I swapped out the light weight Voodoo3 card, for a Radeon 9800 Pro, but the results are much less impressive. When I use CPUMSR.exe to bump up the multi from 6x to the XP-M's official max value of 13.5x, and then run either 3DMark 99 or 3DMark 2000, the system always immediately reboots. I've tried reinstalling these benchmark apps, messing about with BIOS voltage options, setting the BIOS AGP speed to 2x, but nothing fixes this problem.

I notice that Sandra 2002 Pro tells me that the +5V reading is 4.56. That's quite low. I believe there's a PSU "mod" which allows you to increase the voltage on the 5V rail. Perhaps the Radeon 9800 Pro card is using this rail too heavily? (The Radeon is being powered by a 2nd PSU, but maybe it's not up to the task? It's a typical 350W PSU.)

Edit: I removed the Radeon 9800 Pro from the Epox mobo, because its stability was disappointing when I attempted to run the XP-M CPU at (or near) its full speed. (Just had a thought: I wonder if I should have uninstalled then reinstalled G.Breese's Latency Patch?) I replaced it with a Geforce 3 "titanium 200" model card. Now the system appears more stable. I have the BIOS AGP speed set to 4x. This card is not externally powered, and so is effectively putting more of a strain on the PSU which also powers the mobo/CPU. Quick 3DMark 99 Max tests:

6x, 133fsb = 800mhz -> 8165, 16756
13.5x, 133fsb = 1800mhz -> 11598, 29100

13x, 166fsb = 2161mhz -> failed.

Edit 2: Because I couldn't get the GF3 to run 3DMark 99 @ 13x / 166fsb, I removed it, and replaced it with a light weight STB Velocity 4400 card. This card also would not do this test. So, it looks like the (AGP) Voodoo3 is the winner, because it will run 3DMark 99 at up to 169fsb with a CPU multi of 13x.

Voltage readings from the mobo are not worth attention most of the time because their accuracy can be incredibly poor.

You may have a motherboard or VIA chipset AGP compliance/driver problem there. It could also be an inadequate PSU. Radeon 9800 Pro is a fairly demanding card. The vid card itself could be damanged/dieing too.

VIA + AGP = NO!

I wouldn't bank on the voltage reading either, but if you think that's the lowest reading you have seen on the board then I would say that's the answer. Short version:

More MHZ = more power. Higher FSB (even if lower multi/same operating fequency) = more power.

I have a kt333 and an nforce2 based board; I had to go from a 30a to a 38a on my kt333 to get to and above 1.8ghz. The nforce2 will run with a 30a but it would crash after a day or so of running until I beefed it up with a another 38a.

I have a old PC Power & Cooling Turbocool 300W from 1999 that I think I was running a 9700Pro on. That PSU is pretty decent so it didn't have any trouble. But I went up to a Thermaltake Purepower 420W in 2003, I think. Those two were of the high 5v power era.

I recall putting my 9700 Pro in a friend's Gateway P4 to show him how much his GeForce4 MX sucked. 😀 But that thing only had a ~180W PSU that was obviously at the edge already. The 9700P caused it to reboot frequently, but it would still run games for a bit. I remember we were playing X2 The Threat at the time and it turned out that his P4 2.0 was a big bottleneck.

Thanks very much for everyone's interesting comments. The PSUs I'm using are -

PSU #1 (new): Powers the R9800P vid card, CF HDD, ODD, 2 molex fans.

Output Current
+3.3v +5v +12v
21.2A 30A 16A

-12v -+5vsb -5v
0.8A 2.0A 0.3A

+3.3v & 5v Power 220W
+3.3, 5, 12v Power 330W Max

-12v Power -
+5vsb Power -

Total Power 350W

PSU #2 (used): Enermax 475W (40A on 5v rail). Powers mobo+CPU+RAM, and FDD.

That's an insane amount of PSU power for that hardware. 😁 A 300 Watt PSU was what ATI listed as a minimum requirement. But if you go overkill and still have problems, I suppose that will tell you the problem is elsewhere.

I have wasted another evening on a Kinetiz 7E-A. I tried a Geode NX but simply could not get it to post. According to some posts, a Geode should work just fine on a KT133a, but apparently not on QDI.

Fun thread 😀

I got a couple A7V133's laying around.
And an Abit KV7V which has given me endless problems...

Too bad none of my socket A's have an ISA slot.

ux-3 wrote:

I have wasted another evening on a Kinetiz 7E-A. I tried a Geode NX but simply could not get it to post. According to some posts, a Geode should work just fine on a KT133a, but apparently not on QDI.

