Jump to content

Building a stable NF4 and Geforce System


Recommended Posts

How to build a Stable NF4 System

Verify that your Video Card might the problem with stability

 

If you follow all the steps below and in the end come out with a system that still crashes when playing 3D games, the problem is definitely your video card. This information is built from months of testing all tracked in the following thread:

 

http://www.xfxforce.com/forums/thread.jspa...tart=0&tstart=0

 

The information is based mostly on NF4 AMD Athlon systems, but in testing has been found to apply equally to Pentium 4, 4EE, M and D systems as well. No Information is yet available for Core Solo or Duo systems, if this information becomes available

I will update the article. It is also primarily concerned with NVIDIA Geforce Series 6/7 video cards. (ie 6800,6600,7800GT,7800GTX)

 

See below, how to get in touch with me to report your own testing results, and help me keep the information in this up to date and relevant, as well as how to always find the most current version. I try to surf the forums on a regular basis but contacting me directly is a much better way to be sure I get the information.

 

This article assumes an intermediate to advanced knowledge of and comfort in working with PC systems.

 

System Construction:

 

1. When choosing a power supply for the system (or if upgrading system) use a PSU which is SLI approved even if you are only building a single card system.

 

2. Mount all basic hardware, enough to make an initial system boot, Mainboard, CPU/Heatsink, One Stick of Ram (or two if your system requires at least two), CD/DVD drive, Hard Drive, and Video Card.

 

3. Acquire memtest86+ from: http://www.memtest86.com/ and create the Boot CD.

Boot up the system and run the Memtest86 tests looping at least overnight. If this happens with no failures then the ram and the system are stable. Add additional ram one

stick (or set of sticks as required by your system) at a time, testing for 2-3 hours after each addition. After the system ram is fully installed run again overnight. If any of the tests fail you have suspect ram or potentially mainboard.

Swap ram around until you have runs with no failures. Note which sticks were installed when failures occured, this is likely bad ram RMA it.

 

Note: It is assumed here that you have properly configured the ram settings in the PC Bios. For NF4 systems, make sure the timings match the MFG listed ram timings, the Command rate is usually "1T" for two sticks, and "2T" for four sticks. This article also assumes NO Over-clocking. (Other than MFG stock over-clocking often used in many geforce cards) Feel free to use it as a guide to test an over clocked system but your results are strictly YMMV.

 

If the failures (either a failure listed in memtest86 or system locks/crashes/freezes) occur with all Ram sticks then:

 

a. First try changing USB support from 2.0+1.1 to just 1.1

b. If that didn't work try turning off USB entirely.

c. if it still doesn't work then the mainboard is suspect, RMA it.

 

Note: On some NF4 boards problems with the USB 2.0 support have been noted, steps a,b,c help to figure out if yours is one of those boards. (Abit, Gigabyte, and MSI models have been noted to have this issue). It should also be noted that often the problem does not manifest until an OS is installed so testing at this stage is not conclusive of the problem.

 

 

4. Now that we have a known stable base system set make sure the USB support in the BIOS is set to 1.1 only. See note above for mode information on why we are doing this. Later in the build/test process we will set it back. Install an OS, XP Pro or Home most likely. There is nothing special of note during the main OS install it should hopefully proceed smoothly.

 

5. Now that the OS is installed, proceed with installing additional drivers.

 

a. Start with the Nvidia Platform drivers. Install only those required for your hardware. (Current recommended version 6.85, its labeled as x16 drivers, but that’s a mislabel as what it really represents is that the driver set has special support for the newer Dual x16 SLI boards. However it IS the current driver set for ALL NF4 based boards.)

 

- Yes install the IDE driver, it no longer causes any problems and SHOULD be installed as your PATA and SATA drives will work better with the NVIDIA IDE driver installed.

- No don't install the Network Access Manager. Sorry we still cannot recommend it, it still has issues in the testing that has been done.

 

b. Install the AMD CPU (Cool 'n Quiet) drivers. They support more than just the Cool and Quiet functionality and are required for the proper functioning of an AMD CPU based machine. If you have an Intel based machine the same goes for any additional Intel Platform/CPU drivers.

 

c. Install any other drivers for installed equipment; do not install the Video Card Drivers Yet!

 

d. Leave the system powered on and let it sit idle for at least 12 hours, see if anything goes wrong.

-If so go directly to step 6.

e. Change the USB support to 2.0+1.1, again let the system sit 12 hours or so see if anything goes wrong.

-If so go directly to step 6.

 

Note on Driver changes during testing:

 

When and/or if you change driver versions during testing it is vital that you completely remove the old drivers before installing new ones.

