Since this morning, have upped the CPU to 450MHz and tightened the timings back to how they were before. My memtest floppy has stopped working (or the floppy drive) so I can’t check the RAM again. But I’m still getting into windows and doing benchmarks no problem.

3577 3d marks (forgot to take a pic) and 7304 CPU marks. The 3d score is about 5% higher than 83 x 5.5

In dos, my results are a little worse than the benchmarks I made back in Jan this year

Current

Earlier this year

I seem to have lost some FSB somewhere. Not sure where it’s gone 😅 it sometimes shows a 99MHz FSB but I’ve not seen it over 100 since I’ve been playing with it again. Weird! But this will stop me hitting 600Mhz if I can’t find the cause

A small snag but something I was expecting - I have lost a lot of base mem by having the IDE card installed; 568k with cd and mouse driver. I was at around 600k before

I was expecting to lose some from the upper men, but It’s taken 11k from the total base memory and well, I now only have 629k instead of 640k. I searched on here and this is apparently something that I should have expected. The memory map on checkit says there is no data here. Confirmed with debug. What’s it doing??

The reserved area starts from 9d3f:0010. First hex value is set as 0B, then nothing. It doesn’t always start at the same place between boot-ups

Today I got 256mb working. Swapped to a single (2 rank) 256mb stick from the previous setup of 2 x 128mb (2ranks each, 4 ranks total) only issue is I’ve stolen this RAM from the Athlon and I’ll have to replace it!

No update on the socket 7. Still no new CPU. I hope it turns up as I bought it a good while ago now

I was tempted into buying another CPU for the Athlon build. The Athlon build has been a bit of a disaster (certainly financially!) as I don’t seem to be able to decide what level of performance I will be happy with.

On the GPU side of things all is well. But the motherboard CPU and RAM are changing again!

Reminder of timeline so far:

After I bought the socket 7 I also started building an Athlon system as I knew the socket 7 would not do everything I wanted.

I decided on a kt133A based system for backwards compatibility (ISA slot)
I bought a Soltek SL75-KAV but it had nothing else with it, not even an I/O shield.
I needed to buy all of the components to build the system individually (£££). Unless I can find a donor!
I found a donor. MESH case, Athlon-ready PSU, ASUS A7v333 board, Palomino 2000+ Audigy 1 and fx5900xt - almost everything I needed. This was easily the best deal in the whole build.

I wanted to run a Barton on the kt133A, bought a 2600+ and did the pin multiplier mod and patched the bios to get it to work. Had loads of issues, tried a 3000+, couldn’t get it to work reliably. Decided to give up.

I then put the 3000+ in the a7v333. Wow these ddr boards are much faster, plus it’s stable. I will build my system around this board instead and sell the kt133A. Swapped everything over. Then the problems start. No sound. No usb 2.0 controller. Half of the PCI cards I tried didn’t work (including the Audigy and an Audigy 2) something very wrong with the pci subsystem. Very annoying as otherwise the board works perfectly!

So by now I have 2 Athlon systems. One is broken in one way and one is broken in another. So I decided to make one good system. I put the kt133A system back to stock and ran it with the palomino 2000+. This of course now runs perfectly fine but it’s a bit dull. All of the other components have sat quietly in storage. I was tempted to try a recap of the a7v333 but advice on here was not to bother as the issue is probably within the southbridge.

An uneventful 7-8 months have gone by running with the 2000+ now, I’d been on the lookout for a mobile Athlon. Absolutely none have come up for sale (in the UK) and i figured I would have to accept having a mediocre Athlon build for the foreseeable. But not so!

By the way. Here is my computing area. Realise I don’t put enough pics up of the builds I’m doing. It is a bit chaotic at the moment and the xp machine is missing (on the floor out of shot)

As you can see it’s somewhat chaotic at the moment while I’m in the middle of doing stuff to the Athlon and the socket 7 build. The socket 7 desktop normally sits under the desk while the silver/black Athlon tower and missing-from-picture all-black matx i7 sit side by side on the left.

Some chipsets are a bit picky with conflicts on the PCI slots. Have you tried moving the audigy to different slots?

