Re: [Fractint] Slashdot deep zooming
Hey JOT, On 5/7/2015 2:00 PM, fractint-request@mailman.xmission.com wrote:
I suspect treachery. Just think about it - the math behind each Mandelbrot pixel. And doing that with a number with hundreds or thousands of decimals. No.
There are "shortcut" methods and algorithm hacks to do it , like "interpolating" between 2 images. I am a purist - do it right or it don't count!
FractInt has some real tight "C" and "ASM" code - and it took me 3 years on several systems to get past E+100.
It is possible. Looking into his discussion area he mentions "Mandel Machine". https://www.youtube.com/channel/UCsRbcv18VcWJVtfAtbKV1Vw/discussion Mandel Machine looks like it is a high precision fractal explorer which I don't see on Paul Lee's list. Looking at the history they mention using 80 bit precision. I have tried running it but it uses 64 bit Java and I have 32 bit. I am debating installing 64 bit as the claim they can co-exist but I want to do more research on that. http://web.t-online.hu/kbotond/mandelmachine/ There are ways to get much higher, even open ended precision. The GMP library can do this and I am sure while they have made it efficient, there has to be payback to performance. https://gmplib.org/ While I have the utmost respect for Fractint, it as a legacy system. For any fractal program writers out there, this question. Can your program exceed the efficiency of Fractint, i.e. draw a fractal faster at the same resolution? If so, what resources did you use to do it?
Roger Kaufman wrote:
Mandel Machine looks like it is a high precision fractal explorer which I don't see on Paul N. Lee's list.
I have it listed within my database of fractal applications and their users. But my online listing only shows a small number of the actual total available, because of the quantity involved (over 600). I list only those programs with the most users, starting at the top of the list and working my way down. You can also read some more information, including several of the user's comments and a gallery of images, at the following site: http://www.fractalforums.com/mandel-machine/?PHPSESSID=4564ced4d613a70d07fae... Sincerely, P.N.L.
On 05/07/2015 08:46 AM, Roger Kaufman wrote:
Hey JOT,
On 5/7/2015 2:00 PM, fractint-request@mailman.xmission.com wrote:
I suspect treachery. Just think about it - the math behind each Mandelbrot pixel. And doing that with a number with hundreds or thousands of decimals. No.
There are "shortcut" methods and algorithm hacks to do it , like "interpolating" between 2 images. I am a purist - do it right or it don't count!
FractInt has some real tight "C" and "ASM" code - and it took me 3 years on several systems to get past E+100.
It is possible.
Looking into his discussion area he mentions "Mandel Machine".
https://www.youtube.com/channel/UCsRbcv18VcWJVtfAtbKV1Vw/discussion
Mandel Machine looks like it is a high precision fractal explorer which I don't see on Paul Lee's list. Looking at the history they mention using 80 bit precision. I have tried running it but it uses 64 bit Java and I have 32 bit. I am debating installing 64 bit as the claim they can co-exist but I want to do more research on that.
I was going to test that since I have 64-bit Java on 64-bit Linux, but it looks like it's Windows-only? Anyway, the Java source is available, now just need to add the JDK and try javac-ing to see if it works. Don't know if I need the C/ASM files, too, not a programmer. Among the "Optimizations to speed up calculations" it lists are: * "Tweaked Marian/Silver algorithm to guess areas with monotonic iteration counts" * Extensive use of SSE2/3/AVX * Multicore support (up to 32 threads) * "pixel grouping to fully saturate execution units of modern CPUs (up to 16 pixels with SSE2, up to 32 with AVX-capable CPUs)" -- David W. Jones gnome@hawaii.rr.com authenticity, honesty, community http://dancingtreefrog.com
participants (3)
-
david -
Paul N. Lee -
Roger Kaufman