I've got an idea. How about I try some different graphics cards in the KT133A mobo? I could rerun some of those tests. But this time, is there anyone out there who could double-check the new 3DMark 2001 s.e. scores? I have these graphics cards available -

HiS Radeon ATI 7500
MSI Ti4200 (64 MB, 4x AGP)
Gainward MX 440 (128 MB, 8x AGP)
Sapphire? Radeon 9800 Pro (128 MB, 256-bit)

Also available -

Gainward FX5200 (256 MB, 128-bit) - Not very common graphics card.
Voodoo 3 2000 AGP - Not very good for 3DMark 2001 s.e. tests.

Has anyone got a KT133A mobo and one of these graphics cards? I guess the graphics card doesn't have to match *exactly*.

elfuego wrote:

Retro - are the 3dmark 2001 se scores updated? The score between 7600 and 8867 seems pretty lousy for that video card and KT133a. I mean, I've had more then 10 000 score with a radeon 8500 which is a much worse video card then the ti4200! And that was done on a Asus A7V133-c w/ palomino 1800+. I was expecting something between 11000 and 15000 in your case.

I have been looking on the net for Ti4200 reviews using the KT133A chipset. I haven't found any reviews yet, but I did find this forum post -

http://www.forum-3dcenter.org/vbulletin/archi … hp/t-55548.html

It's in German. From what I can gather, the poster is using the same graphics card (MSI G4Ti4200 TD-64), and the same type of RAM: SDRAM 133MHz, and also the KT133 chipset (although it is not KT133A), along with an XP1800+ CPU. I appreciate that his rig is not the same as mine, but it's the closest I can find at the moment. The poster seems to be surprised that his 3DMark 2001 score is only approximately 8500.

Further down this webpage, another poster says that he can (only) get 9150 3Dmarks, with a Ti4200. His system looks better than mine -

AMD Athlon 1333@1470 (147MHz Cas2)
512MB SDRAM
Geforce4Ti4200 (280/600)
MSI KT Turbo (VIA KT133A)

Is there anything else on this forum webpage that is interesting? I can't read German very well, so I may have missed something. Thanks.

According to the posts in 3dcenter the score is ok.
If I remember correctly, my setup with a KT266A (that has also the common via chipst problem) was like this
4in1 driver V 4.43 (not newer)
Georges PCI Latency Patch: http://www.georgebreese.com/net/software/#PCI
Georges Memory Interleave Enabler: http://www.georgebreese.com/net/software/#INT
if it is running stable and fast after this, the 'do not touch a running system' rule applies.

OK, thanks a lot. I will now do the following -

1) Uninstall the nVidia graphics driver

2) Install 4in1 driver v 4.43. Hopefully, this will simply "overwrite" the existing 4.35 4in1 driver. When I install the 4.43 driver, I will select "turbo mode" for AGP, if this option exists. I will also select "yes" to install the other 3 segments of the 4in1 driver. ie, I will install all 4 options. (If this is a bad idea, please let me know.)

3) Install George's PCI latency patch. Currently, this is not installed because I didn't fully appreciate its importance.

4) Install George's Memory interleave enabler. Again, this is not currently installed. (BTW, after I install this, do I make any changes to the BIOS set up area memory timing options?)

5) Reinstall the nVidia graphics driver, version 45.something.

6) Rerun 3DMark 2001 s.e. again.

Well, just read out the memorys SPD EEPROM for each DIMM seperately and then check if the BIOS has the correct setting.
If you are unsure, run Memtest86 afterwards.

You will see if the Memory Interleave Enabler works if the memory performance goes up, which also increases 3dmark scores.

Has anyone got a KT133A mobo and one of these graphics cards? I guess the graphics card doesn't have to match *exactly*.

FWIW if you just want a ballpark estimation, my T-Bred 2400 (266fsb) in an Nforce2 mobo with a 9800Pro used to score around 13,000 in 3DMark01se. With a Thunderbird and SDRAM, I'd expect you'd have a hard time breaking 10K. So 8000 seems about right for a Ti4200 with everything at stock speed.

Also, you mentioned the QDI board a few posts back... you're right, it's not a well designed board. QDI is pretty low-tier in general, and the Kinetiz was particularly troublesome. However... I don't recall it being too bad of a performer when (if) you could get it working. If you're wanting to run a newer CPU, though, look for a different board. There's no techinical limitation preventing one from running a T-Bred or Barton in a KT133/a board, it all comes down to whether the BIOS will support it.

I have upgraded the VIA 4in1 driver, from 4.35 to 4.43. I have also installed George Breese's "Latency" and "Interleave" patches. The results below are updated benchmarks -

Mobo C tests

• Test 1 - No FSB overclocking:

3Dmark '99 = 10035, 20291 (Was 9959, 20319)
3Dmark 2000 = 9271 (Was 8930)
3Dmark 2001 fe = 7835 (Was 7559)
3Dmark 2001 se = 7928 (Was 7671)

It looks like the VIA chipset driver upgrade & George Breese patches have improved performance by approximately 3%.

