My recommendation is to set the I/O to 3.45. The CPU should be okay, if you really wanted to test and isolate you can get Prime95 and run the "small fft" stress test. If it's good for about 30 mins, then you can run the "in-place fft" to see if it has more to do with the memory/chipset. If you can show the CPU and RAM to be stable then only the card is suspect.

Here are some other tweaks I made to achieve stability using the 9800xt:
Drive Strength: changed from D0 to DC
Fast Writes: Off
AGP Aperture: 64mb

Edit:
Don't go and change the drive strength unless you look up some recommended values for the particular card. the value of "DC" is only know safe to me for the 9600/9800.

I have managed to get the system stable. I tried many different settings (BIOS settings and nVidia driver versions) for the Abit Ti4200, but it remained unstable using 3DMark 2001 s.e. The Gainward Ti4200 arrived this morning, so it was a good excuse to remove the Abit Ti4200, and try the Gainward Ti4200 instead. It's a 64 MB model, without a DVI connector - which is a shame. Nevermind, because it works. I get a score of 8621. (That's using the t-bred 2400+ rated CPU, @ 133 FSB.) Currently, I have -

Fast writes = Disabled inside the BIOS.
I am using the 29.something nVidia driver.
IO voltage = 3.5 (It was 3.4, which is the default value. BTW, there is no 3.45 option.)
AGP Aperature = 64

Now the system works OK, I will alter some of the settings above, and see if the system remains stable. For instance, I could Enable "Fast Writes". I could also experiment with a later nVidia driver version. I could also reset the IO voltage to its default setting. Etc.

One thing to note: Before I removed the nVidia 6800 GT card, I tried using the last version of the nVidia driver with it, which is 81.something. It worked OK. In fact, I got approximately 200 more points on the 3DMark 2001 s.e. score. (But my guess is that if I used this 81.xx version with very old DirectX games, it wouldn't work too well.)

Edit: I decided to make the following BIOS changes, and see how that affected testing -

BIOS IO voltage = Decreased from 3.5, to 3.4 (default)
BIOS Fast Writes = Changed from "Not Supported" to "Supported"
3DMark 2001 s.e. = Graphics were corrupted. Test aborted.

BIOS Fast Writes = Changed from "Supported" to "Not Supported"
3DMark 2001 s.e. = Some tests ran unusually slowly. 3DMark froze on the results screen - score could not be seen - now ain't that frustrating!

BIOS IO voltage = Increased from 3.4 (default), to 3.5
BIOS Fast Writes = Changed from "Not Supported" to "Supported"
3DMark 2001 s.e. = 8585

Conclusion - the Abit KT7A mobo + Ti4200 + t-bred 2400 rated CPU needs a BIOS IO voltage "boost" by one "notch". In this case, an increase from its default value of 3.4 to 3.5. Enabling the BIOS "Fast Writes" setting seemed to make no improvement to the 3DMark score.

Edit 2: Time for some more "tweaks" and retests. Below, when I say "the screen goes mad" - I mean that the screen looks very corrupted, with weird colours and graphics splatted all over the screen.

Uninstalled nVidia 29.xx driver, installed nVidia 81.xx driver.
3DMark 2001 s.e. = after a period of time - blank screen. Reboot time.

BIOS core voltage = Increased from 6.5 (default), to 6.75
BIOS Fast Writes = Changed from "Supported" to "Not Supported"
3DMark 2001 s.e. = Screen goes "mad". Reboot time. BIOS screen still looks "mad". Switch off power!

Uninstalled nVidia 81.xx driver, installed nVidia 45.xx driver.
3DMark 2001 s.e. = Screen goes "mad". Switch off power.

BIOS core voltage = Decreased from 6.75, to 6.5 (default)
3DMark 2001 s.e. = Screen goes "mad". Switch off power.

Uninstalled nVidia 45.xx driver, installed nVidia 29.xx driver.
3DMark 2001 s.e. = Freezes. Reboot time.

That's a pity. This means that I am back to square one. Currently, I have *no* stable set of settings.

BIOS general settings set to "fail safe".
BIOS IO voltage = Increased from 3.4 (default), to 3.5
3DMark 2001 s.e. = Screen goes "mad". Switch off power.