As for work area, yes, it's chaotic... looks like heaven. 😀

Yeah I tried everything. Different slots, different cards, all working in other machines

Lots of devices just didn’t show up in the device list at the boot screen. Tried different bioses. Nothing. Even with nothing plugged in the board apart from a gpu, there were several devices present on the board missing from the list. One time the ide controller went missing but it still booted 😅 I could’ve managed if a sound card had worked. The onboard one also failed 🙈

So with this new CPU. I decided to give the KT133A one last try with a Barton CPU with a new approach; see if I can get it to clock up in software. It should be possible with an XP-M right?

Installed the new CPU. Changed the multiplier select jumper on the board to “auto”

The new CPU fired up at 1600MHz - this is what the multiplier jumpers were actually set to (12x 133) so not sure if the auto/bypass setting actually works.

Now to try and increase it. The correct multiplier for this CPU is 16x (2133MHz)

The setmul thread on here was super helpful in achieving this. The KT133A requires an additional step to enable multiplier changes using a tool called pciset. I tried without this step and the system just froze

SetMul - Multiplier control for VIA C3 / AMD K6+7+8 Mobile / Cyrix 5x86

Test run after running the pciset command

Success!!

I just followed the instructions and added the appropriate lines to autoexec.bat so it applies the settings on boot

It works! And it’s stable AND I don’t have to mess around with the motherboard to get it to boot every single time 🥳

Some overclocking then happened!

A few minor issues became apparent

As far as I can tell, multiplier is locked above 16x. Trying to set any higher just results in a setting of 16x. It’s quite a late CPU, not sure if that has any bearing on the situation or not. Would an earlier xp-m 2500+ have the same issue?? If so, I would have been limited to 1867MHz on this board without modifying the CPU (if it’s even possible?)

The FSB settings are quite temperamental. Again using software seems to be the way to go. The bios has “redstorm overclocking” to go up in 1Mhz increments. If I try and shortcut it and start at 140 FSB or 150 FSB using the dip switches, I get no post (at 150) and if I use 140 + bios overclocking it’s unstable at anything above 140. But starting at 133 and going up from there in software, I can go up to around 149-150 before I get errors in memtest86+

I ran some benches, 3dmark99: 16,000 3dmarks and 37000 CPU. This is still about 10% behind the results I got with the stock 3000+ Barton in the a7v333

So it seems I’ve topped out this board. I half expected this and there is more to come

FullYes wrote on 2026-09-11, 18:19:

So with this new CPU. I decided to give the KT133A one last try with a Barton CPU with a new approach; see if I can get it to cl […]
Show full quote

So with this new CPU. I decided to give the KT133A one last try with a Barton CPU with a new approach; see if I can get it to clock up in software. It should be possible with an XP-M right?

Installed the new CPU. Changed the multiplier select jumper on the board to “auto”

The new CPU fired up at 1600MHz - this is what the multiplier jumpers were actually set to (12x 133) so not sure if the auto/bypass setting actually works.

Now to try and increase it. The correct multiplier for this CPU is 16x (2133MHz)

The setmul thread on here was super helpful in achieving this. The KT133A requires an additional step to enable multiplier changes using a tool called pciset. I tried without this step and the system just froze

SetMul - Multiplier control for VIA C3 / AMD K6+7+8 Mobile / Cyrix 5x86

Test run after running the pciset command

The attachment IMG_9850.jpeg is no longer available

Success!!

I just followed the instructions and added the appropriate lines to autoexec.bat so it applies the settings on boot

The attachment IMG_9852.jpeg is no longer available

It works! And it’s stable AND I don’t have to mess around with the motherboard to get it to boot every single time 🥳

I also built a KT133A build recently with GA-7ZXE board, but I used a regular Athlon XP 2500+. My board has no overclocking settings whatsoever, so I had to modify bridges to change multiplier to 13,5x in order to keep it at 1.8GHz, but other than that it worked just fine with the CPU out of the box. I chose to not overclock it as I thought that 133MHz SDR memory might be too much of a b0ttleneck.

I did have to disable AGP sideband addressing with Rivatuner to make it work at 4x and there were some DMA related crashes until I installed a third-party PCI latency patch.

