Bizarre replay freezing

[gameragodzilla]

I recorded a Replay for Ghost Recon 1 on Fever Claw, yet every time I try to play the Replay at 8x speed, the replay freezes at various points. I am using mods, but I never ran into this issue before and my older replays on the same level with the same mods don't have this problem. Running at 4x seemed to have no issue, but why does 8x have it? Does anyone know what's going on and why it's happening?

[Zeealex]

I wonder if it could be an engine/rendering API issue. Where it's having to make so many new draw calls per frame that the engine simply can't keep up. 

Its a theory, but it would make sense given the age of the game.

So, iirc replays in GR work super weird. It's not technically a video replay but actually just a record of every move every character took and every shot that was made and what the effect of that shot was being replayed through the mission file, pretty neato, but it's essentially rendering everything again in real time, hence enemy characters might actually be different than when you played.

If you're running mods that have higher resolution textures and higher poly models, at 8x speed, it's having to flush and redraw a lot more than, say 1x speed. 

 

For example, you're in iron dragon, at the very start you can't see the caves, so the rendering API doesn't draw the caves and the engine has little need to commit the textures or models to VRAM, because you simply can't see it.

However, as you get closer, to the point where you can see it, the rendering API makes a draw call, the engine commits vertex indices and textures to VRAM and the caves are drawn on your screen. At 1x speed, the engine and the rendering API really don't need to change that much to draw a new frame, or load too much into VRAM. you might see a new character or a piece of scenery, however that's a few new draw calls at most. 

Now, at 8x speed, the difference between frames might be huge, you might go from the start of the mission all the way up to the caves in just a few seconds, which means the engine is having to work much harder to quickly call the rendering API to 'draw that character in this frame' then commit the necessary textures and model files to RAM, instead of just a few times per second, it's now having to potentially do this dozens, if not hundreds of times per second.

Add non-billboard, higher resolution textures and higher poly models to the mix, the game may struggle to flush these out when not needed or appropriately allocate memory for each model and texture file. 

While your computer may be able to handle it, the engine will have an upper limit where performance begins to diminish, Ike was optimised for fewer lower resolution models and textures. And I think it usually begins to see a performance drop at 20,000 vertices per frame, despite having an index buffer which would allow for 65535 vertices per model.

I'm sure someone will come over and debunk this theory as GR has many little niggles I'm still wrapping my head around after many, many years, it could be that simply the mods don't play too nice with replay files. 

 

[gameragodzilla]

30 minutes ago, Zeealex said:

I wonder if it could be an engine/rendering API issue. Where it's having to make so many new draw calls per frame that the engine simply can't keep up. 

Its a theory, but it would make sense given the age of the game.

So, iirc replays in GR work super weird. It's not technically a video replay but actually just a record of every move every character took and every shot that was made and what the effect of that shot was being replayed through the mission file, pretty neato, but it's essentially rendering everything again in real time, hence enemy characters might actually be different than when you played.

If you're running mods that have higher resolution textures and higher poly models, at 8x speed, it's having to flush and redraw a lot more than, say 1x speed. 

 

For example, you're in iron dragon, at the very start you can't see the caves, so the rendering API doesn't draw the caves and the engine has little need to commit the textures or models to VRAM, because you simply can't see it.

However, as you get closer, to the point where you can see it, the rendering API makes a draw call, the engine commits vertex indices and textures to VRAM and the caves are drawn on your screen. At 1x speed, the engine and the rendering API really don't need to change that much to draw a new frame, or load too much into VRAM. you might see a new character or a piece of scenery, however that's a few new draw calls at most. 

Now, at 8x speed, the difference between frames might be huge, you might go from the start of the mission all the way up to the caves in just a few seconds, which means the engine is having to work much harder to quickly call the rendering API to 'draw that character in this frame' then commit the necessary textures and model files to RAM, instead of just a few times per second, it's now having to potentially do this dozens, if not hundreds of times per second.

Add non-billboard, higher resolution textures and higher poly models to the mix, the game may struggle to flush these out when not needed or appropriately allocate memory for each model and texture file. 

While your computer may be able to handle it, the engine will have an upper limit where performance begins to diminish, Ike was optimised for fewer lower resolution models and textures. And I think it usually begins to see a performance drop at 20,000 vertices per frame, despite having an index buffer which would allow for 65535 vertices per model.