Uninstalled George Breese's "Memory Interleave enabler"
3DMark 2001 s.e. = Graphics were corrupted. Test aborted.

Removed the t-bred 2400+ rated CPU, replaced with an unlocked mobile barton CPU. Set to its default "BIOS speed" of 6x @ 133 = 800 mhz.
3DMark 2001 s.e. = Screen goes "mad". Switch off power.

Uninstalled George Breese's "Latency Patch"
3DMark 2001 s.e. = Quits back to desktop

BIOS IO voltage = Increased from 3.4 (default), to 3.5
3DMark 2001 s.e. = Screen goes "mad". Switch off power.

Conclusion - there could be some kind of incompatibility with the Ti4200 card(s) and the Abit KT7A mobo. The (MSI) Ti4200 tests worked fine in the other KT133A (QDI Kinetiz) board.

Edit 3: I think I have got the system stable. In order to get the Abit KT7A mobo more stable, I removed the (Gainward) Ti4200 card, and replaced it with a "crappy" FX5200 card. It's a 128-bit 256 MB card, so it isn't the worst card in the world. 3DMark 2001 s.e. now works without going mad.

I also did the following -

Reinstalled the "Latency Patch"
Reinstalled the "Memory interleave patch"
BIOS Fast write - Set to "Not Supported" (I also tried "Supported" and this worked OK)
BIOS IO voltage - Set to 3.50 (I also tried 3.4 and this worked OK)
CPUMSR - set multiplier to 15x (giving real clock speed of 2 GHZ, at 133 FSB)

3DMark 2001 s.e. = 5483

I also tried setting CPUMSR to 16x and also 16.5x multiplier, but the 3DMark program quit back to the desktop.

Edit 4: I am determined to understand why this hardware combination is so problematic: Abit KT7A + Ti4200. I just tried another Ti4200 card, and it failed to run 3DMark 2001 s.e. This card is another Gainward, but it's got a DVI output connector on it. I'll check in a minute, but it might be a "slower clocked" 128 MB version. I went to the mobo's BIOS set up area, and when I saw a field that said "Enabled", I set it to "Disabled". So, I created a kind of "over the top" "fail safe" overall setting. Then, I reinstalled the 29.xx nVidia driver. I continue to use the unlocked mobile barton CPU, but its speed has not been increased by the CPUMSR.exe utility. Instead, it runs at the BIOS default speed of 6x/133 FSB = 800 Mhz. 3DMark 2001 s.e. now works OK. I think the score was about 6000, something like that. I wasn't paying attention, because I now need to use the CPUMSR.exe to increase the CPU's speed. Then, I can get a proper score...

retro games 100 wrote:

..I could Enable "Fast Writes"....

On a VIA chipset, this is certainly going to lead to doom. Don't bother with it. Even if it did work, the preformance boost would be almost non-existent.

Very impressed otherwise 😀

I am confident that I have finally got the system stable now, with a Ti4200 graphics card. To recap, I am currently using -

Abit KT7A mobo, PCB revision 1.3
Unlocked mobile barton CPU (2400+ rated)
CPUMSR.exe set to 15x (@ 133 FSB), real clock speed = 2 Ghz
Coolon.exe cooling enabled
Ti4200 graphics card
nVidia driver version 29.xx
VIA 4in1 version 4.43
George Breese's memory interleave and latency patches installed
1 stick of 256 MB PC133 CL3 SDRAM

BIOS settings (the following general settings are essential, otherwise the system is not stable). I simply trawled through all of the BIOS settings, and set all of the "advanced chipset" settings to "Disabled". It doesn't appear to make a jot of difference to the speed performance of the system, because I've had many of these advanced settings set to "Enabled" on another KT133A based system, and it's made no impact on the 3DMark scores. Also, I set the voltage to Default, and the Fast Write to "no support". The AGP "speed" was set to 4x. AGP aperature was 64. AGP drive strength was "Auto". (I haven't investigated what the Ti4200 is able to accept in terms of its AGP drive strength.) Also, CPU command decode = Normal, and Fast RW turnaround was Off.

To summarise the above BIOS related paragraph, I set the BIOS settings to "ultra fail safe".