 

Recommended driver changing procedure

 

- Use Add/Remove programs to access the Nvidia Drivers Add/change/Remove functionality

 

- If you are changing the platform drivers uninstall everything including the Video Driver if you have reached that testing stage.

 

- If you are only changing the video driver uninstall only the video driver.

 

- After doing the driver uninstalls do not reboot yet. Run a driver cleaner program (Drivercleaner pro is recommended and was used for

all testing in this article) and set it to clean up whatever drivers you have removed (Driver cleaner pro has separate filters for the

Platform drivers and the Video drivers use either or both which ever is proper, for whatever drivers you have removed)

 

- After the system reboots install new drivers, if you are installing platform drivers it is recommended you do the platform drivers install,

reboot and then install the Video drivers.

 

- When installing the video drivers, install the drivers, reboot, then enable SLI if you are setting up an SLI system, reboot.

 

 

6. What to do if anything goes wrong. Problems might have started occuring at this point, here are listed some things to try to re-stabilize the system. Hopefully they will do the trick; this section should also be referred back to later in the build test process if anything goes wrong. If you have not had anything go wrong yet, freezes, crashes, lockups, glitches, strange error messages, etc, this step can probably be skipped for now. After making any of the changes/fixes detailed here repeat all testing steps that have previously been applied until you are past the point the error occurred, 3 times.

It has been observed that often on the first retest the error appears solved, and it sometimes takes 2-3 reruns of the testing before it re-occurs.

 

a. Make sure that your mainboard BIOS is up to date if it is not download and install the latest bios from the MFG's web site.

 

b. Download and install the Microsoft dual core hotfix if you have a dual core cpu.

 

Information on the Hotfix:

 

Info:

 

http://support.microsoft.com/Default.aspx?id=896256

 

The Hostfix itself:

 

http://www.sendmefile.com/00157879

 

* Make sure to add the /usepmtimer switch to boot.ini the instructions are not fully clear on adding this switch, it does appear to be required for proper functionality.

 

* The article on the hotfix mentions a registry change. Unfortunately this is an issue that seems to go either way. Some systems have needed it, others have not.

Most HAVE. Its going to be a judgment call, and it might have to be made further along in the process of testing. If after you start testing, with 3DMark and/or various games problems with stuttering of the video or sound start occurring you might need to make the registry change, just be aware of it.

 

* The Article on the Hotfix has a linked article about sound problems, on XP systems it’s unlikely you will have to do anything based on this article, but you should familiarize yourself with it in just incase.

 

7. If you have reached this point then do one final test before installing the video drivers. Download and install the Prime95 program from:

http://www.mersenne.org/freesoft.htm

 

Prime95's real world purpose is in the interest of science, read through the page if you are of that bent in life it might be interesting.

However our interest is in system testing. Prime95 has a torture test mode which makes an excellent system burn in test.

Start up a torture test and walk away for 12 hours or more. The longer you wait the better the burn in.

 

Any failures, diagnose (repeat step 5,6 instructions), Fix or RMA as needed.

Hopefully (and usually 90% or the time) nothing unusual pops up at this point and you can move on.

 

8. Finally, we know you have been waiting for this, for maybe 2 to 3 days during all this testing and waiting. Install the Nforce Video Drivers.

 

If all has gone well to this point if you have even reached this point any problems which occur are 99.99% chance of being an actual video card problem.

Current release version at the time of writing this is: 81.98 and recommended beta version is 83.40. Neither of these has an serious observed problems in the testing process documented. A lockup/crash/glitch during testing after following any problem correction steps points to a Video card hardware issue.

 

Please note at this stage that this process is known to work for the programs detailed, other programs might have configurations and settings of their own, which might need to be set correctly before doing any testing with them. Testing with versions of the software detailed other than the listed version may give different or incomplete results. Testing with different configurations than those listed (All testing done with Program defaults actually, so if you modify the defaults....) might give you different and/or incomplete results.

 

Test Software suite at this stage.

 

3DMark03 Registered Pro Edition

3DMark05 Registered Pro Edition

3DMark06 Registered Pro Edition

Doom 3

HalfLife 2

 

If you don't have one or the other of these programs either skip that step or substitute a similar program. In so far as the 3Dmarks are concerned the demo versions will sometimes trip the problems, however it has been observed that the registered pro editions are far more likely to trip the problems on a consistent reliable basis.

The Pro editions enable extra tests so the loops last longer and are more intense to the hardware. The 3Dmark tests usually find problems in 10-15 minutes that can take hours to occur in actual games in some cases.