I'm sure someone will come over and debunk this theory as GR has many little niggles I'm still wrapping my head around after many, many years, it could be that simply the mods don't play too nice with replay files. 

 

Interesting, though the odd part for me is that the previous replays using the exact same mods on the exact same level never had this issue. Maybe I’m getting better at the game and therefore running forward faster and shooting more things quickly?

Specifically, it’s Fever Claw that’s giving me problems. Didn’t have an issue with my previous full recording of the level but I saved a new one that I thought was better and it gives constant crashes. There are fairly specific areas where it crashes, though whether it does crash or not is totally random. And sometimes I am able to get through the entire level. I am using the Rambo mod and an absurdly high res SPAS-12, but again, I never had this issue with the prior Replay. Maybe it was right on the cusp of crashing before and now that I’m better at the game, rushing ahead more and blasting enemies quicker, maybe it’s too much. I dunno.

I wonder if there’s a way to mod the engine to increase this limit. I played around with compatibility settings, Nvidia profile inspector, etc. to no avail.

EDIT: I tried testing the vertices theory by replacing my custom SPAS-12 model with a much lower poly version from another mod and it still freezes, and at the same area no less.

Edited April 18, 2022 by gameragodzilla

[Zeealex]

1 hour ago, gameragodzilla said:

I wonder if there’s a way to mod the engine to increase this limit. I played around with compatibility settings, Nvidia profile inspector, etc. to no avail.

Unfortunately not, unlike more modular games such as Arma 3 it is not possible to change the memory allocator for the Ike engine. There are injectors such as enbseries which can inject some functionality of DirectX 9 however this does not do much except add bloom and other global screen effects (at one point it did fix the game for Windows 10). The rest of the rendering goodness such as the new optimizations 9c offered is limited by how the engine uses it, as Ghost Recon was designed and built around DirectX 8.1 it cannot take advantage of the newer API functions. 

You would essentially need to have the source code, a really good reverse engineer, or rebuild the game from the ground up in order to have it manage memory in a more modern manner.

Process injection can only go so far, and in doing so you would be opening the game up for further instability as you could inject data into the wrong memory address at any time.

[gameragodzilla]

17 hours ago, Zeealex said:

Unfortunately not, unlike more modular games such as Arma 3 it is not possible to change the memory allocator for the Ike engine. There are injectors such as enbseries which can inject some functionality of DirectX 9 however this does not do much except add bloom and other global screen effects (at one point it did fix the game for Windows 10). The rest of the rendering goodness such as the new optimizations 9c offered is limited by how the engine uses it, as Ghost Recon was designed and built around DirectX 8.1 it cannot take advantage of the newer API functions. 

You would essentially need to have the source code, a really good reverse engineer, or rebuild the game from the ground up in order to have it manage memory in a more modern manner.

Process injection can only go so far, and in doing so you would be opening the game up for further instability as you could inject data into the wrong memory address at any time.

I see. In that case, are there any ways to get around the limits? Like I said, I tried using a different SPAS-12 model, tried limiting the frame rate and even CPU cores, tried fiddling around with compatibility settings and Nvidia profile, is there anything I can do here?

EDIT: I don't know if this means anything, but these are the things I found in ike.log and ikecrash.log

ike.log:

***** User's system configuration ***** 
      CPU: AuthenticAMD 3900 MHz Recent AMD
      RAM: 2048 MB
      O/S: Microsoft Windows NT Family 6.2  build 9200
Ghost Recon (RELEASE)  version = 1.4.0.0(0)
RSDisplayMgr::CreateGameWindow: Creating a 1920x1080 window at 0,0 on monitor Generic PnP Monitor attached to NVIDIA GeForce RTX 2080 SUPER
RSDisplayMgr::CreateGameWindow: Game running on primary display Generic PnP Monitor attached to NVIDIA GeForce RTX 2080 SUPER
HAL (hw vp): NVIDIA GeForce RTX 2080 SUPER X8R8G8B8
RSFILEDATA: FindFile could not find file ike_fx_fire_type2.tga
RSFILEDATA: FindFile could not find file ike_fx_fire_type2.rsb
Unable to find ike_fx_fire_type2.tga anywhere.
RSFILEDATA: FindFile could not find file lighthalo_effect3.rsb
RSFILEDATA: FindFile could not find file lighthalo_effect3.rsb
Unable to find lighthalo_effect3.rsb anywhere.
RSSoundMgrPC: Selecting sound device Speakers (High Definition Audio Device)
   driver is {0.0.0.00000000}.{348a0099-3a87-4065-96aa-76e804bef8a2}
Media Player: Unable to Connect Video Pins Unknown
RSSoundMgrPC: Selecting sound device Speakers (High Definition Audio Device)
   driver is {0.0.0.00000000}.{348a0099-3a87-4065-96aa-76e804bef8a2}
 

