alex@cs.qmw.ac.uk (Alex Kashko) (02/01/90)
In article <VM16}-@rpi.edu> hiebeler@cs.rpi.edu (Dave Hiebeler) writes: > Well, you are taking 0.118 seconds to update an array 4 times as >large as the CAM array (twice as wide, twice as high), so it would >take you (0.118 / 4) or 0.0295 seconds to update the CAM array. > > The CAM does it in about 0.0166 seconds. > > So you'd be getting about 33 frames/second on a 256x256 array, where >CAM gives you 60 frames/second, almost twice as fast. Thanks. The reason I said a "version" was that the Lichens with death type pattern arose as a special case of a more general program. I went back to T&M and programmed exactly what they said they did. I started with a symmetrical seed and so gor symmetrical results rather thatn the random looking result they give in the book (They don't as far as I can tell, mention what seed they used to give their figure 5.2b). I checked the algorithm against the book and it seems ok. This decicated algorithmruns in 43.2 seconds. This is some 2.7 times faster, so the DAP would be faster then the 256 by 256 CAM for this problem. The DAP610 would obvious ly run 4 times faster than this. > > Keep in mind that the CAM costs on the order of $2000, and that you >can buy a few of them and plug them in together. So you could buy >four CAMs and plug them in together to update a 512x512 array at 60 >frames/sec, which would be 8 times as fast as the DAP. Also, the CAM Or twice as fast as the DAP610.On the revised execution figures above the CAM should still be marginally faster. This is not surprising for the DAP is a general purpose computer, so the dedicated hardware should be faster. The data analysis capacity is something we don't have on the DAP. The Lego `-like extendibility is also missing. I suggested to AMT (Over a lunch break) that such extendibility would be nice, but appartently there are problems. On the costs above a 1024 by 1024 CAM would cost about $32 ,000 which is quite cheap. I must see if my boss will put in for one. Can you email me a contact adress or (and?) other details. > > About how much does a DAP such as the one you are using cost? We >have one of their machines here at RPI, but I haven't gotten around to >using it yet, although it sounds like a pretty decent machine. I >generally use the CAM, but I use Connection Machines when I need to >do something too big for CAM. I believe a DAP510 is about 100,000 sterling, but dont't quote meas I cound be wrong. There is an academic discount of around 40% (again don't quote me). So on a power per pound basis the CAM wins. But the DAP is more versatile. I am using DAP's for cellular automaton models of Polymers, and growth models are a byproduct of this. If anyone can email me references to cellular automaton and other types of simple polymer modelling (Anything welcome but I am short on electrical and electronic properties in particular) or Langmuir Blodgett films I would be grateful. >-- >Dave Hiebeler hiebeler@turing.cs.rpi.edu >Computer Science Dept. hiebeler@cardinal.lanl.gov >Amos Eaton Bldg. "xue zai shao" -- Huang Ying Ying >Rensselaer Polytechnic Institute / Troy, NY 12180-3590