@Zuldan You need a AGP 7900GS or an AGP 7950GT ...they are hard to find but it should boost your score a bit 😜

I have only ever seen one AGP 7950GT on EvilBay and it was a few months back . .was going for an obscene price too but it was verified test and working.

I own a AGP 7900GS and can happily say it beats pretty much everything aside from the weird 3850 and 4670 AGP cards which I doubt are allowed for this little competition.

Its nice to see you having a blast !

Trashbytes wrote on 2026-07-22, 09:32:

@Zuldan You need a AGP 7900GS or an AGP 7950GT ...they are hard to find but it should boost your score a bit :P […]
Show full quote

@Zuldan You need a AGP 7900GS or an AGP 7950GT ...they are hard to find but it should boost your score a bit 😜

I have only ever seen one AGP 7950GT on EvilBay and it was a few months back . .was going for an obscene price too but it was verified test and working.

I own a AGP 7900GS and can happily say it beats pretty much everything aside from the weird 3850 and 4670 AGP cards which I doubt are allowed for this little competition.

Its nice to see you having a blast !

Hey Trashbytes! I am having a blast 🎇

I do have that Gainward 7800GS (N71) that I’m trying to fix (Gainward BLISS 7800 GS+ Golden Sample - Broken) but other than that I’m not ready yet to drop big bucks on a 3850, which is allowed 😉

Now why aren’t you joining the competition? I figured this would be right up your alley with your love for AGP cards.

zuldan wrote on 2026-07-22, 09:50:

Hey Trashbytes! I am having a blast 🎇 […]
Show full quote

Trashbytes wrote on 2026-07-22, 09:32:

@Zuldan You need a AGP 7900GS or an AGP 7950GT ...they are hard to find but it should boost your score a bit :P […]
Show full quote

@Zuldan You need a AGP 7900GS or an AGP 7950GT ...they are hard to find but it should boost your score a bit 😜

I have only ever seen one AGP 7950GT on EvilBay and it was a few months back . .was going for an obscene price too but it was verified test and working.

I own a AGP 7900GS and can happily say it beats pretty much everything aside from the weird 3850 and 4670 AGP cards which I doubt are allowed for this little competition.

Its nice to see you having a blast !

Hey Trashbytes! I am having a blast 🎇

I do have that Gainward 7800GS (N71) that I’m trying to fix (Gainward BLISS 7800 GS+ Golden Sample - Broken) but other than that I’m not ready yet to drop big bucks on a 3850, which is allowed 😉

Now why aren’t you joining the competition? I figured this would be right up your alley with your love for AGP cards.

Honestly I just don't have the free time to throw in the hat or the free space, actually don't have enough free space right now to even store all the parts I have so most of them have been moved to a storage lockup. I did need to downsize a bit so most of my pre P2 parts moved to a new home with a lad who is right into tinkering with 286/386 and 486 parts.

I am still working on dual X99 side project but its far too modern for this forum, but it does involve a pair of 22 core 2699A V4 CPUs and forcing them to run fast. (They really don't want to 🤣 )

Ill get back to Athlon XP eventually as I have a dual Athlon MP system in parts that I would love to put together and overclock, but it might be a while. So for now I get to enjoy seeing you battle it out here.

Was testing with my NF7 v2.0 board and ended up corrupting the Windows install. Because of this, I am going to go ahead and use my 32GB ACARD DDR2 SATA drive. It has a CF slot for backup. That way, when the Windows install gets corrupted again, I can just hit the restore button and it will be back up in running in a matter of minutes.

I really wish that somebody had made 8GB ECC DDR2 (not registered or fully buffered) as the drive theoretically supports 64GB.

cyclone3d wrote on 2026-07-24, 04:23:

Was testing with my NF7 v2.0 board and ended up corrupting the Windows install. Because of this, I am going to go ahead and use my 32GB ACARD DDR2 SATA drive. It has a CF slot for backup. That way, when the Windows install gets corrupted again, I can just hit the restore button and it will be back up in running in a matter of minutes.

I really wish that somebody had made 8GB ECC DDR2 (not registered or fully buffered) as the drive theoretically supports 64GB.

Interesting device. I just use a Kingston 240GB SSD. On my Socket A overclocking adventures its always the registry that dies. So once the System is in my desired state, i just copy the registry files into a backup folder and use them to restore my OS if it dies.

I use SATA ssds which I can plug into my main machine and create a backup there. USB SATA adapter does the job. Macrium Reflect free v7.x is fast (10 mins at most) for a simple win xp install. I even split my disk into multiple partitions, so I can restore only the OS (c:\) and keep installed programs (d:\) and tools (e:\) to make restore even faster.

If you also save your benchmark results and screenshots on d: or e: they will survive the restore too.

Bonus:
If you encounter the cpu-z freeze bug with versions >1.59, you can also restore the old image and it’ll work again.