ikecrash.log:

GhostRecon caused an Access Violation in module GhostRecon.exe at 0023:0051c320.
Exception handler called in Main Thread.
Error occurred at 4/18/2022 01:42:28.
D:\GOG Games\Tom Clancys Ghost Recon\GhostRecon.exe, run by georg.
16 processor(s), type 586.
2048 MBytes physical memory.
Read from location 00000008 caused an access violation.

Registers:
EAX=00000000 CS=0023 EIP=0051c320 EFLGS=00010246
EBX=036daebc SS=002b ESP=0019fd08 EBP=000000a8
ECX=00000000 DS=002b ESI=03a09468 FS=0053
EDX=00854750 ES=002b EDI=00000004 GS=002b
Bytes at CS:EIP:
8b 50 08 8d 48 08 ff 52 04 8a 86 f0 03 00 00 84 

Stack trace:
0051C320: GhostRecon.exe! <no symbol>, <no file>, <no line>

    Module list: names, addresses, sizes, time stamps and file times:
D:\GOG Games\Tom Clancys Ghost Recon\GhostRecon.exe, loaded at 0x00400000 - 6402084 bytes - 73655965 - file date is 4/21/2019 18:39:36
C:\WINDOWS\system32\nvspcap.dll, loaded at 0x072a0000 - 2201800 bytes - 61f3d491 - file date is 1/28/2022 06:50:32
C:\Program Files (x86)\RivaTuner Statistics Server\RTSSHooks.dll, loaded at 0x10000000 - 835128 bytes - 61aa1c68 - file date is 12/3/2021 08:34:16
C:\WINDOWS\SYSTEM32\winmmbase.dll, loaded at 0x50000000 - 116240 bytes - 1b544b3e - file date is 6/5/2021 07:05:46
C:\WINDOWS\SYSTEM32\CoreUIComponents.dll, loaded at 0x51610000 - 2708560 bytes - 1bfcc075 - file date is 11/4/2021 09:01:04
C:\WINDOWS\SYSTEM32\WINSPOOL.DRV, loaded at 0x51fa0000 - 456192 bytes - 05037899 - file date is 1/9/2022 21:18:32
C:\WINDOWS\SYSTEM32\textinputframework.dll, loaded at 0x5e410000 - 916792 bytes - efd444de - file date is 11/4/2021 09:57:00
C:\WINDOWS\SYSTEM32\CoreMessaging.dll, loaded at 0x62510000 - 824848 bytes - 8f8fadfc - file date is 11/4/2021 09:01:08
C:\WINDOWS\SYSTEM32\inputhost.dll, loaded at 0x625e0000 - 1243256 bytes - cab22d32 - file date is 11/4/2021 09:01:06
C:\WINDOWS\SYSTEM32\DINPUT8.dll, loaded at 0x62710000 - 174080 bytes - e9c598fc - file date is 6/5/2021 07:06:22
C:\WINDOWS\SYSTEM32\drvstore.dll, loaded at 0x62b60000 - 1059344 bytes - 1ccb5b07 - file date is 11/4/2021 09:01:10
C:\WINDOWS\SYSTEM32\dxcore.dll, loaded at 0x63d80000 - 172520 bytes - 4561672a - file date is 6/5/2021 07:05:50
C:\WINDOWS\SYSTEM32\AcLayers.DLL, loaded at 0x64ec0000 - 406528 bytes - 7bf03655 - file date is 11/4/2021 09:01:18
C:\WINDOWS\SYSTEM32\iertutil.dll, loaded at 0x657b0000 - 2284520 bytes - 92ec4170 - file date is 4/4/2022 23:01:32
C:\WINDOWS\SYSTEM32\urlmon.dll, loaded at 0x659e0000 - 1657856 bytes - 72a7626f - file date is 2/17/2022 11:55:38
C:\WINDOWS\SYSTEM32\srvcli.dll, loaded at 0x67010000 - 79528 bytes - 388a208b - file date is 4/12/2022 22:15:12
C:\WINDOWS\SYSTEM32\wldp.dll, loaded at 0x681a0000 - 221528 bytes - 137d281c - file date is 11/4/2021 09:01:08
C:\WINDOWS\SYSTEM32\cryptnet.dll, loaded at 0x68200000 - 147752 bytes - e66e3166 - file date is 6/5/2021 07:05:54
C:\WINDOWS\system32\rsaenh.dll, loaded at 0x6ddb0000 - 186960 bytes - 1851a46e - file date is 11/4/2021 09:57:02
C:\WINDOWS\SYSTEM32\CRYPTSP.dll, loaded at 0x6dde0000 - 69232 bytes - 7efe3540 - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\profapi.dll, loaded at 0x6de00000 - 89368 bytes - 2bf86304 - file date is 6/5/2021 07:05:46
