Fractal collaborations - Best machine for fractals?

[DOMEHEAD]

From the varied experience of others on fractal forum I am looking for advice on the best computer on which to run fractal calculations.  From October I am putting on night time shows of films on a ten metre digital dome, including fractals at high resolution.

I'll be purchasing a machine on which others can generate fractal zooms.  What would be the best machine for such work?

The machine I have in mind is a Dell Power Edge Dual Xeon Quad Core 3 GHz server with plenty of memory, running Windows server.  Will the commonly used fractal software automatically make use of all this processing power? 

Really grateful for the advice of those that know, before I make an expensive mistake! 

[taurus]

hzi domehead,

most "commonly used" fractal software (uf, mb3d, bulber) should use all available cores.

the standard benchmark, you can orient on, is cinebench, as it is heavy on processor power. most interesting is the multi-thread-value, it measures the fp power of the system - using all cores available. for most computers (or at least for the processor used) a cinebench value can be found in the internet, so it's worth a bit of investigation. maybe an opteron system is more suitable in your case of use and for the money, but this can change every hour (when new hardware is released).

good luck  
taurus

ps: if you use 32bit software a huge ram might not be useful, as a single 32bit task can only use about 1,8gb of memory.

[ker2x]

it's really software dependent... and budget
eg : http://www.supermicro.com/newsroom/pressreleases/2011/press110531_Computex.cfm

[Syntopia]

Your system will get you about 100 GFLOPS of computational power:

3GHz x 8(cores) x 2 (SSE: 2 double precision floats per cycle) x 2 (MAD instructions per cycle) = 96 GFLOPS

(Assuming your software uses SSE instructions. If you get a Sandy Bridge Xeon with AVX, you can in principle double the GFLOPS, but that requires your software supports AVX - which is very unlikely as of now).

For comparison here is a few GPU's:
Nvidia GTX 460 (~200$): 112 GFLOPS (for doubles).
AMD 5870 (~400$): 544 GFLOPS (for doubles).
NVidia Tesla C2050 (~2400$): 515 GFLOPS (for doubles) .

No doubt GPU's offer much better performance for the money, and they are perfectly suited for 2D and 3D fractals. Unfortunately, the GPU fractal software is far behind the mature CPU fractal software at the moment.

Personally, I would go for a less high-end product: a standard Dell i7@3.4GHz desktop will cost you ~1000$ and provide about 50 GFlops :-)

[cbuchner1]

I am currently investigating multiprecision integer mulltiplication on the GPU as well as Intel AVX. Will post some source code when I've got it running satisfactory. It could be useful for in deep fractal zooms. At the moment I do not yet know whether the CUDA or the AVX versions will be faster.

If you get Xeon CPUs, aim for the new Sandy Bridge E models. Nearly twice the floating point throughput compared to earlier models. Software won't automatically use it, but it will most certainly in future releases. You need Windows 7 SP1 or Windows Server 2008 R2 SP1 for the AVX support.

[DOMEHEAD]

Thanks very much for all the comments above.  Will let you know which machine I go for.