AlexZ wrote on 2026-05-18, 14:51:

You should get Athlon XP 2500+ Barton (333FSB, AXDA2500DKV4D) with 11x multi and run it at 400 FSB to reach 2200Mhz. Then see i […]
Show full quote

You should get Athlon XP 2500+ Barton (333FSB, AXDA2500DKV4D) with 11x multi and run it at 400 FSB to reach 2200Mhz. Then see if it runs at 420-440FSB.

If you want higher OC, you get AMD Athlon XP 2600+ Barton (333FSB, AXDA2600DKV4D) with 11.5multi. You get 2.3Ghz at 400 FSB.

Athlon XP 3000+ Barton is the best silicon, but only useful if you want to do world records. More of collector's item.

Beware that TDP is dependent exponentially on voltage and linearly on frequency. You can easily exceed it by 50%.

The one in my photo below? If so I'll try with that one rather than the 3200+ version! The 3000 appears to be multiplier unlocked too!

So here we go.. some interesting observations:
The board is having fun with "high" FSB clocks. I checked some clocks directly at the CY28372 PLL and found the following:

- at 189MHz FSB the PCI clock is around 35MHz (no sweat) - system is stable no issues
- if you switch to 190MHz FSB and save, the PCI clock suddenly jumps to 48MHz and the system does not boot
- if you restart the system (hard power off + on) the PCI clock remains at 48MHz
- if you restart the system via the reset button, the system boots with 35MHz PCI clock, asks you to check the BIOS settings
- you can go to the BIOS, save the 190MHz again and et-voila, it boots just fine

I think ASROCK borked the BIOS very hard on this one. Fear not.. a REV 3.0 for 10€ is on it's way.. the poor A7N8X will still have to wait. The Rev 3.0 is actually the one I had back in the days.. so it'll feel more authentic for me 😁

Btw.: Car-Chase stutter was indeed a driver issue. Catalyst 7.7 is working just fine, no frame-drops anymore.

More time for tweaking and also going barton next week.. Hopefully I can get the water cooling installed then as well.

PcBytes wrote on 2026-05-20, 10:08:

I'm quite surprised to be still the only one with T-Bird on the score 🤣 - was T-Bird really that unpopular? I thought it'd be a good choice given it's kind of "free roam" in terms of FSB and multipliers.

I have a few. ill see what i can do with them in a few weeks.

Here's a run on Windows 98. GPU-Z has compatibility problems, so I had to use Everest for the GPU information.
6800 GT @ 450/1256-MHz

Updated!!!

I think I have no excuses...

The Barton XP-M 2400 is binned to run at 35W, but before I can OC and bench any of this, I first have to find a GPU* and northbridge cooler, and decide whether I want to try this on air (Zalman CNPS 7000CU clone) or water (240mm custom loop with 4000 RPM finger blenders) - since sub ambient is forbidden, I won't get to try out my stupid idea of tapwater cooling? Or does sub ambient mean the CPU itself is not allowed to go sub ambient? The tap water is probably around 10°C.

*unfortuntely, very few GPU coolers fit a Geforce 7900 GS AGP, so that might be an argument for integrating it a custom water loop - that way I could just ziptie some whatever cooler onto it ghetto style

Another question: does an AMD Geode NX count? It's a Thoroughbred-B that is basically equivalent to the mobile version, except it is binned to do 1.4 GHz at 1.25V or 1 GHz at 1.00V.
They might be really interesting in terms of achieved clocks.

NorsteinB wrote on 2026-05-21, 09:37:

Another question: does an AMD Geode NX count? It's a Thoroughbred-B that is basically equivalent to the mobile version, except it is binned to do 1.4 GHz at 1.25V or 1 GHz at 1.00V.
They might be really interesting in terms of achieved clocks.

Water with ice cubes is fine. Thats available to everyone. Yes the geode counts.

Watch out for condensation when running the cooler colder than ambient.

supercordo wrote on 2026-05-21, 16:53:

Water with ice cubes is fine. Thats available to everyone. Yes the geode counts.

I literally mean the water tap on my terrace 🤣, I'd be too scared of condensation if I cooled it down further, like the previous poster pointed out.

If I was using ice cubes, I'd fill a big bucket with those and the radiator, add water for perfect contact, then add salt to make the cubes melt and decrease the water temperature even further. Probably antifreeze needed in the actual cooling loop then.

But I definitely won't.

In fact, I think I'll go with strong air cooling for fairness, then afterwards make an attempt with tap water and post the frequency increase only, without benchmark scores.

Hobby Rocket Scientist from germany flooded his house in an attempt of Retro-OC

I'm getting there:

Jesus that's some serious OC.
How did you cool that Barton?

It's not even a good sample. DKV4D DTR (Desktop Replacement with 1.65V). Try to find an IQYHA Mobile Barton with 1.45V - they run like hell. 2.7-2.8-GHz range.
submerged water pump in a cooking pot. For the first few runs it's decent but after some time the water is getting pretty warm, then I have to put a bag of ice in the pot. Yes - watch out for condensation cause it will happen quickly.

I only got an IDYHA, but it's a 1.35V 35W model, so I got somewhat high hopes for it anyway.

Who knows. Give it a try. The 7900GS is much faster than my 6800GT. Good Luck!

The Sandman wrote on 2026-05-21, 18:32:

Hobby Rocket Scientist from germany flooded his house in an attempt of Retro-OC

I'm getting there:

What motherboard are you using, Sandman?

DFI LanParty Ultra B. Several times fixed by _digitalbath 😉
There are only two Socket A Boards that don't show their real name in CPU-z
DFI Infinity and Ultra B

Nice! I have been looking for a 7900 GS and they are impossible to find. let alone a 7950 gt.

NorsteinB wrote on 2026-05-21, 09:37:

Another question: does an AMD Geode NX count? It's a Thoroughbred-B that is basically equivalent to the mobile version, except it is binned to do 1.4 GHz at 1.25V or 1 GHz at 1.00V.
They might be really interesting in terms of achieved clocks.

I got a few Geode NX - those usually won't scale much. Tested a dozen and the best sample reached 2200Mhz at high volts. Geode NX is Tbred B binned for low volts and low consumption, not neccessarily binned for low leakage at high clocks.

Here's a score of mine which i did a few years ago. The goal was to prove FSB scaling in 3DM01, so this is not a maxed out score, but the cpu was kept at 2500Mhz. FSB is at 262MHz and the gpu is only a 9800Pro at stock frequency.

paradigital wrote on 2026-05-16, 08:54:

Recapped the A7N8X Deluxe, installed the Uber 1007D bios,

Please use a proper bios and not the pointless Uber. Choose at least the Trats one or (even better) use the newer bios made by digitalbath, preferably the ED55 one (https://biosbude.de/%21%20BIOS%20VAULT/Asus/A … .0/digitalbath/). Also mind that:

1. NF2 overclocking needs a lot of chipset/silicon lottery luck - board mostly doesn't matter, but chipset quality does
2. keep the FSB in sync (1:1) to the ram - that's always the fastest and most stable option
3. always use single sided ram when going for (very high) FSB at or above 250MHz
4. NF2 loves low latecy - so Winbond BH-5 as 2x256mb sticks running at 2-2-2-5 is the best option for highscores

And of course:
5. NF2 does absolutely need a mod bios, as it won't load faster/more stable FSB timings ("romsips" or "straps") with stock bios. So when using a FSB133 Cpu you won't hit high fsb without a modded bios.
6. even with modded bios some boards (looking at you asus...) will hit a fsb wall as chipset voltage isn't sufficient. So voltmods are needed. Other boards like DFI or abit got an option in bios.