C:\WINDOWS\SYSTEM32\UMPDC.dll, loaded at 0x6de20000 - 50088 bytes - fe870edd - file date is 6/5/2021 07:05:52
C:\WINDOWS\SYSTEM32\devobj.dll, loaded at 0x6de50000 - 141000 bytes - 373790ac - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\cfgmgr32.dll, loaded at 0x6de80000 - 237568 bytes - 6371ef08 - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\ntmarta.dll, loaded at 0x6df90000 - 155464 bytes - 8ee36d91 - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\CRYPTBASE.DLL, loaded at 0x6dfc0000 - 32088 bytes - f24f0968 - file date is 6/5/2021 07:05:46
C:\WINDOWS\SYSTEM32\kernel.appcore.dll, loaded at 0x6e790000 - 68160 bytes - f69c7055 - file date is 11/4/2021 09:01:08
C:\WINDOWS\SYSTEM32\wintypes.dll, loaded at 0x6e7b0000 - 959632 bytes - f93e2d16 - file date is 2/17/2022 11:55:36
C:\WINDOWS\SYSTEM32\windows.storage.dll, loaded at 0x6e8a0000 - 6994968 bytes - 4ce232bd - file date is 4/12/2022 22:15:10
C:\WINDOWS\SYSTEM32\msasn1.dll, loaded at 0x6ef50000 - 51120 bytes - 346de8d3 - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\samcli.dll, loaded at 0x6ef60000 - 72192 bytes - 4cd9a172 - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\SspiCli.dll, loaded at 0x6efc0000 - 141536 bytes - c9b32fb1 - file date is 3/8/2022 13:04:50
C:\WINDOWS\SYSTEM32\netutils.dll, loaded at 0x6f040000 - 39256 bytes - 985c2155 - file date is 1/11/2022 13:09:48
D:\GOG Games\Tom Clancys Ghost Recon\DBGHELP.dll, loaded at 0x72a00000 - 163088 bytes - 3af32067 - file date is 4/21/2019 18:42:52
C:\WINDOWS\SYSTEM32\bcrypt.dll, loaded at 0x73a40000 - 97584 bytes - 09b341ac - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\UxTheme.dll, loaded at 0x73a60000 - 513536 bytes - 931f24d6 - file date is 11/4/2021 09:01:04
C:\WINDOWS\SYSTEM32\dwmapi.dll, loaded at 0x73af0000 - 138424 bytes - 314ccf01 - file date is 11/4/2021 09:01:08
C:\WINDOWS\SYSTEM32\MPR.dll, loaded at 0x73bd0000 - 97632 bytes - f208a759 - file date is 2/17/2022 11:55:36
C:\WINDOWS\SYSTEM32\powrprof.dll, loaded at 0x73d30000 - 272736 bytes - f91c170c - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\VERSION.dll, loaded at 0x73d80000 - 27848 bytes - a1b61f3e - file date is 6/5/2021 07:06:02
C:\WINDOWS\SYSTEM32\WINMM.dll, loaded at 0x73dc0000 - 185456 bytes - 3b22fd01 - file date is 6/5/2021 07:05:46
C:\WINDOWS\SYSTEM32\AcGenral.DLL, loaded at 0x74920000 - 2487976 bytes - 0a8709b6 - file date is 4/4/2022 23:01:40
C:\WINDOWS\SYSTEM32\USERENV.dll, loaded at 0x75320000 - 122896 bytes - 4b06e4bc - file date is 6/5/2021 07:05:54
C:\WINDOWS\SYSTEM32\WSOCK32.dll, loaded at 0x75340000 - 16896 bytes - 3ade28c4 - file date is 6/5/2021 07:05:50
C:\WINDOWS\SYSTEM32\apphelp.dll, loaded at 0x759d0000 - 653344 bytes - 1ece0f7a - file date is 11/4/2021 09:57:02
C:\WINDOWS\System32\msvcrt.dll, loaded at 0x75ad0000 - 787136 bytes - 4078cfe5 - file date is 6/5/2021 07:05:46
C:\WINDOWS\System32\USER32.dll, loaded at 0x75ba0000 - 1755064 bytes - 51a17bf5 - file date is 4/4/2022 23:01:32
C:\WINDOWS\System32\RPCRT4.dll, loaded at 0x75d50000 - 768536 bytes - 820ca38f - file date is 4/12/2022 22:15:10
C:\WINDOWS\System32\OLEAUT32.dll, loaded at 0x75e20000 - 629632 bytes - 821a6441 - file date is 6/5/2021 07:05:54
