Author Topic: Dual Core and SLI troubles (Read 667 times)

Vasco · « **on:** June 04, 2007, 04:21:21 AM »

Hi there,

I'm having the following issues with the latest version of the game:

1. Game lockup when entering the arena or the cockpit, if the AH process isn't tied to a single CPU core. I have to limit the affinity using the Process Explorer to any of both cores after starting the game. Windows runs with /pmtimer switch, so RDTSC isn't used as the timebase but the ACPI power management timer is. I suspect this is a pure in-game multithreading issue, rather than a time-synchronization effect. Definitively reproducible.

2. Game lockup when setting FPS to 60 (TFT rate is 60Hz too) and VSYNC on. Works flawlessly with FPS set to 45 or 0 (Unlimited). Simply doesn't work when set to 60. Maybe SLI-related. Definitively reproducible.

3. The animated page of the Clipboard flatters around the clipboard in any angle. This may be SLI-related, since VSYNC is off and SLI mode is set to Alternate Frame Rendering. The buttons and sliders are at the correct position and work (if you can hit them), the animated page flatters around either continously (during game start) or is stuck in any (mostly unreadable) angle. So E6B or roster or map details aren't really readable. I may provide a screenshot. I suspect this is a VSYNC and/or SLI-AFR issue. Definitively reproducible.

4. Completely f'upped Viewpoint the first time I enter the cockpit with the head direction moving randomly in any direction. The heading of the plane in the map is synched to it. Happens almost after every game start (definitively reproducible). This effect can be circumvented by calling the video settings and just pressing OK.

System: AMD Athlon64 X2 4600+ (2.4GHz Dual Core, 512K L2-Cache)
Video: 2x NVIDIA 7900GT 256M SLI
Drivers: 93.71 WHQL or latest beta
Audio: Creative SB Audigy2 ZS
OS: Windows XP SP 2 (German)

I have lots of developer stuff installed so I may provide further infos.
The NVIDIA drivers have a hard-coded profile for AH restricting SLI to AFR which isn't changeable by the user. Older drivers hadn't had that profile and ran with split-screen rendering.

Cheers,
Vasco

Rolex · « **Reply #1 on:** June 04, 2007, 04:53:10 AM »

Hi Wuerger,

I recommend you check this thread which contains a link for an AMD patch:

http://forums.hitechcreations.com/forums/showthread.php?s=&threadid=181720

Vasco · « **Reply #2 on:** June 04, 2007, 05:00:19 AM »

I have the AMD processor driver installed (actually to have CnQ working).

AMD's "Dual Core Optimizer" does nothing more than add the /pmtimer switch to the boot.ini to move from TSC-based timing to the ACPI PM-timer (which is stable under different cpu clockspeeds - in contrast to the TSC which may get out of synch on both cores).

Cheers,
Vasco

Vasco · « **Reply #3 on:** June 04, 2007, 05:07:40 AM »

Well i had AMD's DC Optimizer installed a while ago but deinstalled it for not having any noticeable effect on the apps I ran back then.

I'll give it a try again this evening.

Maybe it doesn't just add the /pmtimer switch but also syncs the TSCs somehow. If AH uses RDTSC direclty it should better switch to QueryPerformanceCounter() anways...

Cheers,
Vasco

Kev367th · « **Reply #4 on:** June 07, 2007, 06:47:17 PM »

For XP and AH2 you HAVE to have the Dual Core Optimizer installed also.

Vasco · « **Reply #5 on:** June 08, 2007, 03:02:21 AM »

This DC Optimizer is just a workaround for old apps (which use the RDTSC directly and fail on Athlon X2s [with CnQ]) who don't receive sw updates anymore.

Proper written apps (read as "contemporary") shouldn't be affected by TSC drift. There's even a note in the directx sdk regarding TSC issues and how to circumvent it.

Cheers,
Vasco

Krusty · « **Reply #6 on:** June 08, 2007, 11:43:51 AM »

The problem lies not in the application code. The problem lies in either the chip or the chip driver.

Otherwise it would happen for ALL dual core chips. It only happens with some AMD chips. Thus it is not a software problem. You need the patch to fix/correct/bypass/whatever the problem. Just because you didn't notice anything different doesn't mean it wasn't working

I say reinstall it and leave it there. You need it. It helps. Thus leave it installed (my thought logic on the matter)

Vasco · « **Reply #7 on:** June 08, 2007, 12:19:49 PM »

it applies to AMD DC's only.. AFAIK Intel DCs keep their TSCs in sync..

Anyways. App coders are discouraged from hitting the hw directly to get timing infos... And I suspect the problem there...

Nevermind, I can read the E6B upside down too... elsewise I just guess...

Cheers,
Vasco

Krusty · « **Reply #8 on:** June 08, 2007, 12:23:55 PM »

#3 is similar to a dual-core AMD problem. Others might be SLI. Can you disable one of the cards (run in compatibility mode)? Yank one from the case for a moment, if you cannot, and then try running AH. If ALL of your problems persist, it's all about SLI. [EDIT: I mean yank/disable one as a trouble shooting step, you can re-enable or re-insert it after you run the test, it goes without saying.]

I suspect, however, that #3 will persist, as it's similar to the description of the "spinning clipboard" multi-threading issue.

Kev367th · « **Reply #9 on:** June 11, 2007, 11:23:34 PM »

Quote

Originally posted by Krusty
#3 is similar to a dual-core AMD problem. Others might be SLI. Can you disable one of the cards (run in compatibility mode)? Yank one from the case for a moment, if you cannot, and then try running AH. If ALL of your problems persist, it's all about SLI. [EDIT: I mean yank/disable one as a trouble shooting step, you can re-enable or re-insert it after you run the test, it goes without saying.]

I suspect, however, that #3 will persist, as it's similar to the description of the "spinning clipboard" multi-threading issue.

I just re-installed EVERYTHING recently.

Added both the dual core optimiser fix and the dual core driver, never had a spinning clipboard.

Try this see if it fixes your lockups -
On the video options in AH disable hardware vertex processing, and both optimisation options.

If it doesn't lock-up anymore -
The fix for an AGP card was to increase the PCI latency to 128 or 256, but I have no idea how to set it on PCI-E.
Once you had done that you could re-enable everything and play without lockups.

Skuzzy · « **Reply #10 on:** June 12, 2007, 06:45:01 AM »

It does not matter how you read the timer. It has to do with being multi-threaded and using a high resolution clock. On any successive read of the high frequency counter you cannot be gauranteed the results will be in proper sequence. And when you are multi-threaded, this can cause a lot of bizarre issues.

The software patch from AMD corrects the problem by inserting another layer between the CPU and the OS which hides the out of sync issue.

If we were reading it directly, no layer of software, short of a protected mode trap, could hide the problem.