3DMark 2001 s.e. (default settings) = 8675, 8725 (Run twice)

Now, it may be interesting to Enable each Disabled BIOS setting at a time, and rerun 3DMark 2001 s.e., to see when the 3DMark program fails. I will edit this post later, to detail all of the currently Disabled BIOS settings, and suggest (guess!) which settings I should Enable first...

Edit: I have made just one change from the settings listed above - I have increased the CPUMSR multiplier setting from 15x, to 16x. (This now gives a real clock speed of approximately 2.13 Ghz, from 2 Ghz.)

3DMark 2001 s.e. = 8750, 8816 (Run twice)

The system is still stable.

Edit 2: I made just one more change. I uninstalled the nVidia 29.xx driver, and installed the 81.xx driver. This causes the 3DMark to fail - during testing, the screen goes blank and I have to reboot the machine. So, I uninstall the 81.xx driver, and try the 45.xx driver instead. I then rerun 3DMark 2001 s.e., twice in a row. Scores are 9026 and 9008. That's not bad! And the system is stable.

Edit 3: I have made just one change - I have increased the CPUMSR multiplier setting from 16x, to 16.5x. (This now gives a real clock speed of approximately 2.2 Ghz, from 2.13 Ghz.)

3DMark 2001 s.e. score = Fails. It quits back to the desktop. My guess is - at this point, you may need to increase the voltage setting(s) in the BIOS. But I'm not going to do that. Instead, I decrease the CPUMSR multiplier setting from 16.5x to 16x, because I think that's good enough.

Edit 4: I'm looking at my O.P. from this thread, and it says:

• Test 4 - Upgrade CPU to an "unlocked" (16x multiplier) mobile barton, running at 2176 Mhz. FSB also OC'd to 136 FSB, from 133 FSB.

3Dmark 2001 se = 8867

I'm confident that I have used the same nVidia driver for my recent tests above, and that is version 45.xx. Those tests above provided scores of 9026 and 9008. (In fact, I've just run 3DMark for a 3rd time, and I get 9020.) So, with the Abit board's advanced chipset BIOS settings set to Disabled, plus I'm now using the VIA 4.43 driver (not version 4.35), plus I'm now using the 2 George Breese patches, I improve the 3DMark scores from those earlier tests (which were done on a QDI Kinetiz) board. But what's interesting is that those earlier tests had the FSB OC'd a little to 136. The recent "9000+ scores" have been done without any FSB OC'ing.

One thing you want to consider is that the BIOS, and the VIA chipset drivers will work with a lot of the same settings and functions. So essentially you have left your BIOS blank and let the chipset drivers do the programming. Probably not a bad way to go about it. 4-in-1's have a large effect on the programming of the PCI registers. If you really want to see what's going on, you can turn to wpcredit and simple .pcr file, like I had mentioned earlier.

You should still try:
Fast Command Decode.
Fast RW Turnaround

I'm also guessing your waiting on the CL2 sticks?

prophase_j wrote:

One thing you want to consider is that the BIOS, and the VIA chipset drivers will work with a lot of the same settings and funct […]
Show full quote

One thing you want to consider is that the BIOS, and the VIA chipset drivers will work with a lot of the same settings and functions. So essentially you have left your BIOS blank and let the chipset drivers do the programming. Probably not a bad way to go about it. 4-in-1's have a large effect on the programming of the PCI registers. If you really want to see what's going on, you can turn to wpcredit and simple .pcr file, like I had mentioned earlier.

You should still try:
Fast Command Decode.
Fast RW Turnaround

I'm also guessing your waiting on the CL2 sticks?

Yes that's a good point. Also, I will try both of those BIOS settings, and post back with the results. Plus, I am waiting on the CL2 sticks. Hopefully, I will get them early next week.

Edit: I just tried setting CPU Command Decode = Fast, and RW Turnaround to Enabled inside the mobo's BIOS. I then reran 3DMark 2001 s.e., with the multiplier set to 16x, and the good news is that the system is still stable. The slightly less than good news I suppose is that the score obtained is 9022. That's approximately what I got before, when these two BIOS fields were set to "Normal" and Disabled. Oh well. Perhaps the VIA 4-in-1 4.43 driver takes care of these two BIOS fields anyway? 😀

Edit 2: I'm really confident the system is now stable. I just removed the Gainward Ti4200, and replaced it with a previously troublesome Abit Ti4200 card. This one has 64 MB, but I think it's clocked a little bit higher than the Gainward 128 MB card. I never got this card to run 3DMark 2001 s.e. once without some problem occuring. Now, it runs the test without any problem. The score is a bit higher too - 9140.