C:\WINDOWS\System32\SETUPAPI.dll, loaded at 0x75ec0000 - 4468256 bytes - 878bd64d - file date is 1/26/2022 18:17:22
C:\WINDOWS\System32\CRYPT32.dll, loaded at 0x76300000 - 1030400 bytes - 44e34f98 - file date is 1/9/2022 21:18:36
C:\WINDOWS\System32\SHLWAPI.dll, loaded at 0x76630000 - 300064 bytes - 47e73a8a - file date is 6/5/2021 07:06:02
C:\WINDOWS\System32\gdi32full.dll, loaded at 0x76690000 - 909512 bytes - dac47a21 - file date is 4/12/2022 22:15:12
C:\WINDOWS\System32\GDI32.dll, loaded at 0x76770000 - 130672 bytes - 7e972d45 - file date is 6/5/2021 07:05:50
C:\WINDOWS\System32\ADVAPI32.dll, loaded at 0x768c0000 - 500984 bytes - dd0096c9 - file date is 1/11/2022 13:09:46
C:\WINDOWS\System32\SHCORE.DLL, loaded at 0x76940000 - 785104 bytes - 20bf6f8f - file date is 4/12/2022 22:15:10
C:\WINDOWS\System32\imagehlp.dll, loaded at 0x76a10000 - 99112 bytes - 94cde7cb - file date is 6/5/2021 07:05:26
C:\WINDOWS\System32\sechost.dll, loaded at 0x76a90000 - 494824 bytes - c2e9401b - file date is 3/8/2022 13:04:50
C:\WINDOWS\System32\combase.dll, loaded at 0x76b10000 - 2679544 bytes - a296a824 - file date is 2/17/2022 11:55:36
C:\WINDOWS\System32\WINTRUST.DLL, loaded at 0x76da0000 - 330000 bytes - 507bdfa0 - file date is 2/17/2022 11:55:36
C:\WINDOWS\System32\SHELL32.dll, loaded at 0x76e00000 - 6332768 bytes - 45062c92 - file date is 4/4/2022 23:01:34
C:\WINDOWS\System32\KERNEL32.DLL, loaded at 0x77410000 - 646008 bytes - 7fcb6c15 - file date is 4/12/2022 22:15:10
C:\WINDOWS\System32\win32u.dll, loaded at 0x77500000 - 101752 bytes - 8a05e54c - file date is 11/4/2021 09:01:08
C:\WINDOWS\System32\msvcp_win.dll, loaded at 0x77520000 - 495320 bytes - 84039c2e - file date is 6/5/2021 07:05:54
C:\WINDOWS\System32\KERNELBASE.dll, loaded at 0x775a0000 - 2440448 bytes - b8da8cf0 - file date is 4/12/2022 22:15:10
C:\WINDOWS\System32\ole32.dll, loaded at 0x77800000 - 1363024 bytes - a89da16b - file date is 11/4/2021 09:01:08
C:\WINDOWS\System32\ucrtbase.dll, loaded at 0x77950000 - 1127472 bytes - cb34958a - file date is 6/5/2021 07:05:54
C:\WINDOWS\System32\bcryptPrimitives.dll, loaded at 0x77a70000 - 405952 bytes - 0cfdd9d2 - file date is 1/9/2022 21:18:26
C:\WINDOWS\System32\IMM32.DLL, loaded at 0x77ae0000 - 145640 bytes - 00a3f851 - file date is 6/5/2021 07:05:54
C:\WINDOWS\System32\MSCTF.dll, loaded at 0x77b70000 - 886296 bytes - d3c78d6e - file date is 2/17/2022 11:55:34
C:\WINDOWS\System32\WS2_32.dll, loaded at 0x77c50000 - 407520 bytes - 0da3b025 - file date is 6/5/2021 07:05:46
C:\WINDOWS\SYSTEM32\ntdll.dll, loaded at 0x77d40000 - 1722104 bytes - 4ae92803 - file date is 2/17/2022 11:55:34
C:\WINDOWS\SYSTEM32\DSOUND.dll, loaded at 0x79c90000 - 444416 bytes - 21d1b3c6 - file date is 6/5/2021 07:05:44
C:\WINDOWS\SYSTEM32\ResampleDmo.DLL, loaded at 0x79d10000 - 625456 bytes - 303609ff - file date is 6/5/2021 08:17:10
C:\WINDOWS\SYSTEM32\msdmo.dll, loaded at 0x79db0000 - 31992 bytes - e73b72f1 - file date is 6/5/2021 07:05:46
D:\GOG Games\Tom Clancys Ghost Recon\d3d8.dll, loaded at 0x7c140000 - 378368 bytes - 618667fe - file date is 11/6/2021 06:43:04
C:\WINDOWS\SYSTEM32\MSACM32.dll, loaded at 0x7c600000 - 98136 bytes - 1d9cc909 - file date is 6/5/2021 07:05:46
 

