dionb wrote:

Can you solder? K6-2/3+ Success on Asus P5A 1.06 - simple mod

This would fix it if it really is the same issue 😀

Thanks for the tip - I recently purchased such a board.

Andry wrote:

Immediately after reading your post I remembered that I read somewhere that ALL AMD K6 series CPUs (2, 3, +) produced after 03/2 […]
Show full quote

meljor wrote:

Learn something new everyday, didn't know this myself and the p5a series is my favorite....thought only the ''+'' were affected. […]
Show full quote

Learn something new everyday, didn't know this myself and the p5a series is my favorite....thought only the ''+'' were affected. Found this on the web somewhere, if your test with a slower/older cpu works please confirm. Perhaps the mod i linked to earlier will also fix this problem?

Found this:

it depends on the chip to some extent the first few batchs of 500, 533 and 550 chips will run fine on the any P5A but the newer batch's sence around the time the + line of chips came out do not. I know some people running these combos and there fine I also now some people who tryed these combo's and they did not work ?

Any way this is what I do Know
AMD made a change to the core instruction set to all newer K6 chips that makes them incompatible with the ICC chip on the revision 1.5 and 1.6 P5A and P5A-B motherboards and the 1011_4 beta will show the correct speed and chip and even identifiy the + chips correctly but the problem is not with the bios it's with the Hardware on the board as in the ICC chip

if your getting really low benchmark results as in your system is benching like a P200 then you have this problem and no bios or tweak program is going to fix it I tryed them all on 3 diffurent Asus P5A's rev. 1.6 motherboards with 2 diffurent K6-2-500's and a K6-3+ 450
pop the same chips in a POS Amptron all in wonder, a FIC 503+, and a Acorp Ali5 and they run fine and bench normal put then in a P5A boots fine acts'like it's running fine just kinda slow bench them and it says there running like P100's ?

you can try the beta bios and all the tweak prog's out there but I'd bet it's not going to help.

Immediately after reading your post I remembered that I read somewhere that ALL AMD K6 series CPUs (2, 3, +) produced after 03/2000 are not suitable for Asus P5A 1.06 or later.

However, I think that this info should be revised, because my K6-2 533 MHz is from 5th week of 2000 (0005 datecode). I supposed 03/2000 intend March 2000, so from 9th week. If it means from 3rd week, it seems to be OK (do I own one from first P5A 10.6 incompatible CPUs batch ??? 😵 )

I replaced the CPU by a 400 MHz one (year 1999) and the bench is around 1300, as expected.

I think I'd search a 1999 made K6-2 500 - 533 - 550 CPU.. 😀

Thanks to everyone for your help!

Andrea

Nice! Glad to hear the mystery is solved. The mod from the link should fix the problem so maybe you can give that a try, otherwise good luck finding the right cpu (shouldn't be too hard).

dionb wrote:

Can you solder? K6-2/3+ Success on Asus P5A 1.06 - simple mod

This would fix it if it really is the same issue 😀

OOHHHH YYYEEEESSSSSSSS!!! It does work!!! 😀

You are the man mate, thank you very much! 😀

Nope, meljor is - I was just repeating the link he gave earlier 😉

Very interesting to see that it's not just the K6Plus CPUs but any late-stepping K6 impacted - and great that you've proved in the best way possible what to do in that case.

Wow, quality control at ASUS must have gone down the shitter if all P5A/P5A-B rev 1.05 & 1.06 run like this 🤣

Nice that it works, thanx for the feedback!

F2bnp wrote:

Wow, quality control at ASUS must have gone down the shitter if all P5A/P5A-B rev 1.05 & 1.06 run like this 🤣

Yeah, I think so... It is my favorite ss7 board and I couldn't really blame them for not supporting a mobile chip the right way as they are not intended to run on a desktop board anyway but they never even confirmed the problem. Other brands did not even have the issue.
Now the latest batches of the desktop cpu's also seem to be affected... yes, that's pretty lame of Asus.

Question is: why? The early boards run fine with all the cpu's but they are limited to caching 128mb ram due to a problem within the Ali chipset (on every brand, not Asus specific). Ali fixed this with the ''G'' revision and all of a sudden the Asus boards got these problems while others like Gigabyte kept working fine. They do however run fine with the earlier k6-2 and k6-3 cpu's and can also cache 512mb.