Edit 3: The mobo's current (and most interesting) BIOS settings are -

CPU power supply = "CPU default"
- Core voltage = 1.575v (the current XP-M CPU's requirement is only 1.45V!)
- I/O voltage = 3.4v

Fast CPU command decode = Fast
Output drive control = Auto
Enhance chip performance = Disabled
Force 4-way interleave = Disabled
Enable DRAM 4k-page mode = Disabled
DRAM clock = Host CLK

-----

Bank 0/1 DRAM timing = SDRAM 8/10ns
Bank 2/3 DRAM timing = SDRAM 8/10ns
Bank 4/5 DRAM timing = SDRAM 8/10ns

DRAM bank interleave = Disabled
Delay DRAM read latch = Auto
MD driving strength = Lo
SDRAM cycle length = 3
Memory hole = Disabled
PCI master pipeline req = Disabled
P2C/C2P concurrency = Disabled
Fast R-W turn around = Enabled
System BIOS cacheable = Disabled
Video RAM caheable = Disabled
AGP aperture size = 64M
AGP-4x mode = Enabled
AGP driving control = Auto
Fast write supported = No support
CPU to PCI write buffer = Disabled
PCI dynamic bursting = Disabled
PCI master 0 WS write = Disabled
PCI delay transaction = Disabled
Delay transaction = Disabled
PCI master read caching = Disabled
PCI#2 access #1 retry = Disabled
AGP master 1 WS write = Disabled
AGP master 1 WS read = Disabled
Are you reading this = Enabled
PCI master bus time-out = 1

-----

I have made just one change - I removed the specially unlocked mobile barton XP-M CPU (2400+ rated, running at 16x @133 FSB = ~2.13 Ghz real speed), and replaced it with a t-bred 2400+ rated CPU (running at 15x @133 FSB = 2 Ghz real speed). The system appears to be stable.

3DMark 2001 s.e. = 8571, 8552.

In comparison, the unlocked barton (~2.13 Ghz) scored 9026, 9008, 9020.

I have two 2400+ t-breds, both of which are in a questionable condition. The first CPU I tried failed to run 3DMark for more than about 10-15 seconds. I then ran Prime95 (small FFTs test) on it, and it failed after about 1 minute. These 2 CPUs can be seen in this Vogons post -

Can you have 2 power supplies for 1 computer? (CPUs shown 1/4 down on page)

The CPU which fails is the bottom CPU. The current CPU is the top CPU. Ideally, I should clean or replace them both! Edit: The current top CPU always displays a corrupted and unstable desktop when I use any Radeon 7500 or 9800 card.

Edit: I decided to decrease the CPU's core voltage in the BIOS. It was on "default", which is 1.575. Considering that the mobile barton CPU only requires 1.45v, I decided to decrease this BIOS value to 1.5v. I reckon that's OK. And the system continues to be completely stable, running at 16x multiplier. I thought I'd give the Radeon 7500 and Voodoo 3 2000 a quick "head to head".

CPU = XP-M; 16x @ 133fsb (2.13 Ghz)

Radeon 7500; GPU clock = 290 Mhz, Real clock = 230 Mhz (DDR), AGP 2x, Cat driver 6.2

3DMark 99 Max = 9361, 32378

Voodoo 3 2000 SD RAM; 166 Mhz, AGP 2x, Amiga Merlin driver 2.9

3DMark 99 Max = 6841, 32439

prophase_j wrote:

I'm also guessing your waiting on the CL2 sticks?

