I don't know if this is the root of your problem, but I believe you should be dividing by 1,048,576 (1024*1024).

Thanks. It makes the info display more accurate, instead of an approximation, but indeed it doesn't get to the core issue. The 'technique' I'm currently using as a workaround is to apply a safety barrier (One), to make sure memory usage doesn't go overboard. Of course you could also use the percentage for this. Given that barrier I do try to use up the free memory within that batch first (Two). However when running the below code the end result (of course no OOM, so that's good) is:

bufferSize: 1342 | max: 12444 | total: 10671 (85%) | free: 46

Which means I go from 14GB assigned to 12.4GB max to 10.6GB actually used. Even given the fact that the java app uses more in the task manager (no problem, there is 16GB in the comp), I'm still wondering why I have to give up 3.4GB. To max out further I can increase the Processing-Preferences towards 16GB.

But in the end the problem remains the same, namely that this workaround relies on trial-and-error & magic numbers. Where of course ideally there is a solution that always works without manual tweaking.

Current workaround

1. ArrayList <PImage> outputBuffer = new ArrayList <PImage> ();
2. PImage tempImage = createImage(1920, 1080, RGB);
3. int maxMemory, totalMemory, freeMemory, percentage;
4. int MEGABYTE = 1024*1024;
5. int safeBarrierOne = 2000;
6. int safeBarrierTwo = 50;
7.  
8. maxMemory = int(Runtime.getRuntime().maxMemory()/MEGABYTE);
9. for (int i=0; i<10000000; i++) {
10.   outputBuffer.add(tempImage.get());
11.   totalMemory = int(Runtime.getRuntime().totalMemory()/MEGABYTE);
12.   freeMemory = int(Runtime.getRuntime().freeMemory()/MEGABYTE);
13.   percentage = int((float)totalMemory/maxMemory*100);
14.   println("bufferSize: " + outputBuffer.size() + " | max: " + maxMemory + " | total: " + totalMemory + " (" + percentage + "%)" + " | free: " + freeMemory);
15.   if (totalMemory >= maxMemory-safeBarrierOne) {
16.     if (freeMemory < safeBarrierTwo) {
17.       println("done");
18.       break;
19.     }
20.   }
21. }
22. exit();

In this StackOverflow answer, it says to accurately calculate the amount of used memory you should use totalMemory()-freeMemory(). I substituted that in your code for the percentage and replaced the break condition with (percentage>98) which allowed me to store 123 1080 images using 99% of my 989 (1024) megabytes. Using (percentage>99) gives me out of memory.

Edit: I see what you mean about the max not being the same as what was assigned. I would guess it has something to do with addressing the amount of memory being used. It looked like 9 megabytes per 256 was lost on my computer. I didn't experiment with larger amounts yet though.

Edit: This looks like it might be helpful / related:
http://mark.koli.ch/2009/01/understanding-javas-perm-gen-maxpermsize-heap-space-etc.html

Thanks for your time and efforts. I tried your usedMemory calculation, but it still crashes.

At lower image dimensions and lower memory size the crash moment approaches 99% and sometimes works. In fact I never noticed this problem until I pushed it toward these high mem conditions. However the bigger the image size, the lower the crash percentage. It's crashing in the 70's and 80's. So a percentage of 98% is never even reached.

It seems there is some issue with Processing / Java not being able to free up memory to the maxMemory. But if maxMemory is not the max, then how do I find the actual max? And why is there so much overhead?

Hope someone has some insights.