i acquired a couple athlon xp barton 2500+ cpus (AXDA2500DKV4D): 166mhz fsb, 11x multiplier, 1.65v. what are the chances of getting these to run on the abit or any other kt7a board with the rom.by bios patcher? what about a barton 2500+ with a 14x multiplier?

stanwebber wrote on 2023-11-22, 17:03:

i acquired a couple athlon xp barton 2500+ cpus (AXDA2500DKV4D): 166mhz fsb, 11x multiplier, 1.65v. what are the chances of getting these to run on the abit or any other kt7a board with the rom.by bios patcher? what about a barton 2500+ with a 14x multiplier?

The chances are very good. If it's a late revision chip, unlock the multiplier with the bridge mod (omit this on early revision chips -- the datecode isn't exact, I think it's anything pre 0238). Then just use the pin mod guide. I recommend 16x133 for the 2500+.

If you want to use setmul in DOS with it, modify the bridge to turn it into a mobile (all revisions require this).

I'm having problems with SETMUL on a KT7A rev. 1.0. I tried two different CPUs (XP-M 2800+ Barton and a Athlon Mobile/Palomino) but each time I try to change the multiplier, the system just freezes.
I changed reg 55h/bit2 to 1, but no difference.
Both CPUs work fine on a ECS K7VZA and can be re-clocked on the fly with setmul.
What am I missing here? Is rev.1.0 not compatible with multiplier changes?

pyrogx wrote on 2023-12-03, 20:23:

I'm having problems with SETMUL on a KT7A rev. 1.0. I tried two different CPUs (XP-M 2800+ Barton and a Athlon Mobile/Palomino) […]
Show full quote

I'm having problems with SETMUL on a KT7A rev. 1.0. I tried two different CPUs (XP-M 2800+ Barton and a Athlon Mobile/Palomino) but each time I try to change the multiplier, the system just freezes.
I changed reg 55h/bit2 to 1, but no difference.
Both CPUs work fine on a ECS K7VZA and can be re-clocked on the fly with setmul.
What am I missing here? Is rev.1.0 not compatible with multiplier changes?

Had the same problem and never found the solution. Now I just change the multiplier in the BIOS.

From the Socket A boards with ISA of my collection (Abit KT7A-RAID V1.0, Abit KT7A V1.3, DFI AK74-EC, Epox EP-8KTA3, Gigabyte GA-7IXE4 and Legend QDI KinetiZ 7EV), I like the Abit KT7A most, both in features and performance. KT133 and KT133A without ISA don't make sense for anyone IMO. I would love to try out the Iwill KK266 or KK266Plus as they are supposed to be good overclockers, with reviews claiming north of 160MHz FSB on reasonable IO voltages (my theory is that they were never sold in Germany as I couldn't find one on eBay in Europe in 2 years. Also, I couldn't find any German reviews of it). On the other hand, the KT133A frequency record on hwbot was achieved on a KT7A at 177MHz, however at a crazy 4.4V IO voltage.
Very few sticks of PC133 will handle that speed. There's Tonicom PC166 Micro BGA memory that doesn't really do 166MHz according to reviews and it is hard to find (managed to score one years ago but looked for a long time). Besides that, Qimonda PC133 sticks seem to overclock well. In general, I recommend at least PC133CL2 modules like Hynix 7ns sticks. This way, one potential cause of instability is ruled out.
My system uses 2x256MB 7ns PC133 memory sticks, Windows 98SE, an Nvidia Quadro FX 3000 AGP (similar to an FX 5900 XT), a Barthon Athlon XP-M 2400+ AXMD2400FJQ4C IDYHA stepping at 12x133=1600MHz 1.25V, Aureal Vortex2 AU8830 PCI and Creative AWE 64 Value CT4520 ISA. I didn't have any problems with the memory but didn't try out more than 512MB because of Win98SE. While some had problems running FX series GPUs on the KT7A on Win98SE, my card has newer caused any problems! Similarly, a 3dfx Voodoo 3 or Voodoo 5 ran well without problems on my board. The Quadro has an external power connector and foregoes the limited AGP power the KT7A seems to have. The PSU is a newer than the capacitor plague with 350W and 20A on 5V. I never recapped my KT7A and all caps look immaculate, so I don't know if I got lucky, they used different caps (brand is Teapo and they appear to be decent) or the previous owner did a recap.
So far, I haven't tried out Setmul or the pinmod to enable higher multipliers. Weirdly, the Athlon refursed to work with the 12.5x multiplier on my board, so I use 12.0x instead.
The flexibility of this platform is awesome for late DOS games up to around 2003. Anything newer runs on Windows XP anyway, so it's my Windows XP PC's territory. I have a hard time deciding on whether to keep the Quadro in the system and use wrappers for Glide games or to replace it with a Voodoo 5. What is your opinion?