I can't seem to find the numbers I had saved for 3dmark on my kt133 and 9800.. but I can tell you everything I did to optimize it:

BIOS:
First load optimized defaults.
Set Fast RW Turnaround on
Set CPU command decode to fast.

Memory:
Having good ram helped alot. When I swtiched over to CL2 ram, sandra showed me about 10-20 mbps theoretical increase, but in far cry I went from an average of 27 to 36 fps. I know Far Cry is a super demanding game and outclassed this hardware, but do the math... that's a 25% increase! The ram I last used was 2x512 Crucial. I was able to set all the memory timings to the lowest setting after raising the voltage one notch.

As far as drivers and patches and such.. I was using 4-in-1 4.37, I read somewhere the older ones worked better for KT133, however I never compared it myself. I also used the PCI latency patch .19d, however .19 should be fine. I don't think the memory interleave patch is needed, since you can set that in the BIOS yourself, and I believe it also set by the 4-in-1.

prophase_j wrote:

I don't think the memory interleave patch is needed, since you can set that in the BIOS yourself, and I believe it also set by the 4-in-1.

This has always been a bit dubious with the via software, what is set and what isn't. I wouldn't always trust BIOS options 100%. Alot of this stuff is programmed in a hurry on beta documentation and buggy as hell.

But this test is easy, just disable the memory interleave patch driver in the device manager (show non-pnp-drivers), set it in bios and compare performance.

elianda wrote:

But this test is easy, just disable the memory interleave patch driver in the device manager (show non-pnp-drivers), set it in bios and compare performance.

Even easier is to check the register setting of the chipset, with a program like wpcredit.

If you want easy just get CPUZ or Everest. Looking through chipset docs and then digging through a register reader is a bit reinventing the wheel.

Old Thrashbarg wrote:

FWIW if you just want a ballpark estimation, my T-Bred 2400 (266fsb) in an Nforce2 mobo with a 9800Pro used to score around 13,000 in 3DMark01se. With a Thunderbird and SDRAM, I'd expect you'd have a hard time breaking 10K. So 8000 seems about right for a Ti4200 with everything at stock speed.

I just checked one of my recent posts, in another thread called "more fun and games with VIA's KT133/A", and found the following info (which has been "summarised") -

Mobo: Epox 8KTA3+Pro
Software: VIA 4-in-1 4.35, no "Breese patches" applied.
Graphics: Radeon 9800 Pro (128 MB, 256-bit)
CPU: XP-M, unlocked multiplier set to 17x (2.27 Ghz real clock speed)
3DMark2001 s.e. (default settings) = 11547

Mobo: Abit KT7A Rev 1.3, non-Raid.
Software: VIA 4-in-1 4.35, no "Breese patches" applied.
Graphics: nVidia 6800 GT (256 MB, 256-bit)
CPU: XP-M, unlocked multiplier set to 16.5x (2.20 Ghz real clock speed)
3DMark2001 s.e. (default settings) = 12053

You say that the "8000 seems about right for a Ti4200 with everything at stock speed." I agree.

I made a few changes to the KT133A based set up, but unfortunately it hasn't made the 3DMark 2001 s.e. score any better.

I removed the QDI Kinetiz board, and replaced it with an Abit KT7A board.

I removed the Infineon 256 MB SDRAM stick, and replaced it with a 512 MB SDRAM ("no name/brand") stick. Reading prophase_j's post above, I think the best thing I can do is to get a stick of "CL2 SDRAM".

I made some BIOS set up area changes (from prophase_j's post above) - First load optimized defaults, then set "Fast RW Turnaround" = on, and also set "CPU command decode" = fast. I also set the AGP aperature from 64 to 128.

The 3DMark 2001 s.e. default settings score is now 7845, which is actually less than the score obtained on the QDI Kinetiz board, which was 7928. Without changing the SDRAM, I don't think that I can get this value to significantly improve..

I might try another graphics cards in this Abit board. Also, I might be able to get another Ti4200 to test. Hopefully tomorrow. This one is by Gainward, and a review can be found here -

http://www.bjorn3d.com/read.php?cID=73&pageID=161 (Page 2 of 2)

What's interesting is that this test shows a 3DMark 2001 s.e. score of 9652. The test rig includes:

* AMD Athlon XP 2000+ (I am using a t-bird 1400 mhz CPU)
* Leadtek K7N420DA nForce motherboard (I am using a KT133A Abit board)
* 2 256MB OCZ PC3000 DDR RAM (I am using 1x512 MB of "crappy" PC-133 SDRAM)

What's interesting is that this test shows a 3DMark 2001 s.e. score of 9652.

You're a few points shy of 8000 on your system, but it's a pretty small difference, and going to good CAS2 RAM will probably put you right at the expected mark. When you're looking at 100 points or less, i.e., 1% or so, you can generally chalk that up to normal statistical variations.

That test you linked has a faster CPU, not just in clock speed but the newer architecture is a bit faster clock-for-clock as well, plus there's a significant boost (at least in benchmarks) from having DDR. Given that 3dMark 2001 is very dependent on CPU speed, those results are perfectly in line with what you're seeing on your system. If you were to put that same 2000+ into your KT133A system, you would probably score somewhere around 8700.