FullYes wrote on 2026-09-11, 19:01:

The FSB settings are quite temperamental. Again using software seems to be the way to go. The bios has “redstorm overclocking” to go up in 1Mhz increments. If I try and shortcut it and start at 140 FSB or 150 FSB using the dip switches, I get no post (at 150) and if I use 140 + bios overclocking it’s unstable at anything above 140. But starting at 133 and going up from there in software, I can go up to around 149-150 before I get errors in memtest86+

I ran some benches, 3dmark99: 16,000 3dmarks and 37000 CPU. This is still about 10% behind the results I got with the stock 3000+ Barton in the a7v333

So it seems I’ve topped out this board. I half expected this and there is more to come

Have you tried 3DMark 2001 SE? I also have a FX 5900XT so it would be interesting to compare results, I got 10197 points. In another thread people were telling me the GPU is too powerful for a KT133A build as it's not going to be useful for games after 2001, but I don't really buy it. 😀

I also did some benchmarking with a nForce2 board to try to determine just how big of a bottleneck the SDR memory is, and it looks like it causes about 15-20% performance penalty in 3DMark and UT2003 Demo benchmark.

I’ll have a look. Done a few Benchmarks in 2001se today but not with the new cpu or with the o/c. I seem to recall the 2000+ being around 10,000 in 2001SE. The driver version matters quite a bit. I’m running inf modded 45.23 drivers. I’ve not had any issues with the AGP side of things. I am not sure if side band addressing is enabled or not

I had an a7n8x-x back in period and that absolutely flew. I seem to recall memory bandwidth being pretty much the theoretical numbers quoted on the sticks

The kt133A memory bandwidth seems to be around 500MB/s. The a7v333 around 800 (at 333fsb) I’m sure the nf2 boards were up around 2Gb/s. I would get an nf2 based system but I know they are bad for backwards compatability

Mike_ wrote on 2026-09-11, 20:55:

I also built a KT133A build recently with GA-7ZXE board, but I used a regular Athlon XP 2500+. My board has no overclocking sett […]
Show full quote

FullYes wrote on 2026-09-11, 18:19:

So with this new CPU. I decided to give the KT133A one last try with a Barton CPU with a new approach; see if I can get it to cl […]
Show full quote

So with this new CPU. I decided to give the KT133A one last try with a Barton CPU with a new approach; see if I can get it to clock up in software. It should be possible with an XP-M right?

Installed the new CPU. Changed the multiplier select jumper on the board to “auto”

The new CPU fired up at 1600MHz - this is what the multiplier jumpers were actually set to (12x 133) so not sure if the auto/bypass setting actually works.

Now to try and increase it. The correct multiplier for this CPU is 16x (2133MHz)

The setmul thread on here was super helpful in achieving this. The KT133A requires an additional step to enable multiplier changes using a tool called pciset. I tried without this step and the system just froze

SetMul - Multiplier control for VIA C3 / AMD K6+7+8 Mobile / Cyrix 5x86

Test run after running the pciset command

The attachment IMG_9850.jpeg is no longer available

Success!!

I just followed the instructions and added the appropriate lines to autoexec.bat so it applies the settings on boot

The attachment IMG_9852.jpeg is no longer available

It works! And it’s stable AND I don’t have to mess around with the motherboard to get it to boot every single time 🥳

I also built a KT133A build recently with GA-7ZXE board, but I used a regular Athlon XP 2500+. My board has no overclocking settings whatsoever, so I had to modify bridges to change multiplier to 13,5x in order to keep it at 1.8GHz, but other than that it worked just fine with the CPU out of the box. I chose to not overclock it as I thought that 133MHz SDR memory might be too much of a b0ttleneck.

I did have to disable AGP sideband addressing with Rivatuner to make it work at 4x and there were some DMA related crashes until I installed a third-party PCI latency patch.

FullYes wrote on 2026-09-11, 19:01:

The FSB settings are quite temperamental. Again using software seems to be the way to go. The bios has “redstorm overclocking” to go up in 1Mhz increments. If I try and shortcut it and start at 140 FSB or 150 FSB using the dip switches, I get no post (at 150) and if I use 140 + bios overclocking it’s unstable at anything above 140. But starting at 133 and going up from there in software, I can go up to around 149-150 before I get errors in memtest86+

I ran some benches, 3dmark99: 16,000 3dmarks and 37000 CPU. This is still about 10% behind the results I got with the stock 3000+ Barton in the a7v333

So it seems I’ve topped out this board. I half expected this and there is more to come

Have you tried 3DMark 2001 SE? I also have a FX 5900XT so it would be interesting to compare results, I got 10197 points. In another thread people were telling me the GPU is too powerful for a KT133A build as it's not going to be useful for games after 2001, but I don't really buy it. 😀

I also did some benchmarking with a nForce2 board to try to determine just how big of a bottleneck the SDR memory is, and it looks like it causes about 15-20% performance penalty in 3DMark and UT2003 Demo benchmark.

