Welcome to Fractal Forums

The future of exploring the Mandelbulb definitely involves multiprecision (otherwise known as arbitrary precision) computation on the gpu so that one can more quickly achieve zooms of a given depth than on their cpu. I have spent a lot of time trying to find the state of the art in multiprecision computation using the GPU since no current Mandelbulb viewer supports it. Makin's Ultra Fractal 5 formula supports it on the cpu, but it becomes much slower as a function of precision and can't yet be parallelized across multiple cpus.

The state of the art on this topic, published in great detail just this January, can be found in several articles on Eric Bainville's website (http://www.bealto.com/mp-opencl.html) that include detailed descriptions, code and benchmarks which demonstrate the superiority of the gpu. Eric also has a 128 bit precision gpu implementation of a Mandelbrot viewer, available here (http://www.bealto.com/mp-mandelbrot.html).

I believe we can fully expect multiprecision arithmetic libraries on the gpu in the near future because of its need in the sciences, particularly molecular biology and not to mention the science of fractals ;) As an example, the MPIR project (http://www.mpir.org/), which is a fork of a recent version of the GMP multiprecision library (http://gmplib.org/), has parallel implementation and GPU support as a primary goal. Unfortunately they have not yet produced more than proof of concepts. Nonetheless I plan to keep a keen eye on their progress.

I'm not sure what the minimal subset of GPU optimized functions that we need is. It seems like we might be able to get away with addition and multiplication by using as many trig identities as possible, but I'm not sure. If that is the case, perhaps we can already optimize an existing GPU renderer (for example, the DirectX 11 version (http://www.fractalforums.com/mandelbulb-implementation/directx-11-compute-direct-x-10-pixel-shader-implementation/)), so that it supports zooms of arbitrary depths.

UF does support distributed rendering, so maybe not multi-CPU, but surely multi-machines. (I am not not sure if it still supports that when arbitrary precision is necessary)

cool, exactly what i was looking for

Old topic, but I was searching around google for similar info.

  A slight update.  ChaosPro is distributed across multiple cores (not sure about multiple CPUs).  The new UF may be <distributed across multiple cores> as well???

  Also, the new GPUs now seem to include double precision (see page 5 of this white paper (http://www.nvidia.com/content/PDF/fermi_white_papers/NVIDIA_Fermi_Compute_Architecture_Whitepaper.pdf)) native in the hardware- but I could be reading marketing gimmicks?

 

 

This seems to be an arbitary precision library for CUDA, for linux c++.
I don't even see the build option in Vcc express so somone with Linux c++ would have to check it out.

http://code.google.com/p/gpuprec/

  I'm sure it could be ported...  :D 

  Nice find. 

Nice find! I will keep this in mind the next time I have some free time...perhaps over the summer.