[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
On 29/07/13 19:09, Gordon Henderson wrote:
I have several 'real world' uses where parallelisation is a mere make switch away in the build but they're all pretty non-spectacular (well Blender might be good but I haven’t looked at the code). I was looking for something 'spectacular' to get people interested. I know the Pi's GPU is good but when the code will be available is another thing. By that time they may have the 64 Core parallellas which will be the 100s of Gflops and not a lot of power. Much as I love the Pi this is the way of computing. There is a PCIe card with a 16 core chip on it (no idea how much but should be in the $100 mark too) and if people get the idea that when their computer goes slow they stick in another processor card, or plug in a USB stick and move to the next level we're onto a winner. Parallelisation for a lot of problems is not difficult, and has generally been worked out pretty well for a vast array of problem areas already. Its there and ready to go ... if only you can find the right switch to use in the build process. For the problems where its tricky - your doing it wrong!On Mon, 29 Jul 2013, tom wrote:Well not quite but for $99 you get two 1Ghz ARM cores and 16 fp cores that will do 32GFLops at a push and all for 5W.Meant to say; The Pi's GPU will do 24GFLops - and one day we'll be able to program it..... However as someone who's worked for massively parallel computer companies in the past, all I'll say is: It aint going to be easy... There's no magic wand when it comes to this stuff - the compilers will help, but unless you design the algorithms to parallelise then you're not going to get much use out of it, no matter how many cores you throw at it.Gordon
Tom te tom te tom -- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/listfaq