Here are some results for you:

Just did a quick run at current o/c

Nice comparison in the result browser. The ~40% increase in clock speed gives (at best) 33% more fps, but the average is maybe 20%

Interestingly there is no performance gain in game 4 which suggests it’s GPU limited?

FullYes wrote on 2026-09-11, 21:37:

I’ll have a look. Done a few Benchmarks in 2001se today but not with the new cpu or with the o/c. I seem to recall the 2000+ being around 10,000 in 2001SE. The driver version matters quite a bit. I’m running inf modded 45.23 drivers. I’ve not had any issues with the AGP side of things. I am not sure if side band addressing is enabled or not

I had an a7n8x-x back in period and that absolutely flew. I seem to recall memory bandwidth being pretty much the theoretical numbers quoted on the sticks

The kt133A memory bandwidth seems to be around 500MB/s. The a7v333 around 800 (at 333fsb) I’m sure the nf2 boards were up around 2Gb/s. I would get an nf2 based system but I know they are bad for backwards compatability

That's actually exactly the board I used for benchmarking nForce2. 😀 I needed a cooler for my KT133A build and somebody was selling that board, cooler, Athlon XP 2600+ and memory for 20 euros including postage, so it was cheaper than buying just the cooler. 😁 But as you said, it's not as good for retro use. For starters, you don't get an ISA slot...

I'm also using 45.23 drivers, though I just forced Windows to use drivers for regular FX 5900, as it's based on the same chip. As for the AGP issue, it could be that your BIOS disables sideband addressing, or it might just be a board-specific bug in my case.

FullYes wrote on 2026-09-11, 22:28:

Here are some results for you: […]
Show full quote

Here are some results for you:

Just did a quick run at current o/c

The attachment IMG_9857.jpeg is no longer available

Nice comparison in the result browser. The ~40% increase in clock speed gives (at best) 33% more fps, but the average is maybe 20%

The attachment IMG_9858.jpeg is no longer available

Interestingly there is no performance gain in game 4 which suggests it’s GPU limited?

Thanks. It looks like KT133A is not as badly memory bottlenecked as I thought as you still got 20% bigger score with 30% higher clock. Although it's surprising that you got more points with Barton at 1.667GHz than I got at 1.8GHz... and I'm using the same drivers. What is the 3DMark 2001SE build you are using?

Here's my result for Athlon XP 2600+ (1.917GHz, unfortunately multiplier-locked, so I could not change it to match clock speed) on A7N8X-X and FX 5900XT. Note that the GPU shows as regular FX 5900 as I forced Windows to install drivers for that one.

I ran my A7N8X-x for a good 4-5 years as my main rig, I just did not want to part with it. Eventually paired with an x1950GT. My main issue with it was the CPU. I ran it with a thoroughbred 2800+ (2250MHz) ran hot and zero o/c headroom. If I had known a bit more about the xp-m situation I would’ve replaced it

I will delve into the settings to see whether sideband is enabled or not.

I have PC133 CL2 memory. The timings are as tight as they’ll go, in all 3 of those benchmarks. It seems to get better results with 2 matching sticks (4 ranks)

I believe it’s build 330 of 2001se but i will check. What are the fps scores in your game tests?

I can try and match the speed of my machine with your Barton 2600+ for a proper comparison? Also let’s compare GPU clocks

FullYes wrote on Yesterday, 10:25:

I ran my A7N8X-x for a good 4-5 years as my main rig, I just did not want to part with it. Eventually paired with an x1950GT. My […]
Show full quote

I ran my A7N8X-x for a good 4-5 years as my main rig, I just did not want to part with it. Eventually paired with an x1950GT. My main issue with it was the CPU. I ran it with a thoroughbred 2800+ (2250MHz) ran hot and zero o/c headroom. If I had known a bit more about the xp-m situation I would’ve replaced it

I will delve into the settings to see whether sideband is enabled or not.

