Simulacra X: Update (CUDA/CTM)
well, my new "compatialbe" battery for my hp laptop finally arrived from hong kong a couple of days back so it's gonna be a linux install/set-up weekend. still not decided on which distro to install yet but i'm leaning towards vector linux.
simulacraX is still moving at a snail's pace. most of the low-level stuff is in place but not implemented (stuff such as the memory management, window management, timing... are all waiting on linux). the maths library is virtually done though. speaking of maths. i came across this just over a week ago. apparently nvidia along with their sworn rivals, amd/ati, are seriously making efforts to turn the GPU into a complete coprocessor by providing low-level access to their hardware and SDKs for development. basically there is nvidia's CUDA and amd/ati's CTM. CUDA provides a C-based tool chain for development. CTM on the other hand is much more open and low-level, allowing programmers access via assembly. IMO, this is some good shit as this allows programmers to write complex and computational algorithms for the hardware using C rather than using shaders (which are primarily for graphics algorithms). this means programmers can write libraries for stuff like AI on the GPU quite easily as well as physics (infact this is as good as having a programmable PPU). and with microsoft's vista up and about, GPUs will be invading computers near you and typically, developers will want to take advantage of these monster processors if they can help it.
well, i'm looking at how to take advantage of this for simulacraX - oh yeah! it's a bit tricky though, because the issue of platform-indepent code comes into play. these APIs are strictly for their respective hardware so logically the best way to go would be to use CTM, and implement a version of CUDA for the amd/ati cards but i have no assembly programming skills. besides a CUDA implementation is a big project on it's own i guess. with my C/C++ background, CUDA is much more easier for me to get into. i just downloaded both the CUDA programming guide and the CTM techincal reference manual... maybe i'll read them tommorrow... zzz
BTW, first we had a morpheme release party thursday and boy, did i get f**ked up soaking my head in multiple pints of kirin and hoegaarden. nasty stuff - i still haven't fully recovered.
simulacraX is still moving at a snail's pace. most of the low-level stuff is in place but not implemented (stuff such as the memory management, window management, timing... are all waiting on linux). the maths library is virtually done though. speaking of maths. i came across this just over a week ago. apparently nvidia along with their sworn rivals, amd/ati, are seriously making efforts to turn the GPU into a complete coprocessor by providing low-level access to their hardware and SDKs for development. basically there is nvidia's CUDA and amd/ati's CTM. CUDA provides a C-based tool chain for development. CTM on the other hand is much more open and low-level, allowing programmers access via assembly. IMO, this is some good shit as this allows programmers to write complex and computational algorithms for the hardware using C rather than using shaders (which are primarily for graphics algorithms). this means programmers can write libraries for stuff like AI on the GPU quite easily as well as physics (infact this is as good as having a programmable PPU). and with microsoft's vista up and about, GPUs will be invading computers near you and typically, developers will want to take advantage of these monster processors if they can help it.
well, i'm looking at how to take advantage of this for simulacraX - oh yeah! it's a bit tricky though, because the issue of platform-indepent code comes into play. these APIs are strictly for their respective hardware so logically the best way to go would be to use CTM, and implement a version of CUDA for the amd/ati cards but i have no assembly programming skills. besides a CUDA implementation is a big project on it's own i guess. with my C/C++ background, CUDA is much more easier for me to get into. i just downloaded both the CUDA programming guide and the CTM techincal reference manual... maybe i'll read them tommorrow... zzz
BTW, first we had a morpheme release party thursday and boy, did i get f**ked up soaking my head in multiple pints of kirin and hoegaarden. nasty stuff - i still haven't fully recovered.
Labels: coding, simulacra_x
posted by c0dec at 00:28
3 Comments:
Experiment time, I think you should try Dream Linux...
Nice blog.
You should check out PeakStream as well. This is a way to program GPUs (and CPUs) that is portable across processor types.
cheers dude. peakstream looks really interesting.
Post a Comment
<< Home