 

a. Install the Video Card Drivers, 81.98 and 83.40 are recommended versions that are known good for these tests. For the sake of completeness 77.71, 77.77, 78.01, 81.95,

82.12, 82.65, 82.94, 83.08, 83.10, 83.20, and 83.60 were also used in a rather extensive testing sweep within the last two days. All of these versions also tested fine; however these versions are either non-current or had secondary issues (...don't support SLI, or do not have setup.exe installers).

 

If you are building and testing an SLI system:

 

- At this time only have one card installed. Complete all tests on one card.

- Uninstall the first card install the other card complete all tests again.

- Install both cards, make the BIOS changes and/or flip the hardware SLI key/switch, configure the driver for SLI support, enable the show GPU load balancing option to make sure SLI is actually working. It was discovered during testing that some beta drivers mainly the 82.* series were giving false OK results during SLI testing due to the fact they were not actually using SLI and the tests were only happening on one card/GPU. Complete all tests again. If the first two tests proceeded OK you probably have good hardware individually, however a failure at this stage more than likely indicates a problem outside of the actual video hardware. This is likely a driver issue, sometimes using a different driver version can cure this, sometimes not. 82.* drivers are not recommended for curing SLI issues as all tested drivers did not actually support SLI even though they showed support and configuration options/messages for it.

 

b. Your probably starting to hate hearing it, but let the system sit a while at idle.

 

c. Reboot.

 

d. Run 1 loop each of 3Dmark03,05,06. Reboot between each run. This will likely go off with no problems. Note the Benchmark scores at this stage.

 

e. Run 3 loops each of 3Dmark03,05,06. Reboot between each run. 50% chance of problems at this stage if you have bad hardware. The benchmark scores should be nearly the same, perhaps 2-3% lower than the single loop runs. Significant drops in the scores MIGHT indicate a problem.

 

f. Run 10 loops each of 3Dmark03,05,06. Reboot between each run. 78% chance of problems at this stage if you have bad hardware. The benchmark scores should be nearly the same, perhaps 2-5% lower than the single loop runs. Significant drops in the scores MIGHT indicate a problem.

 

g. If you made it this far your hardware is very likely good. If you want to confirm it for sure, Run 20 loops each of 3Dmark03,05,06. Reboot between each run.

95% chance of problems at this stage if you have bad hardware. The benchmark scores should be nearly the same, perhaps 2-7% lower than the single loop runs.

Significant drops in the scores MIGHT indicate a problem. Additional runs beyond 20 loops (30,40,50) have not been found to be significantly more revealing of problems.

 

h. Now go have some fun play Doom 3 (or a game such as Quake 4 based on the same engine), or Halflife2, or any other 3D game. It is recommended that you play for at least 2 hours.

 

i. If you made it this far 99.99% chance you have good hardware.

 

j. At this point I will address what I consider to be a contentious issue in this matter. That is the UNDER-Clocking of the Video card as a fix for these issues. You can try it, however if you have tried everything else in this article and under-clocking is the ONLY thing that fixes your problems (other than the SLI issues, consider only single card configs here) there is something wrong with your card RMA it, do not live with an under-clocked card and lose value for the money you paid for that card.

 

What to do if something went wrong at this stage (At any point during the step 8 testing):

 

a. If you have more than one card to test, swap the cards and test the other one see if that one fails in the same or similar situations.

 

b. If you have the machine in USB 2.0+1.1 mode, change it to 1.1 or USB disabled. (If you have a USB Keyboard/Mouse than obviously turning it all the way off will require a USB PCI add on card)

 

c. Try the following excerpted/paraphrased from the NVIDIA Driver release notes for the 80.X series, and also found on many web forums.

 

It seems the root of the Stuck in thread error (which you may or may not see when the system crashes/locks/BSOD) may be a problem with dual cores and the nforce driver's internal threading model....

 

Therefore the following registry changes will modify the drivers internal threading model:

 

1. Go into Regedit and determine the current primary display card by looking in HKey_Local_MachineHardware/DeviceMapVideo and note the GUID (global unique indentifier assigned by Windows) for the entry "devicevideo0" which is the long string at the end of the entry in brackets { }.

 

2. Edit HKey_Local_MachineSYSTEMCurrentControlSetControlVideo{guid}0000, where {guid} is the number derived from the above step.

 

3. Open the "0000" directory and enter a new key called OGL_ThreadControl and give it a value of 2. This will disable multithreading in the driver for all OGL apps.

 

4. In the same "0000" directory, to disable driver multithreading for all D3D apps, you can enter WTD_EXECMODEL and give a value of 0.