Well, at least we can fix it now, after 20 years 😎

Is Gigabyte GA-5AA limited to 128mb too?
Possible to check how much ram is cached by a software?

Is Gigabyte GA-5AA limited to 128mb too?

TAG RAM is soldered on the board - only up to 128 mb.
TAG RAM is not soldered - 128+ mb.

Easy to identify even on bad photos with unreadable chipset markings. Gigabyte GA-5AA may or may not have TAG chip near the AGP slot.

Here's the photo, would you tell me?

That's a revision '' E'' and can only cache up to 128mb. With a k6-2+, k6-3 or k6-3+ this is not an issue and you can cache much more ram.

I'm having troubles with that mb too.
I have rev 103, using a k6-2 500, 128mb ram and a tnt2 m64. I was doing some benchmark to find the best ali drivers.
Everything was "fine" until I installed the driver agp v1.82, then I had to reinstall the video driver (that was adviced during the installation, reinstall video driver... ok wil do!) so I installed again the detonator_3_w9x-805.
From that time things have gone worse and worse, 3dmark runs fine but when I test other benchmarks like incoming (direct3d) or mdk2 (opengl), the results are halved, Incoming was able to do 70 fps, now it does 44, mdk2 went from 32 to 24 fps.
I tried to format, installed the ali drivers and then video, same bad results...
I never had so many issues with Via MVP3.

meljor wrote:

[...] it depends on the chip to some extent the first few batchs of 500, 533 and 550 chips will run fine on the any P5A but the […]
Show full quote

[...] it depends on the chip to some extent the first few batchs of 500, 533 and 550 chips will run fine on the any P5A but the newer batch's sence around the time the + line of chips came out do not. I know some people running these combos and there fine I also now some people who tryed these combo's and they did not work ?

Any way this is what I do Know
AMD made a change to the core instruction set to all newer K6 chips that makes them incompatible with the ICC chip on the revision 1.5 and 1.6 P5A and P5A-B motherboards and the 1011_4 beta will show the correct speed and chip and even identifiy the + chips correctly but the problem is not with the bios it's with the Hardware on the board as in the ICC chip

if your getting really low benchmark results as in your system is benching like a P200 then you have this problem and no bios or tweak program is going to fix it I tryed them all on 3 diffurent Asus P5A's rev. 1.6 motherboards with 2 diffurent K6-2-500's and a K6-3+ 450
pop the same chips in a POS Amptron all in wonder, a FIC 503+, and a Acorp Ali5 and they run fine and bench normal put then in a P5A boots fine acts'like it's running fine just kinda slow bench them and it says there running like P100's ?

you can try the beta bios and all the tweak prog's out there but I'd bet it's not going to help.

Interesting, is there a good way to figure out which CPU revision I have or whether it's affected? I'm running a non-Plus 500 on a P5A-B 1.05 with the revision G chipset. Have been reading this thread for tweaking tips and learning a ton.

It goes fine & runs how I remember it, but my only other K6-2 experiences were with this board 18 years ago...

I'm getting benchmark results similar to the OP in the first post. Haven't updated to the 2.13 AGP driver or run the tweak utility yet. [Edit: where can I download this specific version? VogonsDrivers only has a 2.20 integrated bundle - will that do?]

Here's my CPU:

I guess 0021 is the date code?? 21st week of 2000? Am I reading that right?

Well that was interesting ... I tried to do some investigation & benched my K6-2 against a Cyrix MII/PR233 (the closest thing I have on hand to another K6) at exactly the same bus speed and multiplier, 2.5 x 75 = 188MHz, as per the Cyrix. Clock for clock the Cyrix beat the K6 in nearly everything, except 3DMark 99MAX's CPU Mark, which is probably benefiting from 3DNow.

I didn't do ANY changes in the BIOS/drivers or jumper settings between the two CPUs except the voltage. It definitely seems like my non-Plus K6-2 is triggering the bug. Check out those Unreal and Descent 3 scores.

If that's annoying to read, I even made a chart:

The K6-2 at 500 is definitely faster, but it doesn't seem like 2.5x faster, and the Cyrix at 188 nearly caught up in some applications.