I have PC133 CL2 memory. The timings are as tight as they’ll go, in all 3 of those benchmarks. It seems to get better results with 2 matching sticks (4 ranks)

I believe it’s build 330 of 2001se but i will check. What are the fps scores in your game tests?

I can try and match the speed of my machine with your Barton 2600+ for a proper comparison? Also let’s compare GPU clocks

I only have PC133 CL3 memory, so that explains at least part of the difference. Though if that's the only reason, it's a bigger difference than I would have expected. AGP sideband addressing shouldn't have measurable effect, as even going from AGP 1x to 4x caused a 3% performance increase. Everest Home Edition says that FX 5900 XT's core and memory clocks are 300MHz and 350/700MHz, which is lower than it should be for that GPU. I guess I'll have to investigate this...

EDIT: Apparently it was just running at 300MHz when running 2D software on desktop. It should be 400MHz on games.

I'm also using build 330 of 3DMark 2001 SE, here's a more detailed result. And yeah, it would be nice to compare to your results at the same clock speed. 😀

Mike_ wrote on Yesterday, 13:00:

I only have PC133 CL3 memory, so that explains at least part of the difference. Though if that's the only reason, it's a bigger […]
Show full quote

FullYes wrote on Yesterday, 10:25:

I ran my A7N8X-x for a good 4-5 years as my main rig, I just did not want to part with it. Eventually paired with an x1950GT. My […]
Show full quote

I ran my A7N8X-x for a good 4-5 years as my main rig, I just did not want to part with it. Eventually paired with an x1950GT. My main issue with it was the CPU. I ran it with a thoroughbred 2800+ (2250MHz) ran hot and zero o/c headroom. If I had known a bit more about the xp-m situation I would’ve replaced it

I will delve into the settings to see whether sideband is enabled or not.

I have PC133 CL2 memory. The timings are as tight as they’ll go, in all 3 of those benchmarks. It seems to get better results with 2 matching sticks (4 ranks)

I believe it’s build 330 of 2001se but i will check. What are the fps scores in your game tests?

I can try and match the speed of my machine with your Barton 2600+ for a proper comparison? Also let’s compare GPU clocks

I only have PC133 CL3 memory, so that explains at least part of the difference. Though if that's the only reason, it's a bigger difference than I would have expected. AGP sideband addressing shouldn't have measurable effect, as even going from AGP 1x to 4x caused a 3% performance increase. Everest Home Edition says that FX 5900 XT's core and memory clocks are 300MHz and 350/700MHz, which is lower than it should be for that GPU. I guess I'll have to investigate this...

EDIT: Apparently it was just running at 300MHz when running 2D software on desktop. It should be 400MHz on games.

I'm also using build 330 of 3DMark 2001 SE, here's a more detailed result. And yeah, it would be nice to compare to your results at the same clock speed. 😀

The attachment 3dmark2001se_fx5900xt_4x_agp.gif is no longer available

Had a look at the settings

Confirmed 2001se v330

Nvidia settings;
Image quality set on performance. (Middle setting) mipmap detail on blend (middle setting). Everything else application controlled. Fog table emulation set to on.

DirectX version installed is 9.0c

According to riva tuner. Side band addressing is enabled
Clocks are also 300 / 700. Could do with GPU-z for win 98 😅 to see what it does when it’s actually rendering!

I’ve done 2 benchmarks around 1917Mhz. Both slightly under, around 1911-1914

147 x 13

137 x 14

Comparison table

FullYes wrote on Yesterday, 16:43:

Had a look at the settings […]
Show full quote

Had a look at the settings

Confirmed 2001se v330

Nvidia settings;
Image quality set on performance. (Middle setting) mipmap detail on blend (middle setting). Everything else application controlled. Fog table emulation set to on.

DirectX version installed is 9.0c

According to riva tuner. Side band addressing is enabled
Clocks are also 300 / 700. Could do with GPU-z for win 98 😅 to see what it does when it’s actually rendering!

I’ve done 2 benchmarks around 1917Mhz. Both slightly under, around 1911-1914

147 x 13

The attachment IMG_9882.jpeg is no longer available

137 x 14

The attachment IMG_9883.jpeg is no longer available

So it's basically just under 15% reduction compared to the nForce2 board. That's actually not too bad for getting an ISA slot and DOS compatibility in exchange.