 

 

What the release notes fail to make a point of is that in an SLI system this actually may need to be done twice once for each card. After you do the first card, go back to:

HKey_Local_Machine/Hardware/DeviceMap/Video there will be another of these keys (there are probably actually 3-5 of these entries) that is going to look like and have a similar sub key structure as the first one. You'll note that the other keys have much more simple structures leave these alone. Find the entry for the second card and repeat the procedure.

 

 

d. If the single card tests came out ok, then your hardware is good and any SLI troubles can be attributed to a driver problem. More than likely you have an AMD or Intel Dual Core CPU.

 

There is a particularly nasty bug in the drivers which Nvidia has not yet worked out affecting SLI and Dual core CPU's. I have heard some limited reports of SLI problems with Hyperthreaded or single core CPUs from both AMD and Intel, but they are far fewer.

 

At this point your only options are:

- To drop to single card/GPU operation, and like the author of this article wait for NVIDIA to fix the problem in newer drivers.

- To try dropping to the 77, or 78 series of the drivers which don't seem to have this bug, and in testing so far (admittedly more limited than the 8x.x series testing) have not had problems with the dual core CPU. However testing shows that support for some of the newer graphics capabilities in some games and benchmarks seems to be hampered by the 7x.xx drivers. In testing the drop back to 77 or 78 series drivers had little effect on performance in Halflife 2 (Original edition, it has a significant effect in Lost Coast), 3Dmark03, or 3dmark05. It did however cause a 3000 point drop in 3Dmark06 results. The main difference in 3Dmark06 is support for HDR (High Dynamic Range). There are not too many things using HDR yet, Halflife2:Lost Coast, and Farcry with a patch applied are the only ones that spring immediately to mind. Therefore at this stage the loss of full HDR support might be only a small problem. This will probably become more of a concern as 2006 progresses, so hopefully we will finally see some sort of a fix in the 8x.xx series of drivers for the Dual core issues soon.

 

 

Some resources that might be of help and are references for this article.

 

Support forum thread where most of the initial testing and process methodology development for this article took place

------------------------------------------------------------------------------------------------------------

 

http://www.xfxforce.com/forums/thread.jspa...D=1677&tstart=0

 

******

 

MS Dual Core CPU Hot Fix

------------------------

Info: http://support.microsoft.com/Default.aspx?id=896256

 

Patch itself: http://www.sendmefile.com/00157879

 

****

 

Threads with Help

-----------------

 

http://www.hardforum.com/showthread.php?t=983781

 

http://forums.vr-zone.com.sg/showthread.php?t=46798

 

****

 

MFG support web sites used in developing this article

----

 

http://forums.nvidia.com/index.php?showforum=24

 

http://www.xfxforce.com/forums/index.jspa

 

http://forum.abit-usa.com/

 

http://www.dfi-street.com

 

 

****

 

RMA site for XFX Video cards:

-----

http://www.xfxforce.com/web/support/showRma.jspa

 

****

 

Other Sites with good information

---------------------------------

 

http://www.guru3d.com - Any utilities, drivers or software that I forgot to include a link to in the main article body can most likely be retrieved from the downloads section at guru3d.

 

*********

 

Contacting the Author to report your own results, and/or provide updated information:

-------------------------------------------------------------------------------------

 

I can be contacted at [email protected] (You will likely get a response from a different address but I don't want to expose that address to anymore spam than it already gets).

Please tell me all the relevant details about your system so I can keep my cross reference tables on problems and solutions up to date.

After 2/6/2006 (Because the first version is just being made available now so its not there yet to be updated) the most current version of this article can always be accessed at:

http://www.epithna.com/index.php?action=load_cat&ID=12

 

I promise the article found at that location will be better formatted as well, since I will have full XHTML/CSS markup capabilities at my disposal, web forums are somewhat limiting in that regard.

 

********************

 

Testing Services

----------------

 

In the development of this article I has several people email me and ask if I could test their card in my known stable system. Well once I had a known stable system anyway.

At this point I am willing to do this on a limited basis; I do have a day job and a family after all. I can test at maximum 2 cards per day one during the day (setting off the test before I leave for work) and one at night (setting off the test at night before I go to bed). If you are interested in me doing this email me at the address above to see if I have the capacity to handle testing your card in a reasonable span of time, and for details on shipping, ETC.

 

********************

 

Revision History

------------------

.01 Posted to XFX forums in January 2006

.02 - .99 Posted to other web forums, and/or emailed to various requestors January 2006- Feb 5th 2006

 

 

1.0 2:49 PM 2/6/2006: First Official Version, Heavily revised and updated, Posted to XFX, NVIDIA, ABIT, DFI Forums and well as www.epithna.com

Share this post


Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
×
×
  • Create New...