diep at xs4all.nl
Tue Sep 8 13:38:47 UTC 2015
On Tue, 8 Sep 2015, folkert wrote:
> Hi Vincent,
>> hi Folkert - you probably want to consider building a small piece of gpgpu
>> No CPU library can rival that.
> Thanks for the elaborate reply.
> My goal is not neccessarily to build the fastest most accurate
> mandelbrot calculator. I was merely looking around for alternatives to
> a "big number" library that I use on an embedded platform. The one I
> use currently uses 1 byte per digit which is rather resource heavy for
> an mcu with only 2KB of ram. Now GMP is probably also a bit too heavy
> but I got carried away :-)
>> BTW very interested in the program you got now to see how with more bits
>> precision mandelbrot looks like - where is it for download?
> I somewhat documented what I did: http://www.vanheusden.com/misc/gmp/
> You'll notice that there is hardly any difference beyond 64 bit.
That's because you're not doing enough iterations.
You're just doing 8 bits of iterations (256) instead of say 16 million or
even more, at which point you'll see a GPU suddenly makes more sense at
256 bitsprecision :)
If you use more bits significance to calculate you also want more
> Folkert van Heusden
More information about the gmp-discuss