Where are the CUDA apps on Linux

[regomodo]

#

[ssam]

http://www.hep.manchester.ac.uk/gpmad/

(I share an office with one of the devs )

[regomodo]

#

[regomodo]

#

[MTTR]

NV should fund some good programmers to get support into some of the biggest number crunchers (ffmpeg, IDS, databasing, etc). I can't imagine they would have to spend a fortune before more of the SOHO/DIY groups started jumping onto the sub-$200 products.

In retrospect I always imagined a flash/CPU/RAM/FPGA PCI package that would allow rapid programming and offloading at consumer prices. Looks like dreams are coming true, but ffmpeg is the dealbreaker IMO.

Regards

[Polygon]

Ah, so somebody has done something with CUDA on Linux then. I've just installed the CUDA SDKand toolkit on windows xp. Damn, that is some seriously impressive stuff. The computing capabilities are crazy.

its not that we dont care, its just the fact that, hey! windows has 90% of the marketshare! of course they have all the cuda apps at the moment, because those developers most likely use windows.

don't worry, in time we will get some, although i don't really want seperate applications, but rather integration with current applications. Ripping a dvd with handbrake + cuda, or transcoding a video with avidemux + cuda would be awesome. Hell, even distributed computing programs like folding@home and all teh BOINC projects could greatly benefit from this. Be patient =)

[Eclipse.]

So nobody cares for the lack of CUDA apps in Linux. Windows gets all the good apps.

Feel free to code some yourself.

I'm pretty sure there will be some CUDA related projects out there.

[Swagman]

Only thing I want it for is crunching Rosetta

[MTTR]

FWIW I am no expert at these things but here are a couple cool ideas (no idea on the viability):

ARCH module to extend CPU capabilities (similar to math coprocessor?) Perhaps ASMP (asymmetrical multiprocessing).

A debugger to CUDA compiler for attaching to running sections of existing code and build portions of offload machine code. Maybe realtime with some serious -fu, more probable would be either threshhold detection or user initiated.

All of this would probably require a lot of co-op between the CPU and GPU, but maybe some of this has already been accomplished elsewhere?

[peterhoeg]

its not that we dont care, its just the fact that, hey! windows has 90% of the marketshare! of course they have all the cuda apps at the moment, because those developers most likely use windows.

don't worry, in time we will get some, although i don't really want seperate applications, but rather integration with current applications. Ripping a dvd with handbrake + cuda, or transcoding a video with avidemux + cuda would be awesome. Hell, even distributed computing programs like folding@home and all teh BOINC projects could greatly benefit from this. Be patient =)

BOINC is already using CUDA: http://boinc.berkeley.edu/cuda.php and claiming "These applications run from 2X to 10X faster than the CPU-only version.".

So we're getting there.