The CL2 SDRAM sticks have arrived. I don't know if this is bad or not, but the RAM chips on these sticks are only on one side. Does that mean it's high density? And does that mean they are not as good as double-sided low density SDRAM sticks? The CL2 SDRAM sticks are all 128MB. I've managed to find 3 sticks which look identical. Each stick says:

(Sticker on the left)
M(logo - Micron?) MT8LSDT1664AG-13EE1 PC133U-222-542-A
US BZABF6B053 200204
128MB, SYNCH, 133MHz, CL2

(Sticker on the right - I haven't included everything, as some of doesn't look important)
128MB simm_nsNS 16Mx64 133MHz 3.3V SDRAM NP
IBM

The SDRAM stick I removed to accomodate these 3 SDRAM sticks above is made by Infineon. It's a double-sided 256MB CL3 stick. Its sticker says: PC133-333-520

With the three new CL2 sticks in place, I switch on the mobo's power. The simple BIOS memory test/count works, but just before the Windows 98 desktop appears, an error message tells that there is insufficient memory to start windows, and the machine must be switched off. (Sorry, I forget the exact wording.) I go to the mobo's BIOS set up area, and change the memory timing "SDRAM cycle length" value from 3 to 2. I reboot. But then the simple BIOS memory test/count fails. I go back to the mobo's BIOS set up area, and change "DRAM bank Interleave" from Disabled to 4-way. I reboot, and now I can get to the Windows desktop.

Please note: those 2 BIOS set up area changes I made above ("SDRAM cycle length": from 3 to 2, and "DRAM bank interveave": from Disabled to 4-way) - they were just guesses to get the mobo to boot correctly in to Windows.

I run Sandra 2002 Pro, and run its Memory Bandwidth test. I get 969, 967. The VIA KT133 PC133 CL2 SDRAM reference chipset/Memory scores are 1012, 1003. So, a little disappointing perhaps? I quit Sandra and the mobo suddenly reboots! Perhaps I need to remove all 3 SDRAM sticks, and then retest one stick at a time using Memtest?

The following BIOS set up fields seem to relate to memory. I have been messing about with their values, and all of the different values that I have tried so far produce a BIOS POST error message of "Memory test fail" during the simple BIOS POST memory test/count.

Force 4-way interleave = Disabled
Enable DRAM 4k-page mode = Disabled
DRAM clock = Host CLK

-----

Bank 0/1 DRAM timing = SDRAM 8/10ns (Other option values are Normal, Medium, Fast, Turbo)
Bank 2/3 DRAM timing = SDRAM 8/10ns (Other option values are Normal, Medium, Fast, Turbo)
Bank 4/5 DRAM timing = SDRAM 8/10ns (Other option values are Normal, Medium, Fast, Turbo)

DRAM bank interleave = Disabled (Other options are 2-Way, 4-Way)
Delay DRAM read latch = Auto (Other options are "No Delay", 0.5ns, 1.0ns, 1.5ns)
MD driving strength = Lo (Other option is Hi) <-- Maybe I should change this to "Hi" - does that mean high density? Edit: Oops, no. This does not mean high density. The MD driving strength = memory data bus' driving strength, and it is recommended to keep this set to Lo for "normal" RAM loads.
SDRAM cycle length = 2 (Other option is 3)
Memory hole = Disabled (Other option is Enabled)

I have now removed 2 sticks, leaving just 1 stick in the mobo. I continue to mess about, and see what happens...

Edit: I have decided to put 1 more stick in the mobo, to make a total of 2 sticks. 2 x 128 = 256MB in total. I'm now running Memtest-86 v3.5 It says the memory speed is 415 MB/s. I will just wait until Memtest completes...

Edit 2: Memtest says: "Pass complete, no errors"

Edit 3: I ran one quick test. At the moment, I have a basic Voodoo 3 card in the mobo. Before I changed the mobo's RAM, I ran this test -

3DMark 99 Max = 6841, 32439 (16x multiplier, CL3 RAM)

After I changed the RAM to CL2 SDRAM, I rerun the same test -

3DMark 99 Max = 6821, 32658 (16x multiplier, CL2 RAM)

Perhaps this test isn't a very good test? (It doesn't seem to show any improvement.)

Edit 4: I reran this same quick test again. The only change I have made is inside the BIOS set up area. I changed the "DRAM bank interleave" value from Disabled, to 2-Way. The new test result is -

3DMark 99 Max = 6835, 32717

It's interesting to note that the 2nd score (the CPU score) is increasing, from 32439 (CL3 RAM) to 32658 (CL2 RAM) to 32717 (CL2 RAM, 2-way DRAM bank interleave).