Incidentally my Voodoo3 runs like a champ at 75MHz (albeit with a little extra fan blowing on it); the system is rock solid with the Cyrix & pretty quick for what it is. I might keep it installed until I can find a non-bug-triggering K6 or a good enough soldering iron to do the resistor mod.

Does the resistor mod completely mitigate the bug?

Bumping this back up. It appears my K6-2/500 with date code 0021 (late May, 2000) is NOT affected by this bug. I was able to see huge gains by installing the right AGP driver, tweaking the hell out of it, and using some faster memory. Knock on wood, but I'm extremely pleased with the performance I managed to get. My "final" 3DMark 2000 score is 1449.

So either not all late non-Plus CPUs got the instruction change, or it doesn't affect them for some reason, or I have a mythical non-bugged revision G chip on my P5A-B 1.05.

I did my thing again (ran a whole pile of benchmarks and put them into a spreadsheet.) Here are the results:

Note that there's a complete Windows reinstall & sound card swap, along with an upgrade to a matched pair of PC133 RAM, between columns 1 and 2.

I saw huge gains just by upgrading from the ASUS-supplied version of the AGP driver (v1.54e) to the 2.13 unified, and even more from tweaking the settings. (The 2.20 unified pack from Vogonsdrivers wouldn't install.)

Here's the final AGP settings I settled on:

The biggest gains came from enabling Write Allocation and K6 EWBEC -> 3 (which I set at the same time.) The other stuff had very little effect, although GAT Mode seemed to net me a couple FPS. Changing the RAM timing from 3-2-2 to 2-2-2 for the final run of benchmarks gave a tiny but measurable boost.

Check out the improvement in bus transfer rate in Final Reality! Exactly the same system specs, only the driver tweaks above applied:

Here's the granular results I got for the four games trying all the driver tweaks, in case you want to scroll through. Fun stuff.

1*** Baseline (no tweaks):
2D3              40 FPS
3UT99            29.63 FPS
4GLQuake         91.4 FPS
5WinQuake        12.8 FPS
6
7*** Tweak attempt 1:
8K6 EWBEC        2 (default)
9GAT Mode        0 (default)
10FB size         ALL
11Write Alloc     Enabled w/o hole
12
13D3              41-43 FPS (small gain)
14UT99            29.29 FPS (same)
15GLQuake         122 FPS (HUGE gain)
16WinQuake        20.7 FPS (big gain)
17
18*** Tweak attempt 2:
19K6 EWBEC        3
20GAT Mode        0 (default)
21FB size         ALL
22Write Alloc     Enabled w/o hole
23
24D3              43-47(!) FPS (small gain)
25UT99            29.69 FPS (same)
26GLQuake         122.7 FPS (same)
27WinQuake        20.7 FPS (same)
28
29*** Tweak attempt 3:
30K6 EWBEC        3
31GAT Mode        1
32FB size         ALL
33Write Alloc     Enabled w/o hole
34
35D3              45 (same)
36UT99            29.35 (same)
37GLQuake         122.3 FPS (same)
38WinQuake        20.7 FPS (same)
39
40*** Tweak attempt 4:
41K6 EWBEC        3
42GAT Mode        2
43FB size         ALL
44Write Alloc     Enabled w/o hole
45
46D3              45 (same)
47UT99            30.15 (same)
48GLQuake         123.1 FPS (same)
49WinQuake        20.7 FPS (same)
50
51
52*** Tweak attempt 5:
53K6 EWBEC        3
54GAT Mode        2
55FB size         ALL
56Write Alloc     Enabled w/o hole
57
58Use UC For WC   enabled
59
60D3              45 (same)

61UT99            29.90 (same)
62GLQuake         122.9 FPS (same)
63WinQuake        20.7 FPS (same)
64
65*** Tweak attempt 6:
66K6 EWBEC        3
67GAT Mode        2
68FB size         ALL
69Write Alloc     Enabled w/o hole
70
71Use UC For WC   enabled
72LGBT Always Fetch enabled
73
74D3              45 (same)
75UT99            29.66 (same)
76GLQuake         122.6 FPS (same)
77WinQuake        20.7 FPS (same)
78
79*** Tweak attempt 7:
80K6 EWBEC        3
81GAT Mode        2
82FB size         ALL
83Write Alloc     Enabled w/o hole
84
85Use UC For WC   disabled
86GTLB Always Fetch disabled
87
88SDRAM timing    2-2-2
89
90D3              45 (same)
91UT99            30.5 (small improvement)
92GLQuake         123.6 FPS (small improvement)
93WinQuake        20.9 FPS (small improvement)

