not long ago these chips were very difficult to program and their flexibility was not enough to run generic purpose algorithms. Now this has changed and with supply of tools easing the translation of linear code to multiple processing cores more and more companies turn to GPUs.
I have recently attended a jaw dropping presentation by Jules Urbach of otoy.com. With real time ray tracing and servers equipped with GPUs they can generate high resolution virtual reality with a movie quality. They have some screenshots on their site, so looking at them you may feel what the quality they have is like, but bear in mind they can do this at 30 frames per second and everything is being synthesized from preprogrammed models. What has never been possible even with the fastest computers so far, thanks to GPUs is now within a reach even with relatively small budget.
So the next time you buy a graphics card for your gaming PC, think of it as of a single board supercomputer :)
A few days ago Wired reported about an astrophysicist who replaced his number - crunching supercomputer with a cluster of PS3s. This trend becomes even more mainstream, as some serious companies offer serious programming tools. Terra Soft Solutions has a version of Linux tailored for the PlayStation and RapidMind offers its stream programming toolkit for the console. This year's acquisition of PeakStream (a competitor to RapidMind) by Google amplified the trend even more. All these platforms and solutions share a common paradigm - multiple processing cores and parallel execution of tasks. What seems now a "new" trend by Intel and AMD, has been mastered for years by the graphics chips designers. Not long ago these chips were very difficult to program and their flexibility was not enough to run generic purpose algorithms. Now this has changed and with supply of tools easing the translation of linear code to multiple processing cores more and more companies turn to GPUs.
I have recently attended a jaw dropping presentation by Jules Urbach of otoy.com. With real time ray tracing and servers equipped with GPUs they can generate high resolution virtual reality with a movie quality. They have some screenshots on their site, so looking at them you may feel what the quality they have is like, but bear in mind they can do this at 30 frames per second and everything is being synthesized from preprogrammed models. What has never been possible even with the fastest computers so far, thanks to GPUs is now within a reach even with relatively small budget.
So the next time you buy a graphics card for your gaming PC, think of it as of a single board supercomputer :)
Did you see this at the Gildertech Conference
ReplyDeleteYes indeed :). They were taping the presentations, so ultimately they should be posted at http://www.telecosmconference.com/
ReplyDelete