Btw, would you like to benchmark it at 133MHz FSB and CPU at 1.8GHz as well? I'm a bit curious about how much slower my CL3 memory is. 😀

Mike_ wrote on Yesterday, 17:01:

FullYes wrote on Yesterday, 16:43:

Had a look at the settings […]
Show full quote

Had a look at the settings

Confirmed 2001se v330

Nvidia settings;
Image quality set on performance. (Middle setting) mipmap detail on blend (middle setting). Everything else application controlled. Fog table emulation set to on.

DirectX version installed is 9.0c

According to riva tuner. Side band addressing is enabled
Clocks are also 300 / 700. Could do with GPU-z for win 98 😅 to see what it does when it’s actually rendering!

I’ve done 2 benchmarks around 1917Mhz. Both slightly under, around 1911-1914

147 x 13

The attachment IMG_9882.jpeg is no longer available

137 x 14

The attachment IMG_9883.jpeg is no longer available

So it's basically just under 15% reduction compared to the nForce2 board. That's actually not too bad for getting an ISA slot and DOS compatibility in exchange.

Btw, would you like to benchmark it at 133MHz FSB and CPU at 1.8GHz as well? I'm a bit curious about how much slower my CL3 memory is. 😀

Here you go!

I also managed to do some benches at 2400Mhz. CPU-z crashes the computer when I open it. Though I was able to confirm 150MHz FSB at a lower multiplier. I may be at the limit of both the CPU and the chipset here. Or it could also be a power issue. This board still uses linear voltage regulators based on all of the heatsincs they are bolted to 😳

Updated comparison table. Diminishing returns. That last 50Mhz barely moved the needle

FullYes wrote on Yesterday, 17:50:

Here you go! […]
Show full quote

Here you go!

The attachment IMG_9890.jpeg is no longer available

I also managed to do some benches at 2400Mhz. CPU-z crashes the computer when I open it. Though I was able to confirm 150MHz FSB at a lower multiplier. I may be at the limit of both the CPU and the chipset here. Or it could also be a power issue. This board still uses linear voltage regulators based on all of the heatsincs they are bolted to 😳

The attachment IMG_9886.jpeg is no longer available

The attachment IMG_9885.jpeg is no longer available

Updated comparison table. Diminishing returns. That last 50Mhz barely moved the needle

The attachment IMG_9891.jpeg is no longer available

Thanks, so CL2 memory gives about 6% performance gain, which is surprisingly good. I wonder if those memory sticks are expensive, though? 😀

Anyways, it looks like overclocking gets very much diminishing returns after 2GHz...

Mike_ wrote on Yesterday, 19:18:

FullYes wrote on Yesterday, 17:50:

Here you go! […]
Show full quote

Here you go!

The attachment IMG_9890.jpeg is no longer available

I also managed to do some benches at 2400Mhz. CPU-z crashes the computer when I open it. Though I was able to confirm 150MHz FSB at a lower multiplier. I may be at the limit of both the CPU and the chipset here. Or it could also be a power issue. This board still uses linear voltage regulators based on all of the heatsincs they are bolted to 😳

The attachment IMG_9886.jpeg is no longer available

The attachment IMG_9885.jpeg is no longer available

Updated comparison table. Diminishing returns. That last 50Mhz barely moved the needle

The attachment IMG_9891.jpeg is no longer available

Thanks, so CL2 memory gives about 6% performance gain, which is surprisingly good. I wonder if those memory sticks are expensive, though? 😀

Anyways, it looks like overclocking gets very much diminishing returns after 2GHz...

I think the ram I have cost me about £12 per 256MB module. The seller still has some left. But the price has gone up to £18 per module plus postage. Not crazy money. But heading in the wrong direction.

I do also have another stick of RAM that is rated CL3 that will work at CL2, but fails if I try any overclocking on it. I’ve picked up a lot of RAM sticks in job lots for cheap, but not found anything as good as what’s in it now.

You really see the limit of this platform at the higher clocks. I would say it’s already moreorless plateaued even at 1.67Ghz; from there, to increase the score by the maximum I’ve achieved (~16%) it’s taken about a 45% higher clock speed. Most of the Gains come out of increasing the FSB. I will run one more test at 16x133 and I think I’m done playing with this system for now. I have something exciting to share on the socket 7 system tomorrow