I made some BIOS set up area changes, relating to the memory -

Bank 0/1 DRAM timing = Fast (Other option values are SDRAM 8/10ns, Normal, Medium, Turbo)
Bank 2/3 DRAM timing = Fast (Other option values are SDRAM 8/10ns, Normal, Medium, Turbo)
Bank 4/5 DRAM timing = Fast (Other option values are SDRAM 8/10ns, Normal, Medium, Turbo)

DRAM bank interleave = 2-Way (Other options are Disabled, 4-Way)
(no change made) Delay DRAM read latch = Auto (Other option is "No Delay", 0.5ns, 1.0ns, 1.5ns)
MD driving strength = Hi (Other option is Lo)
SDRAM cycle length = 2 (Other option is 3)

Now when I run Memtest, the memory speed is shown as 443 MB/s. (Previously, when the "Bank DRAM timing" was set to "SDRAM 8/10ns" and not "Fast", and also when "DRAM bank interleave" was set to Disabled and not 2-Way, Memtest reported the memory speed as 415 MB/s.) Memtest passes OK, with these new faster memory BIOS set up area settings.

I run Sandra 2002 Pro, and run its Memory Bandwidth test. I get 975, 973 which is a little bit better than the old 969, 967 values. (As mentioned above, the VIA KT133 PC133 CL2 SDRAM reference chipset/Memory scores are 1012, 1003.) I then rerun the 3DMark 99 Max test and get 6831, 33366. A-ha! 😀 Now we are getting somewhere. The 2nd score (the 33366 CPU 3DMark value) has risen again, this time by quite a significant amount.

Edit: I've made just one change - I removed the Voodoo 3 card, and put the Abit Ti4200 card back in to the mobo. I then rerun 3DMark 2001 s.e. These new scores can be seen below, alongside the old scores which used CL3 RAM -

3DMark 2001 s.e. = 9026, 9008, 9020. (old scores, using CL3 RAM, 16x multiplier)
3DMark 2001 s.e. = 9344, 9392 (new scores, using CL2 RAM, 16x multiplier, and also using "faster memory settings" inside the BIOS set up area)

The CL2 SDRAM plus the BIOS memory tweaks have resulted in a performance increase of approximately 3.9% inside 3DMark 2001 s.e.

Edit 2: I have made just one change - in the BIOS set up area, for the "Bank DRAM timing" memory speed setting, I have set this value from "Fast", to "Turbo" -

Bank 0/1 DRAM timing = Turbo (Other option values are SDRAM 8/10ns, Normal, Medium, Fast)
Bank 2/3 DRAM timing = Turbo (Other option values are SDRAM 8/10ns, Normal, Medium, Fast)
Bank 4/5 DRAM timing = Turbo (Other option values are SDRAM 8/10ns, Normal, Medium, Fast)

Now when I run Memtest, it says that the memory speed is 476 MB/s. (Previously, when the "Bank DRAM timing" settings were set to "Fast", Memtest reported the memory speed as 443 MB/s. Also, when the "Bank DRAM timing" was on the "SDRAM 8/10ns" setting, Memtest reported the memory speed as 415 MB/s.) Memtest passes OK. I then run these 3 benchies -

Sandra 2002 Pro "Memory Bandwidth" = 986, 982
3DMark 99 Max = 14373, 33573
3DMark 2001 s.e. = 9467

retro games 100 wrote:

I run Sandra 2002 Pro, and run its Memory Bandwidth test. I get 975, 973 which is a little bit better than the old 969, 967 values. (As mentioned above, the VIA KT133 PC133 CL2 SDRAM reference chipset/Memory scores are 1012, 1003.)

For the record, at FSB 149Mhz with timings at 8/10ns you will get a score of ~ 1050/1040 (my setup).

With turbo settings I've only managed 136 FSB and with that setting I get about 1015/1010.

I was able to beat the listed scores in Sandra without overclocking using a the 8KTA3:

Those memory scores are very good. Perhaps my "new" CL2 RAM sticks are not the best quality? I have made 2 changes inside the BIOS -

DRAM bank interleave = 4-Way (Other options are Disabled, 2-Way)
MD driving strength = Lo (Other option is Hi)

Memtest reports that the memory is the same speed, at 476 MB/s. The Memtest test also passes OK.

