The organization of PCI INTs to PCI slots is unique to every board really. Sometimes the manual spells it all out very nicely. Sometimes I've even seen it indicated in the BIOS PnP area.

I dug out my KT133A based QDI Kinetiz mobo again. I put the mobile barton XP-M 2400+ rated CPU (code AXMH2400FQQ4C) back in to it. I also reinstalled the LLA driver device using Windows 98 "add hardware wizard", as the LLA download package is missing its help file, and consequently the setup.exe fails to run correctly. I run wpcredit. Using elfuego's screenshot shown earlier in this thread, I set bit 3 from the right to 1 for registry entry number 55. This was crucial because now everything works. When I launch cpumsr.exe, it still reports the error message of "Processor could not be changed! Unknown error occured", but it doesn't appear to be important. I set the multiplier (from 6x) to 13.5x with a voltage of 1.45, and when I click on the Set button, I see the "Transition finished" success message. I ran Sandra to make sure the CPU speed had increased, and it said it was now running at 1.79Ghz.

The next bit of good news is that coolon.exe now "magically" works. When I click the "cooling" checkbox, sure enough the temperature drops inside speedfan.exe. Here are my temperature with cpumsr test results:

I switch on the PC, and the BIOS POST message reports "Mobile AMD Athlon 800mhz" (it's running with a 6x multiplier @ 133)
Temp at Windows 98 desktop = 23.7
Temp 5 minutes later = 25.9

I then use cpumsr to change the multiplier from 6 to 13.5, so that the CPU is now running at 1.79Ghz (from 800Mhz)
Temp = 27.2
Temp 5 minutes later = 31.6

I then tick the coolon checkbox, to enable cooling
Temp 5 minutes later = 24.1 (that's a 23.7% reduction: 7.5 degrees C drop from 31.6 shown above)
Temp 5 minutes later = 23.2 (that's a 26.5% reduction: 8.4 degrees C drop from 31.6)

The temp seems to have stopped dropping now. 23.2 seems to be it's new desktop idle value.
I untick the cooling checkbox, and wait 5 minutes.
Temp = 30.7

Retro, I see we use exactly the same CPU. So why are you torturing it with such low speeds? I mean... Its like having a ferrari and driving it only in the first gear 😉

Yes yes I know that we dont really need more for DOS/Win98, but for the sake of testing - try to run 720p HD video from youtube at 2.2Ghz - I was surprised that I finally have decent FPS for it on retro machine 😉

For example take this link:
http://www.youtube.com/view_play_list?p=1835C … L&v=06LlTvN_Dqc

Its gotten pretty decent now; it was really crappy and laggy before @ 1.7Ghz. Try it 😀

Edit: BTW I got the same error as you did from Win98 and LLA driver; but I didnt install it via add new hardware. I've installed it manually by copying the stuff to system32 directory as it was written in the readme.txt. You get used to reading this stuff when you start working in linux 😉

OK, I need to increase the XP-M's multiplier from 13.5x to 18x. To do this, I need to cut the 3rd L6 bridge. I have a minor problem. My XP-M CPU looks a bit like this -

http://www.cpu-world.com/CPUs/K7/TYPE-Athlon%20XP-M.html (please scroll down to the 2nd photo. It's the one with the red plastic cover on the CPU.)

I can't remove this strange plastic protective cover. Perhaps I could just cut through it?

Also, I have a sharp "generic DIY box cutter" knife. I was thinking of simply pushing the tip of the blade down in to the CPU, rather than making a "moving along the surface" blade cutting movement. What do you think is the best way of making the 3rd L6 bridge cut? Thanks. 😀

BTW, I just found this webpage. I think it's very good. If you scroll down 3/4 of the page, you can see this table: "Reference, L6 : SFID for Mobile"

http://fab51.com/cpu/barton/athlon-e24.html

retro games 100 wrote:

Also, I have a sharp "generic DIY box cutter" knife. I was thinking of simply pushing the tip of the blade down in to the CPU, rather than making a "moving along the surface" blade cutting movement.