Tzk wrote on 2026-07-24, 08:47:

I use SATA ssds which I can plug into my main machine and create a backup there. USB SATA adapter does the job. Macrium Reflect […]
Show full quote

I use SATA ssds which I can plug into my main machine and create a backup there. USB SATA adapter does the job. Macrium Reflect free v7.x is fast (10 mins at most) for a simple win xp install. I even split my disk into multiple partitions, so I can restore only the OS (c:\) and keep installed programs (d:\) and tools (e:\) to make restore even faster.

If you also save your benchmark results and screenshots on d: or e: they will survive the restore too.

Bonus:
If you encounter the cpu-z freeze bug with versions >1.59, you can also restore the old image and it’ll work again.

Yeah, similar process here: Something dead? USB-SATA -> do what's needed -> back into the system.

Do you use an IDE to SATA adapter or onboard SATA or PCI adapter? I landed on a SII3114, as it allows me to switch motherboards and chipsets with the same windows XP install, as the SATA controller remains the same.

Depends. For quick tests I use ide/sata adapter as XP boots off most IDE ports without drivers. For everything else I bought a 10€ SII3512 pci card which is almost the same as 3112/3114/3514. on some boards (a7n8x, DFI, abit) I use the onboard sata controller.

I flashed the pci card with the latest non-raid bios for better compatibility.

I use an IDE DOM.

I tell ya what, the digitalbath BIOS has been a pain to get initially stable enough to install XP. Still have to figure out why it likes to freeze on POST on a reboot.

I've never messed with a BIOS with added options like this before so it is taking a bit to figure it out.

@ _digitalbath - I'm having trouble getting anything stable with the NF7_[B619XT-3]_B4 BIOS. I am using a T-bird and TCCD RAM.

What do all the extra settings default to (Hangs on POST on a reboot and have to reset BIOS via jumper to get past POST on a cold boot)

What should I be trying?

The stock BIOS is stable up to 220Mhz fsb and this one seems to behave in the same unstable way no matter the fsb. I've gone all the way down to 166.

Edit: Doing some research, this may be entirely due to me using a T-Bird CPU. Trying a different BIOS that may behave better.. Trying to get over 222Mhz fsb for benching (ideally 234 or higher so I can also have max stable CPU speed)

The problem with the modded BIOSes are the modded romsips were made for higher frequencies. Using them at default FSB clocks may cause instabilities. You can test in memtest. You may get no errors but random frezzes at some multies. To prevent this issue, I've included to the B4 Revision BIOSes an automatic romsip value 68h. It sets a higher value for lower clocks and it helps to get a stable system. Default BIOSes uses also higher 68h values. Modded BIOSes uses lower values for higher clocks. You can skip the Auto function by selecting "default" value in the romsips 68h setting. Maybe worth to test it at the default value.
TREF, tRC, tRFC and Alpha Timings are dram Timings. For TCCD, I would suggest:
TRC: 11-13
TRFC:14-16
TREF: 2500-3600
Alpha: 4-5-6-x-x-x-x or higher

Drive strength: auto or 2-4 if you prefere manual settings
Slew rate: auto or 7-8 for manual settings. Abit boards like value 7

Romsip table : you can leave it at auto for the table in BIOS. The DFI tables may help in OC.

A hanging system just before the boot might be caused by a wrong setting in the added settings. I would suggest to do a longer clear CMOS. Somewhere like 5 minutes. Then reflash the BIOS. Maybe some of the CMOS values are not cleared propperly. I had this issue on few boards. You can see it in the BIOS when the default values for the "new" settings are not at their default values after a clear CMOS.

You also could try a different BIOS Version. Merlins ED could be a solution.
A responce would be nice as it helps to improve the compability.

Edit.
Thunderbird and Palomino CPUs will limit the FSB clocks to 200 - 230MHz.
Thoroughbred B and Barton CPUs do fine at high clocks.

Ok, cool. I may try that later. I am back to at least 220Mhz fsb stable with the D26 Black Mantaray CPC On BIOS in the attached .zip Testing RAM timings to see if I can get 225Mhz or higher stable. If that doesn't help, I will try the same CPC Off BIOS.

I also tried the trats D27_619R2 and the d26-blackmatarays-xt BIOSes and was seeing the same exact issues as the NF7_[619XT-3]_B4 BIOS.

With the help of ChatGPT it seems like it probably has to do with the CPU interface tables on the newer modded BIOSes being made for CPUs newer than Thunderbird

I used the mantaray BIOS for a while. I liked it! It hast simmilar romsips like the 619XT BIOS (origin is DFI Beta BIOS). So this good! Important for the Mantaray BIOS is, you use the Interface normal settings. At Interface aggressive, the BIOS use the Nvidia default sips and you will limit at your 220MHz.

See at my edit in the previous Post. Thunderbird CPUs will limit the FSB. They were not made for higher frequencies. I even tried to get higher clocks by changing CPU timings in the SIPs. It may help but nothing for a serious setup.

Lower multies like 5 and 6 are slower in CPU timings, maybe you will have more luck with them.

Got up to 232Mhz fsb with the T-Bird and NF7 v2.0 and then while the system was idle, it proceeded to most likely have the PS_ON control transistor die. Because of that, I am going to go ahead and post my highest score attained so far - 16432. I was trying for 234fsb but it kept locking up during the 3DMark 2001SE Lobby benchmark. Could be that that transistor was starting to go out.