Sandra 2002 Pro "Memory Bandwidth" = 986, 982 (no change)
3DMark 99 Max = 14330, 32929 (2nd value has dropped a bit, from 33573)

I then made 3 changes inside the BIOS -

DRAM bank interleave = 2-Way (Other options are Disabled, 4-Way)
CPU decode = Fast (Other option is Normal)
Fast R-W turn around = Enabled (Other option is Disabled)

Sandra 2002 Pro "Memory Bandwidth" = 987, 983 (virtually no change)
3DMark 99 Max = 14392, 33543 (virtually no change from the last "2-Way" test)
3DMark 2001 s.e. = 9477 (virtually no change)

At the moment, I can't seem to make the mobo go any faster. However, I am really pleased that the mobo seems stable. 😀

That memory test is kinda touchy. You would want to make sure you have no other programs running, if you see anything else in the system tray other the the volume control, you should kill them. You can also quickly re-run the test by pressing F5, and limit any input the may alter the end score.

I just set a lot of BIOS options to Enabled. Curiously, doing this has made no difference to the 3DMark scores. Also, the system remains stable. The BIOS options which have been changed from Disabled to Enabled are -

PCI master pipeline req = Enabled
P2C/C2P concurrency = Enabled
CPU to PCI write buffer = Enabled
PCI dynamic bursting = Enabled
PCI master 0 WS write = Enabled
PCI delay transaction = Enabled
Delay transaction = Enabled
PCI master read caching = Enabled
PCI#2 access #1 retry = Enabled
AGP master 1 WS write = Enabled
AGP master 1 WS read = Enabled

(no change) Fast R-W turn around = Enabled
(no change) Fast write supported = No support

Also, I've noticed that the Sandra memory scores have altered very slightly. I now get 993, 966. The first value has gone up, while the second value has gone down.

BTW, I found a webpage which has some KT7A BIOS tweaking information -

http://www.devhardware.com/c/a/Motherboards/K … OS-Tweak-Guide/

Maybe you can try going with a different 4-in-1 driver. Some of these settings were likely enabled by the current driver already, hence no change in performance. Conversely, they may be disable after the driver loads. Some of those options don't have much to do with the memory subsystem, and more to do with the PCI and AGP bus.

prophase_j wrote:

Maybe you can try going with a different 4-in-1 driver. Some of these settings were likely enabled by the current driver already, hence no change in performance. Conversely, they may be disable after the driver loads. Some of those options don't have much to do with the memory subsystem, and more to do with the PCI and AGP bus.

Yes, I see what you're saying. BTW, I did try a different 4-in-1 driver. I uninstalled the nVidia 45.xx graphics driver, then ran the currently installed 4-in-1 driver setup .exe version 4.43 in order to uninstall it. I then installed 4in1 version 4.37, then reinstalled the nVidia 45.xx driver. Unfortunately this has decreased all benchmarking scores. The Sandra memory scores are 987, 961 (they were 993, 966), and the 3DMark 99 scores are now 14314, 33183 (they were 14392, 33543). It's not important. I'm just pleased everything is stable. I think the next test to do is to get some alternative CL2 sticks. Can you recommend any quality CL2 SDRAM? Sorry if you've said this already, but what SDRAM do you use? Also, should I go for a pair of double-sided 256mb sticks (512mb total)? Thanks a lot.

Might I ask, why are you installing 4in1 driver at all? As far as I remember, Win98 SE has pretty stable and up-to-date VIA bus driver. I stopped using that driver since I tried a buggy version a long time ago. Try benchmarking w/o it...

The memory I used was made Crucial, which is the performace brand of Micron. 2x512mb

Actually I think the sicks you have might be very close to what I have. I'm not sure if the double-sided affects the number of banks or not, but I would think that if it is, and each stick is only a single bank, then your not able to take full advantage of the bank-interleave.

elfuego wrote:

Might I ask, why are you installing 4in1 driver at all? As far as I remember, Win98 SE has pretty stable and up-to-date VIA bus driver. I stopped using that driver since I tried a buggy version a long time ago. Try benchmarking w/o it...

(The following post is quite frankly rather dull and rambling.)

