648max + 9700 Solution

Question

HI Everyone,
This message is aimed at anyone who is having stability problems with the MSI648 and ATI 9700 Pro, below is a copy of the e-mail i sent to ATI, Crucial, MSI and SIS stating what i thought the problem was. If your interested read it all...if not just skip to the section titled solution.
I have posted here as MSI/SIS have not responded to any mails i sent asking for this to be investigated/confirmed/fixed so i hope its of use to some of you. My system is now running 100% stable and has been for a week now.
I must also mention that ATI and Crucial replied to this mail (and in ATI's case given it further investigation at the highest level) and have given me excellent service.
*Note this also affects the Asus P4S8X (again the 648 chipset.
From Stuart Davidson
To: tech support at Crucial.com/uk, MSI support, ATI support, SIS Support.
All,
Outlined below is the concern i have experienced and fixed with an incompatability/fault in the SIS 648 chipset when using the ATI Radeon 9700 Pro in AGP8x mode. I am sending this mail to you all so that Crucial and ATI may give the fix to any users that call with a similar concern and so that MSI and SIS can work to ensure new motherboards are fixed. If i have used the incorrect e-mail address for anyone i would appreciate it if you could forward the e-mail internally to the correct department.
I recently had to return a faulty Crucial 9700 pro due to graphical anomolies (chess board pattern in 3d applications overlaying the graphics) and received a replacement on 22nd Jan 03. The new card is a 3.0 revision.
My Main System Specs:
Intel Pentium 4 2.53Ghz
MSI 648 Max (MS-6585) Motherboard
Crucial 128Mb ATI 9700 Pro
512mb DDR333 - Samsung
40Gb IBM Deskstar 120GXP 7200RPM Hard Drive
Q-Tec 400w PSU
Win XP Pro -Sp1
After installing the replacement 9700pro in my PC i checked the glitches i was previously experiencing with the first card were no longer ocurring. This involved running about 10 seconds of a few games - which worked flawlesly. As the card was the newest revision i decided to run 3dMark 2001 SE to see if the score would differ from the previous card- about 1minute into the benchmark the Monitor went into standby. I reset my PC using the reset button on the front of my case as i could not get the monitor out of standby. Upon restarting i ran 3dMark and this time after about 30 seconds the screen froze and the PC would not respond (again using the restart button was the only way to get a response). This continued to occur every time i ran a 3d application on the system (both Open GL and Direct3d)at the most 30 seconds after starting the application/game. From Wednesday evening until Sunday evening i spent a considerable time (20+ hours) trying to fix the concen, here is a list of what i tried (not necisarily in any order):
Reinstall of Win XP Pro
SIS AGP Drivers 1.11, 1.12beta, 1.12final and 1.13, 1.14
Drivers bundled with the card, Catalyst 2.5 and 3.0a
Direct X 8.1 and 9.0
Win XP SP1
Fastwrites on and off using Smartgart(ATI control Panel)
AGP 4x and 8x (changed using Smartgart/Control Panel as my motherboard wont allow you to choose AGP speed in BIOS)
MSI 648Max BIOS 1.0, 1.1, 1.2, 1.3, 1.4
Windows 98SE
changing PSU to a Mercury 300w PSU
Changing HD to a Maxtor 15gb 5400rpm drive
Dismanteling the entire machine and reseating all components
Removing all non necisary hardware (dvd/cdrw, floppy, NIC etc)
Calling Crucial tech support
Underclocking the CPU to 2.0Ghz
Underclocking the DDR to 133mhz
None of the above made any difference to the stability of the 9700Pro.
I was ready to give up and packed the card up ready to be RMA today. Late last night as a last resort i was searching through Rage3d.com's support forum and came across a post by someone with the same concern - they had a "unique" way of fixing this.
*NOTE* There are a huge number of problem reports from people using motherboards based on the SIS 648 chipset and Radeon 9700 on the Internet (doing a google search on "648 9700 pro problem" gets you 10 pages of results. Browsing forums you find even more (not just MSI boards).
Cause of the Concern
It appears that there is a concern with the SIS 648 chipset supplying a constant 0.8volt as per the AGP 3.0 specifications. It seems that when the card is running 3d aps the voltage through the AGP slot dips below the minimum required to power the card correctly.
Solution
As you may or may not know the AGP 2.0 standard uses 1.5volt so theoretically using AGP 4x should give the card enough voltage to run stable (which explains why the monitor ocassionaly entered standby mode). It seems that using catalyst to set AGP to 4x did not work fully instead you have to either
1. Use your BIOS to select AGP 4x or raise the voltage on the AGP slot. (Both of which the MSI board does not allow)
or
2. Make your card be detected as AGP 4x
As i the MSI 648max motherboard does not allow option 1 i had to use option 2.
How to make an AGP 8x Card detect as AGP 4x
During the POST of a bios checks are performed on the AGP slot to confirm :-
1. If the card supports AGP 8x using Pin 3
and
2. If the AGP slot can support AGP 8x using Pin 11
By placing a peice of tape over Pins 3 and 11 i was able to manually make the Motherboard AGP 4x and the card AGP 4x.
The Results
Last night i ran the card on 3dmark for 2 hours with no concern in AGP 4x mode, I was also able to compete several Unreal Tournament benchmarks at maximum detail and 1600*??? (max resolution) , run Earth and Beyond for 1 hour, complete Codecreatures and GL excess benchmarks several times in a loop.
There was no instability experienced and the performance difference was only -1.5%. Today as a last test i have run the un-modified card on a real AGP4x (1.5v) machine and it has worked flawlessly.
Suggested Actions
Crucial: Make the helpline staff aware of this issue and decide if the fix should be recommended to customers or a warning added to your site about the incompatability.
ATI: As with crucial make support personel aware of the concern and possibly assist SIS in resolving the concern in future revisions
MSI: Inform support personel and possibly add a note to the 648max product page advising of this concern. Also to work with SIS to allow BIOS selection of AGP 4x and AGP voltage changes.
SIS: To work with ATI/MSI to resolve this concern in future chipsets and motherboards.
Should any of you have questions i will be more than happy to assist, if needed i am also happy to test new BIOS revisions or any fixes.
Thanks for your time.

Answer 1

HI Whipsnade,
I cant confirm for sure that the AGP voltage is dropping below 0.8v. But it does seem the logical reason when you look at the monitor going into standby and the freezes, yet its 100% stable at 1.5v. (Esp the standby in my optinion shows it most). I have tried some other 9700pros and 3 out of the 4 showed probs at agp 8x (1.1+3.0rev) but none showed probs at agp4x, if i had a non +9700 card that was agp8x i would test it - but i dont. All of the boards showed problems in the Asus p4s8x as well at AGP 8x which is why i think its the chipset to blame.
I had hoped MSI's tech team would have been able to confirm it...or SIS but that seems not to be the case.
ATI are going to speak to MSI now anyway, so i'll find out from ATI what MSI say.
Am i bothered by not being able to use AGP 8x?...enough to RMA the board - no , but i am going to buy a new MOBO as soon as i have some spare cash. And the new mobo will have an Intel chipset -so granite bay looks the best bet. (but due to the non response from MSI im not sure it will be the MSI GB board :-( )
Stu