I tried some newer drivers 81.98 fixes the problem but I am stuck at 75fps. Also 3dmark2000 is now working with no artifacts, but low score, about 10,000 3dmarks when I think it should be ~13,000. Not sure if it’s CPU which causes the low score. I get similar issue in 3dm01, about 10,000 points

CPU update - I’ve officially called time of death on the 2800+ Barton. I found chips on the edge of the die and the die also looks quite pitted. Comparing to the photos on the original eBay listing, these issues are all present so I am fairly convinced that the chip was already dead before it was shipped to me

My pic

Original pic from ad

I’ve just purchased a 2600+ Barton so I can try again. This one should be unlocked too - build week 11 of 2003 according to the sticker.

That's a real shame. These exposed dies were quite vulnerable to damage. Also I think AMD CPUs of this era had no thermal protection, so could very easily cook themselves if heatsink is incorrectly mounted.

That 2800+ Barton had die in a terrible state, I wouldn't buy it no matter what seller says. People need to stop looking for the lowest price but buy the best die for reasonable price.

Yep there is no doubt it’s in very poor condition. The 2600+ looks a lot better as far as I can tell.

RetroPCCupboard wrote on 2025-12-02, 05:28:

That's a real shame. These exposed dies were quite vulnerable to damage. Also I think AMD CPUs of this era had no thermal protection, so could very easily cook themselves if heatsink is incorrectly mounted.

Yes, they’ll fry themselves in no time if the cooler is removed! I remember seeing someone doing it for a video many years ago.

More tests with different drivers on the FX5900XT. I did the 45.23 inf mod to get the 5900XT to install on that driver version, and I’m getting much better 3dmark results now, and no artifacts or vsync issues in 3dm99 😬

3dm99 12140
2000 - 10085
2001SE - 9961

This is still with the 2000+ running stock speed. 2600+ should arrive tomorrow. Got a full refund on the 2800+, didn’t even have to return it!

I’m hoping to play some actual games this weekend 🤣

That 2800+ looked horrible so it made no sence to request return. If it makes no sence to return then I usually request destruction of the item to make it irreparable. With CPU with pins it's as easy as stepping on it.

With 5900XT you can play at 1600x1200, but old games usually do not scale user interface well. My PIII 900 with FX5600 is enough for games up to 2002.

The 2600+ arrived today. Die in perfect condition. Fitted it and got machine to POST no problem! BIOS just says “unknown CPU” but it seems to be fully working and seems fully unlocked! 😬

I’ve set the multiplier to 12.5 for now. I will need to remove the motherboard to pin-mod the cpu socket to enable higher multipliers

Some initial clock for clock comparisons with the Palomino 2000+

3dm99 12140 vs 12905 (133x12.5)
3dm99 12863 vs 13561 (141x12.5)

That extra cache seems to make a big difference in the low end stuff. 2MB texture rendering speed increased from 1562 to 1955 FPS for example.

It’s probably already proven elsewhere but I guess the cache compensates for the SD-RAM a bit; I think the performance between the CPUs would be closer if there was DDR memory on this board

Pin mod done, bios very unhappy though. I thought I’d done something bad and damaged the board or cpu as it would not post for ages. But I have figured out a procedure that works reliably to get it to post, as tedious as it is:

Turn off power on back of PSU
Remove cmos battery
Refit cmos battery
Set jumper on motherboard to auto configure the multipliers
Turn on power at back of PSU
Power on -> wait a min -> power off (no post)
Set the jumper for multipliers to manual
Power on. Hey presto. It works, and I can soft reboot. But power off and I have to do it all again.

So far I’ve got it running at 2133MHz and once it lets me post, it seems stable at stock 1.65V, 52C running prime95

New benchmark results at 2133MHz

3dm99 14489
2000: 12626
2001se: 11363

Hmm. Well if you have to go through that process to boot it, that doesn't sound very usable. Lol.

No, it’s not ideal! As I said I can soft reset (reset button / ctrl+alt+del) no issue but power cycling is a problem

Tried the Q11 and Q12 bios, no change. I might try another PSU just incase there’s something funny going on with it.

This wasn’t originally going to be a overclocking adventure, but I just can’t help myself

Been doing more tests. I can sometimes skip removing the battery in the startup procedure. I’ve had it up to 2400MHz but this speed is not stable at stock voltage, which I’m not planning to go over due to heat. The cooler is a fairly basic one.

I am still testing, but it looks like around 2200MHz is the stability limit. I need to play around with the FSB now to see if I can get any more performance of it. I put a new cooler on the northbridge as the original one was pretty non-existant

Best results on 3dm99, just over 15k now with 139 x 16, but it wasn’t stable in prime95, so i need to drop to a 15 multi and see how close to 150MHz FSB I can get.

I don't think you need to go that far. If higher multipliers don't work properly due to boot issues then 12.5 x 133 will be plenty fast for Windows 98.