I did exactly the same. Especially because with our version of the CPU the contacts are extremely tiny (smaller then on t-bred and normal barton) and you must NOT cut the "surrounding" box of the bridge! We are talking about a few mm here so be very careful. I've butchered an old and already fried XP1700+ for the sake of learning and experimenting. If you have such an old one which you can use as a guinea pig that would be a good start.

I've done the cutting with a simple cutter knife and btw, Fab51 is the site where I also got the info about bridges 😀

About the plastic cover - it doesnt look like too much of a problem. I would just cut through it, although my CPU looks exactly like the green one on the bottom of the page with athlons.

The old ceramic chips were so easy to unlock. You could actually very easily solder the bridges shut. No messing with conductive paint and no messing with pencil graphite that very slowly evaporates when it has current running through it.

I just gave up when they moved to that fiberglass packaging. Meh. Very frustrating.

I've used a socket mod though for a few mobos with limited multis.

Holding a torch with fading battery power, a magnifying glass that came out a christmas cracker back in the 1990s, a heavy metal "box cutter" with a fat tip, shakey sweaty hands and bad eyesight isn't the best combination! And after about 5 minutes of dithering, I decided to just push the blade down in to the CPU and hope for the best. Squinting at my handywork, I could see that the cut made seemed a reasonable "shape" (length and depth), but it was badly positioned to one side and not in the middle of the two "gold bridge ends". Crucially however, the cut made was not immediately next to one of the two "gold bridge ends".

I switch on the PC and no fire occurs. 😀 I run cpumsr, and the multiplier options now say 18x max. (Not 13.5x max) Yay! I select 18x, but the PC goes mad and I have to reboot. I try this again, but this time I select 17x. All seems well! (Please note: I run coolon to help out with the cooling.)

CPU temperature is 24.5. That's awesome.
CPU speed = 2.26Ghz. That's not bad at all! (17x mult, @ 133fsb)

Sandra dry = 6213, wet = 3093
Sandra multi = 12331, 13681

pcpbench mode 105 (1024x768) = 117.2 (using 128-bit FX5200)

This is interesting - after I tell cpumsr to increase the multiplier from 6x to 17x, the core voltage increases from about 1.51 to 1.61. (I double-check AXMH2400FQQ4C on cpu-world, and this CPU's voltage requirement is just 1.45. 😦)

Tomorrow, I'll run some 3DMark benchies, just to make sure the system is stable. I'll probably set up the mobo to use 2 PSUs, just to help out with any extra juice required.

Edit: Please note that I can't view that youtube HD video, because this mobo/PC has no network hardware. (I have no plans for retro PCs + connection to the internet.)

I can't believe this thread is still living. Just get a real (Intel) board and be done with it. The KT133 was terrible.

I think you're missing the point.

I'm super excited now. going to mess with this later tonight when I get home 😀

I have run some more benchies. I have used 2 PSUs. 1 Enermax PSU (475w with a meaty 5v rail) just for the mobo, and 1 Nexus PSU (350w) for the Radeon 9800 Pro card, 1 HDD, 1 DVD, and 2 fans. For the Radeon card, I specifically set its v-sync setting to "always off".

3DMark 99 Max (default settings) = 13533, 32848 CPU 3DMarks.

3DMark 2000 (default settings) = went mad at the start of the first "adventure test" - no result.

I rebooted and then did not crank up the CPU's speed for the next (repeat) test below.

3DMark 2000 (default settings, v-sync reset to its default value) = again, it went mad at the start of the first "adventure test" - no result.

So, perhaps these 2 "bad results" were graphics card driver related problems?

