[Dave Baumann]

A while ago we mentioned the site GPGPU, a site looking at more general purpose applications for programmable 3D graphics accelerators such as the current line of DirectX capable boards from ATI and NVIDIA. One problem with graphics accelerators becoming truly general purpose is that it’s very difficult in terms of programming to achieve anything useful. However, research at Stanford University may make this more of an eventual reality via BrookGPU.

The BrookGPU project brings the ANSI C “Brook” data parallel language across to graphics processors. Programmers can specify which part of the code to run specifically on the graphics processor while the rest is executed normally on the CPU, potentially making the graphics processor a vector co-processor. Being an environment similar to C makes this type of programming much more accessible to a wider range of developers.

Although the uses are fairly limited, there are some applications that would make sense to port across to the graphics processor in the long run – video encoding and decoding could be achieved, lifting the load from the CPU, could be on such example.

[Tim Murray]

A comment on Slashdot seemed right on the money--this is a dream come true for people who rely heavily on software synthesizers.

[Dave B(TotalVR)]

I also seem to remember somebody commenting that an Nvidia Gefroce 4/fx or something had been set up to re-order a database - at 4 times the speed of a P4.

It would be nice to see, with the upcoming concept of commercial dual-core CPU's, to have a CPU that also has the graphics core in the same package and, say, using the other processors on-die memory controller.

If this were to happen then flinging mundane tasks over to the graphics core to handle will be a matter of having the right interface between the two cores. Might suffer from the slow system ram tho, so it might not be a performance answer in terms of your 3D games (though the cost difference could change that) but those long term processor tasks such as Mpeg encoding and things could benifit massively.

[Anonymous]

Originally Posted by DaveBaumann

some applications that would make sense to port across to the graphics processor in the long run – video encoding and decoding could be achieved, lifting the load from the CPU, could be on such example.

on a related note, the Radeon 9500 (and higher) ATI cards already support hardware accelerated post-processing of DivX encoded video..

[Anonymous]

Current and next generation 3d hardware is also close to being ready for hardware accelerated raytracing. There are still some problems around floating point precision, data sorting and some other stuff, but there are already raytracing prototypes running on GPUs at certain american universities acheiving this. There is still excessive overhead for a lack of a few general instructions in HLSL, Cg and the like, but if they come, we could see actual raytracing happening interactively on our desktops.

[Anonymous]

How about limiting adwords to the article text only? Adwords in the headline are difficult to read.