On page 1 of this thread, I mention that the T-Bred 2400+ rated CPU (AXDA2400DKV3C) works on this board. I searched ebay for AXDA2400DKV3C, and found lots of results. If you're not keen on trying an XP-M CPU because of voltage concerns, I think a safer bet would be to try a T-Bred. The voltage requirement of this CPU is 1.65V, as opposed to 1.45V of the 2400+ rated XP-M (which also works on this board - I did kill one once, but I'm reasonably certain it was because of incorrect heatsink placement; I must have cracked the core.)

You should still be able to put 1.65v on a 1.45v chip. Those mobiles are basically higher rated chips which are downclocked and 'downvolted' and have modified bridges to make them behave like mobiles do.

I know all that. My XP-M 2600+ has served years as XP-3400+ at 1,70V.

I don't want to up the speed, I want to slow down. Ideally to 300 MHz with a mobile CPU. I wonder if I could make a mobile mod on the 1400c

I thought I read somewhere that athlon CPU's don't go lower then 600MHz.
This article is not what I remembered, but they draw the same conclusion:
http://www.silentpcreview.com/article37-page1.html

Afaik, it is possible to lower the multi to 3x with software, like Central Brain Identifier. 3x 100 MHz is 300 MHz, and I think that I have done that with my XP-M before. Actually, I just checked, and yes, it works. I tried two XP-M, and they both allow for 300 MHz, provided the mobo does not interfere. In fact, my SIS Chipset allows 95MHz FSB, which delivers 285 MHz final speed. That is slow but I fear it may still be way to fast for DOS.
With cache off, the range translates well to the performance of 286 to 486. But without ISA slot, PCI sound will falter with no cache. So the whole thing would only work in the presence of an ISA card. And we are back on the KT133a. Sigh...

OK, Thanks for clearing up the subject of the athlon's minimum multiplier.
Yes the PCI sound problem without cache.... I experienced that before with a SB-Live.

Update: I could talk my TB1400c into accepting multipliers. I could lower him down to 5.5x, reaching 550 MHz that way.

I also got the Geode working on the KT133A. However, I was not able to change its multi by software at all. So on my KT133A, I see no way to go below 550 MHz real clock.

I also found voltage settings for 1.53 Volts amongst others.

Many post-2000 Intel CPUs have a thermal protection function called clock modulation that can be software controlled. I've seen Pentium M running at 200 MHz and I know Celeron M can do it too. I had a Sony Vaio Pentium M notebook that came with software that could enable a fanless mode that used this feature. RMClock can control it too.

I'm guessing that this is still affected by FSB, so if you want 200 MHz you probably need one of the CPUs with the 100 MHz FSB (aka 400 MHz quad pumped).

This is separate from Speedstep.

ux-3 wrote:

Afaik, it is possible to lower the multi to 3x with software, like Central Brain Identifier. 3x 100 MHz is 300 MHz, and I think that I have done that with my XP-M before. Actually, I just checked, and yes, it works. I tried two XP-M, and they both allow for 300 MHz, provided the mobo does not interfere. In fact, my SIS Chipset allows 95MHz FSB, which delivers 285 MHz final speed. That is slow but I fear it may still be way to fast for DOS.
With cache off, the range translates well to the performance of 286 to 486. But without ISA slot, PCI sound will falter with no cache. So the whole thing would only work in the presence of an ISA card. And we are back on the KT133a. Sigh...

I got an ASUS A7A266 (using an ALI chipset 😳 ), it's capable of using a 66Mhz fsb on Athlons. But alas, no ISA 😖

If we forfeit ISA, why bother?
To me, the moment you give up ISA, you give up DOS. Without ISA you can't get slow enough for DOS. (I never figured out the SB compatibility settings of the Kinetiz Bios - maybe they work without cache on?)

If we we give up DOS, there is really no need to bother with the KT133A. All you need then is Win98se compatibility with a lower speed border of 200-300MHz. You have a lot to choose from with those requirements. All you need to decide is, if you want 3dfx AGP or not. But even if you decide for it, you can use the KT266A which is a very decent chipset, or the KT333, which I have never owned.

I got it working! I didn't use my mobile Barton though. I rather took a Geode, which is a 10.5x133 mobile Thoroughbred B.
Used FSB100 only, set voltage to 1.65V. With crystalCPU, I could lower the thing to 300MHz and ran 3DM2000 with it.

Instant speed changes in the range of 300 to 1050 MHz. Since cache can be disabled and ISA is available, this becomes an interesting DOS/WIN98 option.