I crank up the CPU's speed once more to 2.26Ghz, and begin testing again -

3DMark 2001 first edition (default settings, v-sync at default setting) = 11286 (CPU temp is now 33.4)
3DMark 2001 first edition (default settings, v-sync set to "off") = 11277 (CPU temp is now 33.4)
3DMark 2001 first edition (completely maxed out settings incl. 6x AA, v-sync set to "off") = 7223 (CPU temp is now 33.9)

Just for a bit of extra fun, after these tests completed, I rebooted to get access to the BIOS set up area. I then adjusted the QDI mobo's simple OC'ing facility and set OC'ing to: 140/35Mhz/Off. I then set cpumsr's multiplier to 17x, at which point the PC went mad, and I had to reboot. I adjusted the BIOS OC'ing setting to: 136/34Mhz/Off, and set cpumsr's multiplier to 17x again. That seems to work. Sandra tells me that the CPU's real speed is 2.31Ghz. I then rerun one of the 3DMark2001 tests above, but it quickly goes mad and I have to reboot. Maybe the QDI board is not so good at OC'ing? Or maybe the combination of the new 17x multiplier with an increased FSB is causing some problems? It's really not important, because the 17x multiplier @133FSB (2.26Ghz real speed) seems stable. (BTW, with the 2 PSU set up, I tried setting the multiplier to 18x, but the machine locks up very soon afterwards.)

I'd like to say "big thanks" to prophase_j and elfuego for providing essential guidance to getting this all working. It was great fun! 😀

Imagine how much faster this system would be if I swapped out the "humble" 9800 card for a 6800GT or GS? Having said that, when I ran all of these tests above, the system "felt" nice and stable. The "slower" ATI graphics card (compared to nVidia 5950 and 6800 cards) did not make weird hissing sounds, as if the system was struggling to cope with the power demand.

A photo of the current set up -

Some more benchies, but this time I need to report some bad news, but with a solution. I decided to run 3DMark2001 second edition. Up to now, I've been using the first edition. I completely maxed out its settings, with 6x AA etc. But the "high detail lobby" test repeatedly fails. To fix this, instead of selecting 17x multiplier inside cpumsr, I select 16.5x multiplier instead. This provides 2.19Ghz CPU speed (not 2.26Ghz). Now, testing is OK and the second edition score is 7097.

I wanted to finish my testing with benchmarking Quake 3 Arena. It's a simple game to test, and always provides reasonable frame rates! I maxed out all of its settings, and also set the resolution to 1280x1024. I also set the multiplier to 17x, because I reckon this "simple" game won't upset the system too much. (And testing was OK.)

Using the 17x multiplier @133fsb (2.26ghz) = 115.1 fps.
Using the CPU's default "start up" multiplier of just 6x, @ 133fsb (800mhz) = 66.4 fps. That's not bad really!

I wanted to repeat some of these tests, but using an Epox "retro rocket" mobo, rather than the QDI board. Here's what I noticed -

BIOS POST screen says: 6x multiplier, 133fsb = 800Mhz. (Exactly the same as the QDI board.)

I run cpumsr, and set the multiplier to 18x, which is its maximum setting. Sandra tells me that the CPU is running at 2.4Ghz before the machine freezes. (This was marginally better than the QDI board, where it froze instantly after setting the multiplier to 18x.)

I run cpumsr, and set the multiplier to 17x. I run 3DMark2001 second edition, with every setting maxed out (6x AA, etc), and it fails instantly. This is not as good as the QDI board, because on this board, this test worked for a few minutes before failing.

cpumsr: multiplier set to 16.5x. (This yields a CPU speed of 2.20Ghz) I run 3DMark2001 second edition, with every setting maxed out (6x AA, etc), and I get a score of 7111. The QDI board gave 7097, so the Epox is fractionally better. (The CPU temp at the end of this test was 36.2)

Also, Sandra tells me that:

Core voltage is 1.66. That's rather high for a CPU which expects just 1.45.
3.3 voltage is 3.52. Again, that's a bit high.

Perhaps these values are a bit high because of the failing caps on the mobo? Some of them look a bit "evil".