I simply guessed that the downloadable VIA 4in1 driver package (eg versions 4.35 or 4.37 or 4.43) would be better and also more up to date than the Windows 98 SE "built in" VIA driver package. I have now uninstalled the currently downloaded VIA 4in1 driver (version 4.37), and when the mobo rebooted, Win98SE then installed its own "built in" VIA 4in1 driver package. Also, I decided to reinstall the unofficial Windows 98 service pack, just for good measure. (Afterwards, I noticed that the desktop appeared a bit sluggish, which didn't look right to me.) After reinstalling the nVidia driver package, I then rerun some benchies. I ran Sandra's memory bandwidth test, and the scores were the same as before. I then ran 3DMark 99, and the frame rate displayed in the corner looked lower than usual, and then the screen became badly garbled. Reboot time.

I reinstalled the VIA 4in1 driver package version 4.43, and reran 3DMark 99. The 2nd CPU score seemed a little bit lower than usual, so I reran this test. Then, I noticed that the image on test number 1 (the racing spaceships) flickered a bit, which looked wrong to me. The test completed, and the 2nd CPU score seemed back to normal (about 33500), but the first "3D Marks" score was a bit lower than normal! (about 14000, and not about 14300). I reran 3DMark 99 for a third time, and I noticed that the scores seemed different (a bit lower) again. Maybe this program is not working too well, and generally produces slightly erratic results?

I decide to uncheck DMA mode on my DVD-ROM drive, and rerun 3DMark 99. It completes OK with an "average" score. However, 3DMark 2001 s.e. quits back to the desktop without completing. I then uninstall both George Breese patches, and I also Disable lots of "advanced chipset" options in the BIOS - these options when Enabled did not seem to improve performance anyway. I also increase the CPU core voltage to 1.525 (from 1.5). 3DMark 99 completes OK with a lower than normal score. 3DMark 2001 s.e. completes OK, with a typical score of about 9400. I then reinstall the Breese Latency patch. I also Enable lots of "advanced chipset" options in the BIOS, because some of them do seem important, such as "PCI Delay Transaction" and "Delay Transaction", although I don't know what the difference between these two are. I also decreased the CPU core voltage (from 1.525 to 1.475), but increased the IO voltage to 3.5 (from 3.4). 3DMark 99 passes but provides another erratic (and slightly lower) score. 3DMark 2001 s.e. quits back to the desktop.

With so many different BIOS options, and other factors (such as different VIA driver versions, condition of the mobo, etc), it's very difficult to know precisely what's wrong. It was working great yesterday! I decide to simply select the BIOS "optimized settings", and then just tweak the DRAM memory timings. Sandra memory test gives 985, 957 which is poor. I notice when I selected the BIOS optimized settings, it set the MD driving strength to "Hi". I wonder if that is a problem? I wonder if this should be set to "Lo" instead. (In fact, I've set this to "Lo". Also, I have set the command decode back to "Fast", and the Fast R-W turn around = Enabled. I have also set the PCI caching to Enabled.) Sandra memory benchmark now = 991, 962. That's a bit better. Also, stability has returned to the system. But who knows, maybe because that's because the mobo has just warmed up a bit?

BTW, inside the VIA 4in1 release 4.37 zip package, there is an interesting "history log" called 4in1release.doc, in MS Word format.

(End of dull and rambling post.)

I accidentally killed the XP-M CPU. I removed it from the Abit board, and put it in the QDI Kinetiz board. I've tested this CPU out many times in this board, without any problem. I cleared the CMOS jumper, and when the BIOS POST message appeared on screen, I went to the BIOS set up area. But the keyboard quickly became unresponsive and I heard a faint strange crackling? noise from somewhere near the mobo and then the screen went blank. I've tried many times to get the mobo to POST with this CPU, but no luck. I put the CPU back in the Abit board, but no luck. I tried a different CPU in the Kinetiz board, and I get to POST. It's a real shame the CPU has died, because that's my only one, and it took a bit of patience and luck to successfully cut the L6 on it. Oh well. 🙁

Edit: I wonder if testing a 24 pin to 20 pin PSU adapter plug caused this problem to occur? It's the first time I have tested one. It looks cheap and nasty. Perhaps this adapter failed, and the PSU fed the wrong amount of juice in to the mobo?