You have a leak.

[kayfess]

Methodology:

1. Reboot boxes and launch game on identical Vista and XP boxes both with 4Gb of DDR 1200 RAM and ~1Tb extremely high performance RAID.
2. Load same save game on both boxes, docked at station in Heretic's end and begin the Venus access to where ships jump in.
3. Note performance.
4. Reboot boxes and relaunch game to previous save game.
5. Pause game after loading on both systems.
6. Resume play 36 hours later and note performance when ships jump in.
7. Reboot boxes and relaunch game to previous save game.
8. Alt-tab out while docked at station after loading.
9. Leave boxes entirely alone for 36 hours before alt-tabbing in and noting performance when ships jump in.
10. Reboot boxes and relaunch gme to previous save game.
11. Resume play immediately and note performance when ships jump in as per 2 & 3.

Finding:

At 3 and 11: Performance is nominal - not great but not the worst - minor freezes and pauses.

At 6 and 9, performance is unplayable with single frames between 2-4 second pauses until the mission ends.

Code: Select all

------------------
System Information
------------------
Time of this report: 11/30/2008, 15:30:40
       Machine name: FILET
   Operating System: Windows Vista™ Ultimate (6.0, Build 6001) Service Pack 1 (6001.vistasp1_gdr.080917-1612)
           Language: English (Regional Setting: English)
System Manufacturer: NVIDIA
       System Model: 132-CK-NF78
               BIOS: Phoenix - AwardBIOS v6.00PG
          Processor: Intel(R) Core(TM)2 Duo CPU     E8400  @ 3.00GHz (2 CPUs), ~3.0GHz
             Memory: 2814MB RAM
          Page File: 1117MB used, 4721MB available
        Windows Dir: C:\Windows
    DirectX Version: DirectX 10
DX Setup Parameters: Not found
     DxDiag Version: 6.00.6001.18000 32bit Unicode
...
---------------
Display Devices
---------------
        Card name: NVIDIA GeForce 8800 GT 
     Manufacturer: NVIDIA
        Chip type: GeForce 8800 GT
         DAC type: Integrated RAMDAC
       Device Key: Enum\PCI\VEN_10DE&DEV_0611&SUBSYS_C8013842&REV_A2
   Display Memory: 1651 MB
 Dedicated Memory: 499 MB
    Shared Memory: 1151 MB
     Current Mode: 1920 x 1200 (32 bit) (60Hz)
          Monitor: SyncMaster 245B/245BW/245BPlus(Digital)
...
Driver Attributes: Final Retail   (is that nv or monitor driver?)
 Driver Date/Size: 9/17/2008 22:55:00, 5963776 bytes
      WHQL Logo'd: Yes

Against my own better judgement, I'm currently using the onboard realtek audio. One of my XFi boards died and I'm just coasting on the realtek (running 5.1 surround sound) until I can be bothered to buy two replacement cards.

[kayfess]

Oh, and I disabled all powersaving settings before starting the tests to rule out some odd powersaving issue.

[AjRyder]

yep its a problem at the moment BUT they are working on it!!!!

[kayfess]

Handy Hint:

Repeatedly unloading items you are going to need to immediately reload is probably not a good part of any garbage collection strategy.

I noticed then when re-running the mission (after several of my ships destroyed themselves).

Every time a different weapon fires or a different explosion goes off or a different missile is launched or detonated or every time the radio noises for start and end of transmission played, the game froze and the UI vanished.

It would seem that fundamental UI components - like the UI itself - are either not being properly reference counted (why, after all, would you flush elements that are in use?) or aren't being registered as sticky cache objects. Obviously they are gonna get instantly reloaded, so there's really no point to unloading them. "defragmenting the cache!" no, that's either nonsense or confessions of even worse designs under the hood: those should be the first things loaded into the cache by their very nature; all unloading them is doing is forcing you to do two lots of additional work.

[frymaster]

This appears to be the crux of the issues some people, but not others, are experiencing. If you watch the program's virtual size when you change sectors, you can see it spike up to the 2 gig point, whereupon some stuff is unloaded and the memory usage drops to ~1.2 gigs (and builds up again as other stuff is loaded)

I _never_ have this happen to the extent of the HUD unloading during flight, but some people appear to be hitting the memory ceiling earlier than others. The only difference I've been able to find is that one person had around 80 megabytes larger discrepancy between private bytes and virtual size than me.

It's not _merely_ a matter of GC - I suspect how often an object was used plays a large part in decisions about what to throw out, of course - in that on a 64-bit system with LAA turned on (to make the process use up to 4 gigs of memory), the memory usage will never drop until you hit the 4 gig barrier. So obviously, when new objects need loaded, it only unloads others if it's either close to the limit, or if the memory allocation failed.

Basically what I'm saying is, it's not unloading the HUD completely randomly, it's unloading it because of a critical memory starvation. You're right, of course; the HUD and a few other things should probably be special-cased to not do this, but a better question is "why are only some people experiencing this level of memory starvation?"*

* I'm currently _not_ running with LAA set, in the hope of finding the similarities and differences

[AjRyder]

I have only ever had the big freeze once, and I just reloaded and played the same mission again and everything was ok. Im only running with 2 gig of ram but I do shut off everything else in task manager before I play the game, just to be safe. I dont need to do this as it seems to run ok with everything loaded at start up but since I hit the wall I have been just making sure I have enough free memory!!!! I am sure it will be fixed soon.