Edit: This is a bit unfortunate - although I am using 2 PSUs just to increase my chances of success with these high powered Athlon tests, I was unable to use the Enermax "Athlon friendly" PSU with the mobo. The reason is because the 24 pin power plug has no "snap off" 24 to 20 pin section. I couldn't fit the 24 pin power plug on to the mobo, because there were things in the way, such as caps. So, I used the lower powered Nexus 350W (30A 5V rail) PSU to power the mobo, and the Enermax PSU to power all of the components including the graphics card.

Edit 2: I am having problems with 16.5x multiplier (@133 fsb) on this Epox "evil caps of death" mobo. 3DMark2001 s.e. at max settings keeps failing. So I set the multiplier to 13.5x (1800mhz) using cpumsr, and rerun the test. I get 6898, which is surprisingly high considering I got 7111 with 16.5x multiplier. I set the multiplier to 15x (2000mhz) using cpumsr, and rerun the test again. I get 7019.

Finally, forgetting about high powered graphics tests, I set the multiplier back up to 17x and had a game of JK:MotS. 😀

Fiddling with this chipset is a waste of time. I've had three or four different KT133 boards back in the day (from relatively cheap to really expensive), and they all were buggy, buggy, buggy. Random crashes, errors while copying from IDE to IDE, strange hardware incompatibilities, etc. . - a whole barrel of laughs that was. In the end i was frustrated and got a P4 (first Celeron, then upgraded to a P4). I was happily using that machine for more than 5 years, without any problems at all (even though it was running a rather crappy SiS chipset).

I can see their benefits. Using an unlocked mobile barton running at approximately 2.2Ghz @133 fsb can provide impressive performance, but the most important aspect is the ISA slot for that special retro specification. I'm only interested in messing about with old games, so a random crash here and there won't bother me. 😀

I wonder what you'll say when you're starting to find the first (silently) broken files on your hd (; . My point of view (even when gaming): stability > performance.

ADDiCT wrote:

I wonder what you'll say when you're starting to find the first (silently) broken files on your hd (; . My point of view (even when gaming): stability > performance.

All the time I spent with my rig and it's quetionable sability never once had an issue with data corruption. After getting all the kinks worked out in my system I used i for a over a mnoth straight without a single lockup, do all my daily tasks, playing anything from unreal, to starcraft, to civilation 4, all while running folding at home in the backround.

I assume you were on XP/2K because Win9x and Stability should not be in the same sentence. 😁

ADDiCT wrote:

I wonder what you'll say when you're starting to find the first (silently) broken files on your hd (; . My point of view (even when gaming): stability > performance.

Hmm.. Alright if you say so. I wonder when that will be. It didnt happen for... well almost 10 years now... 😅 ...and I still love this chipset, as its the most powerful chipset with an ISA slot. This means that the intel alternative would be the legendary BX chipset, which in turn only supports CPUs ~500-600Mhz (or ~1Ghz with some modifications). There is no intel CPU <1Ghz which is comparable to Athlon XP >2 Ghz! To be honest, not even P4 can match its performance.

@retro

Congratulations!!! Good job! I'm glad you had at least as much fun as I did 😉

And thanks for sharing the 3Dmark 2001 score! I also wanted to see how far can a Radeon 9800 pro go with this machine. Judging from the score, it really does get pretty limited by the SDRAM. Back at 2001 I've used Radeon 8500 with ASUS A7V133-C and I've gotten about 10500 score. Radeon 9800 pro on the other hand, in a DDR setup (e.g. Nforce2 chipset) reaches about 18 000. That means that SDRAM is a major bloody bottleneck.

Since I do not own a Radeon 9800 anymore and since I have no intention of removing V5 5500 from this machine, would you do another test for me? Go into the BIOS and loosen up the memory timings - set them as low as you can, but enable 4-way interleave. Then set the FSB to at least 150Mhz (or more, if your RAM allows it); then increase the multiplier to 15 (2250Mhz) or even more if CPU allows it, and then run the 3Dmark01 again. I assume that the score should be up by at least 200 marks, if not 500.

Oh and dont worry about voltage. Its ok if its not exceeding 1.625V. My CPU was rated 1.350V, but runs 1.525V currently.