I had backed down to 232 and tightened the RAM back up to 2-2-3-11 and got a slightly higher score but didn't have a chance to screenshot that before the board decided to put itself out of comission.

This is with the D26 Black Mantaray CPC On BIOS.

Really impressive score, especially for a Thunderbird!

Unfortunately, I can't give you my spot on the leaderboard so easily 😉

I am still working through some AGP oddities on this system, to the point that I'm now modifying drivers to make the damn thing boot properly. If there's a brownie point leaderboard, I have the WinDDK toolkit installed on my Precision 410 with 2x P3s, as the NT Kernel debug host, to debug the system boot routine. Gotta have a dual rig mending a dual rig 😀

Still need to get around to working out what wires to solder to get back to 15.5x. Unfortunately, my attention is focused on trying to wall mount my retro PCs and finding a better KVM switch solution.

cyclone3d wrote on 2026-07-25, 20:14:

If that doesn't help, I will try the same CPC Off BIOS.

I wouldn’t bother with cpc off. Usually you need about 15-20Mhz higher FSB to even match cpc on performance. Especially with TBird and Palomino (those are usually fsb limited) cpc on is your best option.

Regarding your issues with digitalbath bios:
The extra timings are applied after post, so booting at safe clocks and raising the FSB in the OS is a reasonable approach. Of course this includes the ability to post at all…

A big todo for those mods is to find a way to apply the custom settings as early as possible in the boot process. We haven’t managed that till now.

cyclone3d wrote on 2026-07-26, 04:32:

Got up to 232Mhz fsb with the T-Bird and NF7 v2.0 and then while the system was idle, it proceeded to most likely have the PS_ON control transistor die. Because of that, I am going to go ahead and post my highest score attained so far - 16432. I was trying for 234fsb but it kept locking up during the 3DMark 2001SE Lobby benchmark. Could be that that transistor was starting to go out.

I had backed down to 232 and tightened the RAM back up to 2-2-3-11 and got a slightly higher score but didn't have a chance to screenshot that before the board decided to put itself out of comission.

This is with the D26 Black Mantaray CPC On BIOS.

Nice Thunderbird score!!

Thanks all.

@geordiepingu - I am really hoping to get over 16500 with this CPU.

I am now getting ready to swap out the apparent dead transistor. The original part is sSN, and a direct replacement is K72, which I have harvested 10 off a dead, much newer Dell board next to the DisplayPort.

Sounds like most / all DisplayPort ports use the same type of setup.

Edit: the transistor I though was the problem is not. Been trying to track down what is going on.

Initial symptoms - locked up and would not power down. Powered down via PSU switch.
Turned PSU back on and powered on via the power button. This led to no POST as well as the green led on the motherboard pulsing and also the cooling fans speed pulsing up and down.
Was testing voltages and stuff and the PS_ON voltages were being wonky.

After probing things, it acted like the transistor was not reading properly.

However, the motherboard was now powering up as soon as I turned power on on the PSU. Tried another known good PSU with the same results.

Transistor has been removed and was determined good (looks like it is possibly just monitoring PS_ON). The PS_ON signal is coming from somewhere else. Working on looking at what is happening with the power button and also other voltages when the power button is pressed and when the PSU auto comes on and the Winbond Super I/O chip... uggghhhh.

This board was sold to me "AS-IS" so maybe it was doing this before and after is sat for however long it started working again for a while.

Gave up on getting the nf7 2.0!for now.

Setting up the EPOX EP-8VTAI (later renamed to 8KRAI Pro). The chipset is AMD KT880. BIOS is weird and I will probably see if I can mod it to at least unlock hidden features.

Edit: Used Modbin6 2.04.03 to unhide all useable options. Defaults are in ()

(Advanced BIOS Features)
CPU L2 Cache ECC checking (disabled)
APIC Mode (enabled)

(Advanced Chipset Features -> DRAM Clock/Drive Control)
DRAM Clock (unhid 100Mhz) - more memory dividers are always useful. The options are 100, 133, 166, 200. They set 1:1 when running those bus speeds and I am guessing up to the next major bus speed. Then the dividers auto change I think. Still need to do more testing.

(Integrated Preipherals -> VIA OnChip PCI Device)
USB Device Function (disabled) - no idea what this does

(Integrated Peripherals -> Super IO Device)
Infrared Port Select (unhid sub options)

(Power Management Setup)
Power Management Option(unhid Disabled) - This may help with overclocking
Suspend Mode (unhid 4 seconds - 40 seconds)
Ac Loss Auto Restart (unhid On)

(PnP/PCI Configurations)
Resources Controlled By (auto)
IRQ Resources (can set IRQs to PCI device or reserved)
PCI/VGA Palette Snoop (disabled)

(Power BIOS Features)
Current Voltage (unhid 1.075v - 1.375v) - yay, we can undervolt even more. Useful for the super low voltage mobile parts. The Geode NX lowest stock voltage is 1.0v