Seeking Help With MAF & Other Crashes

shenryyr said:

v26 with same-day patch

Posting this as a separate thread in order to not clutter any further. Sorry other threads!

I'm using 32bit win7 with 4gb (2x2gb paired) ram and a 1gb video card, viewports are on with size 30x30, and all the graphics settings at low/minumum except medium render quality (on low, I get pink "missing" textures on some tiles)

I have been keeping an eye on memory usage, and the game MAF crashes at around 1.6gb pretty regularly. From what people have already said, this should not happen with so little memory being used, and yet it does. Frequently. What could be causing this?

The non-MAF crashes happen quite randomly, I will surely post some minidumps from those, but they don't seem to be repeatable so save files might not be of use.

Really important to note... it does not crash in early game stages, only once we are about 10% into a snail prehistoric game, when there is more uncovered map, more units, etc, and the crashes come closer together as we progress from that point.

Here's a couple non-maf crashes with minidumps, savefiles, and bbai if that helps at all
http://forums.civfanatics.com/showpost.php?p=11844097&postcount=358

any help would be of course appreciated

Click to expand...

ls612 said:

Are you still using bandicam? I thought that Koshling found that the issue was bandiam's dynamic hooking during runtime, and that he said you should use different video capture software. Are these crashes while you are recording or not?

Click to expand...

shenryyr said:

not recording

Click to expand...

Koshling said:

I have reproduced one of your non-MAF crashes, and it's something to do with viewports, but I have not yet been able ro pin it down. Worst cases, I'll give you an option to turn off mission animations (which are strongly implicated) when viewports are in use. That will probably prevent the (non MAF) crashes, so at least it'll allow you to continue while I try to find a more complete solution. I'll do this tomorrow of I don't get to the bottom of the root cause first.

Click to expand...

shenryyr said:

cool beans.

any ideas on the MAFs? like why would it fail to allocate memory when there's plenty available? I have done the 3gb thing a while back and they still occur(albeit less frequently than without the 3gb trick)

Click to expand...

Koshling said:

Not a clue on why you are getting MAFs at those memory usage levels. I've fixed one non-MAF and am working on another currently (well, after I'm done working today I will be), both from one of your earlier save games.

Click to expand...

ls612 said:

I've been having the same issue in some of my tests, it seems that for a 64 bit OS Civ refuses to allocate more than ~2.5 GB for me, according to the debugger. It might make sense to take a look at what is happening, as I'm pretty sure Win7 isn't asking for 1.5 GB of kernel space.

Click to expand...

Koshling said:

2.5 has always been about my limit also, and we've never figured out why, but it doesn't happen in our code so TBH there's nothing we can do about it apart from trying to live inside that 2.5ish footprint.

I'm wondering (especially for multimaps where processing will probably tend to be map-localized in time) whether it's possible to page entities in from serialization in 64-bit memory space (so dehydrate and reference only by id, then rehydrate on access). Could either use AWE, or named pipes to a 64-bit process. We'd need strong locality of reference though, or the performance overhead would be nasty. I might experiment along these lines sometime...

Click to expand...

AIAndy said:

I wonder, could we force some of our allocations that we know are DLL only to be in the upper part of the virtual address space so if there is any code that cannot deal with full 32bit pointers it still finds some space in the sub 31bit part?

Click to expand...