Trivia on the Athlon XP Mobile: I read somewhere that they were manufactured until mid-2005 and my particular example is from week 8 of 2005 which I thought is really late for socket A. That information can't be right, as there is a picture on cpuworld of such CPU from week 11 2007! Imagine how late it was produced, only to likely end up in a budget laptop. By that time, the 65nm process was used for other CPUs and 45nm came later that year. 130nm was ancient and the yields and process quality must have been awesome at that point. Who knows how well a 2007 chip will overclock? I guess even better than my 2005 which already is good. So when looking for an XP-M, take the one that was manufactured most recently.

Hi, thanks for the detailed post! I don't have much to add to your excellent advice, except that I too can not go above 512 MB (which is already more than enough). Using both a nVIDIA Geforce 4 in combination with Voodoo 3 PCI works great. The main issue is to find a PSU with a strong 5V rail (at least 25A), old ones have capacitor issues, new ones are weak on the 5V rail. If anyone has recommendations on this, let us know 😀

By the way, a capacitor may look great and be close to failure such as causing a short between ground and Vcc. You cannot verify it with a multimeter while it's on the MB, you need to desolder it and check with the capacitor test of your multimeter. In my experience recapping several boards, the capacitors near the CPU and the AGP port are the ones that go bad, many times with no visual hint at all. With a rare and precious motherboard like the KT7A, it's best to do a full recap. But if you've never recapped, buy a cheap motherboard and hone your skills on it. You don't want to destroy copper traces on the KT7A, it's very difficult to repair.

cde wrote on 2024-04-03, 11:20:

The main issue is to find a PSU with a strong 5V rail (at least 25A), old ones have capacitor issues, new ones are weak on the 5V rail. If anyone has recommendations on this, let us know 😀

I recommend 2 different PSUs that have 24-25A on 5V rail, if that's the minimum you require. They are expensive though:

-Corsair 650 cx650m
-EVGA Supernova 650 G6

Thanks!

I did some more stability testing on the KT7A Revision 1.1. I employed the game Flatout 2 for this once again, the game was configured at 1024x768x32bpp with texture quality set to very high. The CPU is an Athlon XP 2800+ (Barton - AXDA2800DKV4D) running at 2Ghz (15x133. This is a slight underclock, but I think it improves stability). The CPU's L5 bridge was shorted to turn it into a mobile Athlon and allow for the custom multiplier. The multiplier was set according to the Interactive Pin-Mod Guide by bridging socket pins on the underside of the motherboard. nVidia driver version used was 45.23. Chipset driver used was Via 4-in-1 version 4.35 with all options set to maximum performance during installation.

