Code Parallel or Die, Part 2

re: Code Parallel or Die, Part 2

Bill C — Mon, 07 Jul 2008 02:21:05 GMT

Bigger, better, faster, more powerful. Interesting scam. I say that because, like so many others, I actully never thought of. Until lately that is.

Since I develop my own Linux OS, Since Linux does everything that I need it to do, why bother with (sub standard technology) windows?

Unless MS is willing to remove the (over 450 MB's) of 'bloat' from a basic XP Pro install, I see no use of it.

BTW, I'm willing to put my money where my mouth is. I removed the 'bloat' repacked the ISO, created a addon-ons DVD, and drivers disk (in case I needed additional drivers) and when it was all said and done, my stripped version of XP installs in under 15 minutes, revitalises old P-III PC's and loads as fast as a intel Dual Core. It took less than a week to hack XP Pro and make it (somewhat) usable. Which begs the question, why hasn't MS taken out the garbage yet? In the end, I went back to my Linux PC. I guess bigger, better, faster isn't really all that great after all.

re: Code Parallel or Die, Part 2

Mark Czubin — Mon, 12 May 2008 21:21:23 GMT

Nice article, not something meant for port25.

Anyways, I've been doing parallel programming with pthreads(similar to winthreads), and well... it's hard. Then something else popped up, OpenMP, but guess how much fans I got around people not using gcc...

But of late I've been using shaders and Cuda, and they are truly facinating.

Today, I finished optimizing an blur filter, which filters an 512x512 image in about 170 fps compared to my old naive cpu (serial) algorithm that did blurring in about 5 seconds / frame ... :)

I truly believe parallel programming is here to stay... but with the majority of coders, it's still a huge problem. But 1 thing I learned from cuda etc. People need a decent api, either integrated in the language or easily avaible.

Cuda for example is integrated, very easy to do parallel programming with a whooping 128 cores! Shaders are harder and external api, but still much easier then winthreads or whatever ugly api.

For everyone else... let's hope the underlying API's will multi thread/multi program their own function. (like opengl/directX ... serial 3D app calling a parallel library).

re: Code Parallel or Die, Part 2

falde — Thu, 17 Apr 2008 16:57:33 GMT

A language like Timber that is designed around message-passing reactive objects would probably make it easier to write code that scales. At least i think we have to wrestle ppl away from languages like C that is designed around serial processing. Writing scalable apps in such languages is usually painful. Even if systems for message-passing and other stuff was added to make C-like languages scale i think that it would be to weak. C/C++/C#/Java etc encourages serial programming.

re: Code Parallel or Die, Part 2

Michael Ahumibe — Wed, 16 Apr 2008 16:46:38 GMT

Sounds just like the CPU 64bit marketing confusion. Not much will happen until people code specifically for it.