When testing boards I just put the board on a shoebox on the ground and PSU next to it. Nothing will get shorted, it's pretty safe.

AlexZ wrote on 2025-12-06, 21:21:

I don't think you need to go that far. If higher multipliers don't work properly due to boot issues then 12.5 x 133 will be plenty fast for Windows 98.

When testing boards I just put the board on a shoebox on the ground and PSU next to it. Nothing will get shorted, it's pretty safe.

I’m seeing what I can do, not what is necessarily sensible or reasonable. Indeed if I can’t find a solution to this posting problem, I should revert it to stock multipliers

My last roll of the dice getting it working properly is patching the bios with bios patcher. I have run the patcher on the bios today and it updated the microcode plus made several other fixes, one which looked like it might be relevant to the issue I’m having. Will be flashing this patched bios later on. Wish me luck!

The board is on top of a slim cardboard box, 25mm thick. I’m still using the drives in the case underneath. Such a setup saved me taking the drives out. I had fully built the case. I hope i can reassemble it soon 👍

Bios successfully patched with the following updates;

And flash was successful! CPU now recognised. No change in POSTing behaviour though unfortunately

Today I had a chance to have a play around again

I went back to 100MHz FSB and discovered that the computer now POSTs normally up to 22x multi
It will not post without the trick described in the previous posts, at any setting above 120MHz FSB. 112 and 115MHz post but are unstable. 120 tries to post but I get a blank screen. This is more than I get at 124, 133 and up. The monitor just goes to sleep when I aim higher.

120 is the highest you can go with PCI = FSB/3 everything above that is PCI=FSB/4

So I am wondering what could cause this behaviour. It can be made to post at 133, 140 and 150 FSB with the trick mentioned in my first post and 133 and 140 are both stable and I can go to about 145MHz before I start to get any stability issues. I am wondering if there is an issue somewhere on the board where the PCI divider isn’t being set correctly. A bad solder joint perhaps?Does anyone have any ideas? I’ll have a poke around to see if I can find anything in the meantime

Decided to make another thread in general chat about the issue to try and get more eyes on it. Someone suggested the clock generator chip or crystal could be faulty. I’m going to be getting an oscilloscope and a post code analyser to do more tests

Meanwhile the reset trick continues to work for FSB>133MHz and it’s possible to boot the system normally using lower FSB settings in the region of 100 - 110 MHz without the trick.

I’ve been running stability tests using prime95. The limit seems to be lower than i first thought; around 2133MHz instead of closer to 2200, with this CPU on stock volts. I’ve got 3 sets of results which actually perform very close to one another.

133 x 16 - 14489 3dmarks / 33540 cpu 3dmarks
141 x 15 - 14915 3dmarks / 34614 cpu
111.8 x 19 - 14569 3dmarks / 33871 cpu (ram running async at 145MHz)

After hijacking someone else’s thread about the similar KT7A board, I made an interesting discovery

The computer will POST normally when set to 133 x 17 - 2266MHz - but at no other 133 setting

Based on my other tests, this is too fast for this CPU on stock voltage. But I can also now POST at 124 x 17 - 2108MHz but this is also not stable for some reason. This speed should be underclocking the buses and I’ve tested other FSBxMulti combinations around a similar overall clock speed and they’re all stable. I need to investigate further, I have a couple of theories:

1) PCI is running at FSB/3 instead of FSB/4 for some reason (not very likely, but i can’t test this at the moment until I get an oscilloscope)
2) 124MHz FSB is actually 130MHz FSB according to a datasheet for the clock generator I found. This would mean the CPU is running at 2210MHz and would explain the instability.

However this theory is also resting on 140MHz actually equalling 145MHz FSB - which has shown to be stable when I try it with 15x multiplier, which I don’t think it should be.

I made a table of the discrepancies of the datasheet vs the manual. Thanks to those who posted in my other thread, they helped me figure this out

Ok well after my overclocking adventures with the Athlon I decided to fire up the socket 7 system today. I’ve been trying to play a few late 90s games on the Athlon machine but I’m having issues with cutscenes not working - the monitor says it is “out of sync” when the cutscenes play on dungeon keeper and Hardwar. Not tried other games yet.

Not tried Hardwar yet on the Pentium machine as it really needs a 3D card, but dungeon keeper cutscenes seem to be working ok on that.

Has anyone tried a Kyro on a Pentium machine? They seem to be around GeForce 2 level. Overkill on a Pentium but there is one on eBay relatively cheap and it’s PCI

Edit - found the answer, the Kyro needs a good CPU it seems. I’ll keep an eye out for a voodoo of some kind

I’ve got the Pentium machine working nicely on all the dos games I’ve been trying. Having around 600k free base mem with all drivers installed. Currently running it with the Vibra16 while the AWE64 sits in the Athlon machine.

I had an issue on both machines with Dungeon Keeper in windows, freezing on the level select window. It works fine in dos on the Pentium though, if a little sluggish running 640 x 480.