END OF AN ERA, FRACTALFORUMS.COM IS CONTINUED ON FRACTALFORUMS.ORG

1st 4 me



http://www.fractalforums.com/index.php?action=gallery;sa=view;id=8677

First Mandelbulb picture I rendered with Java

As far as I know Java is not that much slower than other languages anymore,
look for "java vs. c++ benchmark" and you will find a lot of interesting tests that prove
this wrong.
What I'm sure of is that my program is a lot slower than some GPU-program,
because all these programs seem to calculate the fractals in realtime.

The advantage with my program is that it already uses as many cores as given.
I have a quadcore so the picture below was calculated with ~98 % of all four core's power
and rendered within three seconds.

I'm just curious to see what Java can do with 3D-Rendering and want to improve my program as far
as the 2D-program has.

The best way to make an argument like this is to bring your own evidence to the table; "you can Google for other people's arguments to support my point of view" isn't a very strong statement.

Just saying.

All I really meant to say with the above post is the more general point that you could Google to find support for UFOs, 9/11 being an inside job, the existence of Gods, the Earth being flat, etc Doesn't necessarily mean it's true

But, more specifically, there is a fundamental issue in Java that all array accesses are bounds checked. So what boils down to a single MOV instruction in some languages, becomes a sequence of comparisons with possible exceptions being thrown, etc... this is absolutely fundamental, since in almost all programming situations you're doing a lot of array access (eg. in loops, where all the time is spent).

I'm not saying Java is extremely slow, and I'm not disputing that it's gotten a lot faster recently (eg. they finally introduced a stack instead of allocating everything on the heap). However, there are fundamental reasons why the performance will not be able to match that of lower level languages in many situations.

Your asking me to provide my own evidence is misplacing the burden of proof (I'm not the one making bold claims about language performance), however I've written "a few" ray tracers in my time, and found Java to be a lot slower. That was before Java 6, on older CPU architectures (Athlon 64), so I'm not going to take it as being indicative of what performance is like now... but there are fundamental reasons such as these array bounds checks which should bear consideration. The point you made about accuracy and something being "impossible in C++" is laughable and not worth a proper refutation.

Obviously if you're just doing file I/O and calling system libraries (as the majority of Java apps seem to be doing), you're not going to notice any performance difference; however, we're here to do some serious computation, and we're doing it ourselves.

Well, I can briefly justify my (admittedly sharp) comment about something being impossible to compute with C++, on the basis of it being Turing complete: http://en.wikipedia.org/wiki/Turing_completeness

I really like Java btw, and C# even more so. They're just usually not the best tools for high performance computing, and as you correctly point out, GPU computing is an entirely different matter because the CPU need only fire off the process and do a bit of memory copying

If you're keen to learn about that, there are a number of excellent CUDA (NVIDIA only) and OpenCL (vendor agnostic) tutorials and examples on the internet - this is a good place to ask about specific issues, such as performance for high precision arithmetic (there's a good thread on that topic here).

Funny you mention turing completeness 
Have you ever tried to "write a program" with a turing machine before ? It's fun 
I wrote a simple calculator for binary numbers, e.g.  1011+11010 = ? or 101*11011 = ?
with a selfmade turing simulator. It would be really hard to code the mandelbrot with a turing machine 
But I know where you're going and what you imply by saying C++ can handle everything, in that way it's correct.

That aside I think I might look for CUDA sometime, my NVIDIA could want some hard work to do 

265 sec. on a 3.0GHz Quadcore for 1800x1200 pixels I understand.
With 12*12 supersampling you mean actually 144x the amount of pixels calculated and then antialiased ?
Seems to be a good speed to me and may actually be faster than mine.