Is the memory violation the reason behind this? Granted, the ikecrash.log didn't update anytime the game froze and I see the d3d8.dll file I attempted to use in game from dgVoodoo2, which didn't work. But I wonder if there's anything else here that anyone here can look at. I still don't know why this level in particular causes this problem when the previous Replay of Fever Claw never had this problem.

Edited April 19, 2022 by gameragodzilla

[AlexKimov]

There's nothing to do. As @Zeealex said some programming magic nedded, it can be software (game code, dx api, drivers, windows, any process basically) and/or hardware problem. Try everything, maybe it'll help you .

"Read from location 00000008 caused an access violation." means "GR officially crushed", you'll never know what happened without debugger, profiler and other dev tools.

[gameragodzilla]

10 hours ago, AlexKimov said:

There's nothing to do. As @Zeealex said some programming magic nedded, it can be software (game code, dx api, drivers, windows, any process basically) and/or hardware problem. Try everything, maybe it'll help you .

"Read from location 00000008 caused an access violation." means "GR officially crushed", you'll never know what happened without debugger, profiler and other dev tools.

Blegh, well I was hoping someone could've told me why. Could it possibly be some Windows 11 update that changed certain memory parameters to be restricted, and therefore the game being older created an access violation? Or is this standard for all freezes in Ghost Recon since 2001?

[gameragodzilla]

Wanted to post another dedicated reply because I think I figured out the reason why it happens, though I have no idea why that thing happens or how to fix it.

I noticed that when in replay, the tank column in Fever Claw jitters on the Command Map at certain points. I'm not even sure if the tank jittering is in the world itself as well or just the Command Map, as many times the tanks aren't directly on my screen in gameplay. However, it jitters particularly badly at 8x speed, and the times where it does freeze, it seems like the jittering was enough to cause the tank to clip slightly through the walls of the game world, which may be causing the freezing. There's still some jittering in 4x speed, but nowhere near as much, to the point where I suspect it never gets enough to actually clip out of the game world.

So the question is, does anyone know why this jittering would happen on a Replay? And is there anything I can tweak to make it stop happening? This would explain why other levels seem to not have issues, as this level is dedicated towards escorting a tank column through a blown out city.