Here are my findings:

• One 512MB memory stick seems ideal for this board. 3 sticks totalling 1.5gb worked but made the system crash right away in mprime at a 133mhz FSB. I did not experiment further as 512MB is more than enough for Windows 98
• With all BIOS options set to max, the system is perfectly stable, except for "enhance chip performance" which causes Flatout 2 to crash to desktop during a race. I think this option is too aggressive for later CPUs, and perhaps Abit had this in mind for Thoroughbreds and not Bartons
• Flatout 2 was stable with both a GeForce 4 ti4200 and a GeForce FX 5800 Ultra. So there is in fact enough power delivery/stability on the AGP voltage rail
• The nVidia "AGP" tab in the driver properties (enabled via coolbits hack) shows that fast writes, SBA, and AGP 4x is enabled (and indeed, I maxxed the AGP out in the BIOS)

In conclusion, I finally found a good hardware combination for this board and hopefully settled the issue of the supposed instability and incompatibility of earlier revisions of the KT7A. Yes, they are stable, yes, they work with Bartons, yes they do high multipliers.

If anyone has any other videocards they'd like me to test, feel free to chime in.

EDIT (9/25/2024 23:30EST): Tested with a GeForce3 ti500, completed desert oil field race in Flatout2 with no issue. Tested with GeForce4 MX440 - incompatible with Flatout2 (required at least DirectX 😎, but Quake3 DEMO001 completed without issue. Scored 145fps vs. the ti500's 139FPS). Tested Voodoo3 2000, runs perfectly fine, ran Q3A DEMO001 fine. Everything so far seems perfectly stable with 1 ram stick.

Great! Need to take my KT7A-system from storage and start testing it again after a long time.

I'm actually playing with Asus A7V133-VM at the moment, it has KM133A chipset with integrated ProSavage. I was shocked when I got it working @166FSB with 3-2-3 memory settings. Sticks are Micron PC133 CL3 ECC. Never got KT133A that high, even close.

BTW, what is the easiest method to make an Athlon XP to mobile variant? To make that L5 bridge?

dj_pirtu wrote on 2024-10-15, 12:40:

Great! Need to take my KT7A-system from storage and start testing it again after a long time.

I'm actually playing with Asus A7V133-VM at the moment, it has KM133A chipset with integrated ProSavage. I was shocked when I got it working @166FSB with 3-2-3 memory settings. Sticks are Micron PC133 CL3 ECC. Never got KT133A that high, even close.

BTW, what is the easiest method to make an Athlon XP to mobile variant? To make that L5 bridge?

Cool, but you know you can also use high multipliers instead of high FSB? I have done 20x100 (instead of 133x15) and that worked fine.

For me, yes, bridge the appropriate L5 pads... Then mod the socket pins from the underside of the board to enable high multipliers and to set the multiplier. I am using 15x133 for 2ghz on a 2800+. It's a very slight underclock, but I have found that it is stable at 1.5V (with a significant temperature drop) on the modded desktop Barton.

I'm using conductive paint for this (after taping off the area around and leaving only the area I want painted).

Afterwards, test with Setmul. If you were successful, it will tell you it detected a mobile CPU. Otherwise, it will say that "only L1 can be disabled" or something to that effect (don't forget to run PCISET on the correct register if you want to actually change the multiplier beforehand, or it won't work).

If you are certain you bridged the correct L5 pads but still don't get good results then you have a very late manufactured superlocked Athlon XP.

I also use 3DMark2000 to test. On a stable system, the benchmark loop runs forever (custom benchmark - 1024x768x32bpp). On the Soyo SY-K7VTA PRO, I found I had to disable AGP 4x in the BIOS to get it to pass 3DMark2000.

With motherboard that uses SDR memory it's more than important to push FSB as high as possible, CPU is badly bottlenecked by the memory bandwith.

But, I did the L5 trick with pencil, second last bridge counting from the left so L5 is at upper left corner of the CPU. Now setmul works between x3 and x11 but I wanted x12 or even x13. Trying to google how to unlock higher multipliers but there's stuff of wiretricking etc. I think I don't need a wire trick for the mobo because I'm using setmul. I don't care what multiplier the system will boot. BIOS don't even have any multiplier settings.

BTW, I got 13255 from 3DMark01SE with Radeon X800Pro. CPU at 166x11=1826MHz. Radeon 9700 or 9800 can't handle 83MHz AGP bus speed.

Edit: got it! Both ends from L6 cut so maximum multiplier is now x19.