Edit: Speaking of which, you mentioned earlier having a 2400+ available? Maybe I missed a post about it, but have you tried that chip in the Abit board? That board should be able to run it just fine.

I am using an Abit KT7A revision 1.3 non-Raid board. I just put a t-bred 2400+ rated CPU in to it, and the BIOS POST screen displays:

AMD Athlon XP 2400+

08/22/2002 "B4" (That's part of the BIOS string)

Inside Windows 98, I check out Sandra and she tells me that the FSB is 133, the multiplier is 15x, and the real clock speed is 2.0 Ghz. I am fortunate because this is a 1.3 board. If it was a 1.0, 1.1, or 1.2 board this would not happen. You would still get an excellent 2 Ghz real clock speed, but it would be in this format: FSB 100, multiplier 20x.

With the 2400+ rated CPU installed, I try and rerun 3DMark 2001 s.e., but it repeatedly quits back to the desktop. Perhaps I'm not so fortunate after all? What I'll do now is remove the Ti4200 card, and replace it with a Radeon 9800 Pro card. That way, I can feed it power from the second PSU, so that the first PSU which goes in to the mobo can power the mobo+CPU+RAM and nothing else.

That's really strange... since it recognizes the chip by name the BIOS should be new enough. (Did you clear CMOS after installing the new chip?) The T-Bred uses a lot less power than the T-Bird, so it shouldn't be a PSU issue...

I dunno how deep you want to get into it, but have you tried any of the other usual tweaks with that particular combo, like different 4-in-1 drivers, setting the AGP speed to 2X, upping the Vcore and drive strength by a notch, etc.? Kinda grasping at straws here, but it could just be that there's some setting that works fine on the TBird, but not on the newer CPU for whatever reason.

I decided to "wheel out the big gun" instead. I put an nVidia 6800 GT in to the board, nvidia driver version 77.something. I rerun 3DMark 2001 s.e. (now as always, on default settings), and it freezes. I think there could be something wrong with the board. It can't be a power problem, because I'm using a meaty 5v 40 amp PSU just for the mobo+CPU+RAM.

I remove the t-bred 2400+ CPU (which is in visibly very poor looking condition), and replace it with another t-bred 2400+ CPU. This one looks much better, but *always* goes mad if I use a Radeon 9800 Pro card with it. (Something is wrong with this CPU too.) However, with the nVidia 6800 GT card, I am lucky and the 3DMark 2001 s.e. test runs without error. The score I get is 11615.

>> Did you clear CMOS after installing the new chip?

Hehe, after I smoked my first 2 t-bred 2400+s in this Abit board, I now pop out the button battery AND clear the CMOS jumper. (I don't take any chances with this board any more.) It's a shame, because my 2 t-bred 2400+ replacements I now use are probably both defective in some way.

I have tried various tweaks inside the BIOS. After doing lots of tests, I am never too certain which ones have really worked or not.

What are the volatges ou have set for the I/O and CPU in the BIOS?

prophase_j wrote:

What are the volatges ou have set for the I/O and CPU in the BIOS?

It's currently set to "Default", which is 1.65V for the core and 3.4V for the IO. If I go to the BIOS "health status" section, the readings say: 1.64V for the core, and 3.39V for the IO. Also, the other readings say: 5V=4.95 and 12V=12.30 IMO, all 4 of these values seem OK to me.

Edit: Although I could be wrong of course. 😉 Perhaps I need to take in to account failing caps? Perhaps I need to increase the voltage? The board looks "old and tired". Very dusty. It looks like it's seen a lot of action. Having said that, it now works OK with the other t-bred 2400+ CPU. It's run 3DMark 2001 s.e. twice now without failing.

Old Thrashbarg wrote:

I dunno how deep you want to get into it, but have you tried any of the other usual tweaks with that particular combo, like different 4-in-1 drivers, setting the AGP speed to 2X, upping the Vcore and drive strength by a notch, etc.? Kinda grasping at straws here, but it could just be that there's some setting that works fine on the TBird, but not on the newer CPU for whatever reason.

Tomorrow, I'll remove the nVidia 6800 GT card, and put the Ti4200 back in and try again. I'll leave the 2400+ t-bred in the mobo. Two things to note: 1) Earlier today, I had swapped out the MSI Ti4200 for an Abit Ti4200. 2) I was having problems with this Abit Ti4200 + t-bird 1400 mhz + Abit mobo + 3DMark 2001 s.e. (It would unexpectedly quit back to the desktop.) I will try and get this Abit Ti4200 working reliably with the Abit mobo + t-bred 2400 rated CPU. I will mess about with the BIOS settings, VIA driver versions, nVidia driver versions, reinstall it all again perhaps, etc.

Earlier today, I experimented with different versions of the nVidia driver package - eg, 29.xx, 30.xx, 45.xx, 56.xx, 61.xx. The earlier versions appeared to provide faster results, but the 3DMark tests never finished as they quit back to the desktop. The 45.xx driver seemed the most stable.

Also, I bought some 128 MB PC133 CL2 SDRAM sticks from ebay. They were only £1 each.