BTW, in case this is helpful, I found this article from a Chinese magazine where they go through all the driver settings in detail. I had a hell of a time squeezing this through Google Translate & having all the images show up, so I made a (long!) png capture of the whole thing. It's not too bad to read, surprisingly.

Here's the description of all the settings in the AGP utility from elsewhere on the same site:

ALi Yangzhi chipset motherboard AGP Utility tool latest version 1.40 For Win9x/ME/2000/XP (released on August 31, 2001) Yang Zhi […]
Show full quote

ALi Yangzhi chipset motherboard AGP Utility tool latest version 1.40 For Win9x/ME/2000/XP (released on August 31, 2001) Yang Zhi can always come up with some practical tools, which really makes everyone like it, they introduced earlier The ability to improve disk performance through settings is the IDE Cache Settings Center. This time we are seeing their new work, AGP Utility. Making some appropriate settings for the AGP part of the chipset will enable us to play a greater role in the investment in graphics cards.

After executing AGP Utility, we saw that AGP Utility provides five options: Status, Normal, Turbo, Safe, Manual. Status is used to display the current AGP part settings and working status; Normal is the normal AGP setting; and the Turbo option unopens all applicable AGP settings, this option maximizes performance; Safe option takes The most conservative settings ensure no problems.

Finally, let's talk about Manual. After selecting Manual, there will be a very complicated option. Here we will explain to you: First, the AGP data rate setting, which can adjust the AGP rate (1X, 2X, 4X); Force Side Band Disable, this option can force the sideband addressing function to be turned off; PCI Mode, if selected, runs at the PCI-BUS rate, not recommended; PCI Retry Counter, after opening this function, the Northbridge chip transfer command will retry 0xff times. This function is only available for ALi M1621; Set K6 EWBEC, if you are using K6-2/3 chip system, then choose it, can improve performance; Use UC For WC, can use Uncacheable memory as write-combining memory; GTLB always Fetch, forcing GART Table; Reset AGP command, this option is useful for normal recovery of ATI graphics card from suspend mode; AGP Timing, turn off concurrent access; Frame Buffer Size, set frame buffer, there are multiple options available, respectively Is disable, 1M, 2M, 4M, 8M, 16M, All; Fast Write Function, AGP fast write, can open it; AGP delay Clock, this option control 1541 chip AGP clock digital PLL, easy to cause a crash when not well adjusted; SDRAM Delay Clock, adjust SDRAM DPLL, easy to cause a crash if not adjusted properly; Disable PCI read Prefetch, turn off PCI pre-read; Force Alert Request, support CPU save Take PCI.

I have written so much in one breath, and everyone may have to look tired, but the use of tools is a qualified enthusiast. This tool is recommended for updates from friends who use ALi's related chipsets. Friends can go to our hardware evaluation section to check the detailed instructions for using this tool .

Thank you for the translation, too bad it considers a much more new mb and chipset, I did try to mess with agp utility while doing the round up benchmark comparison but with the tnt2 m64 I was stuck at agp 1x and all other settings didn't so much difference.

Apologies for the thread necro, but wanted to say "thanks" for all the effort to document and test (and test and test). I am getting awesome results on my new-to-me K6 system:

K6-III+ 450 ACZ (at stock speed 4.5x100)
Asus P5A 1.03 (1011 Beta 5 BIOS)
3x128MB Micron PC133 CL2
3dfx Voodoo3 3000 AGP
Sound Blaster AWE64 Gold
Intel PRO/1000 GT

As mentioned here and elsewhere on the forum, the ALi 2.13 drivers were key - the AGP tweaks provide additional gains with no apparent loss of stability. I'm curious to push the clock speed on this chip but want to get to a good baseline OS image before I cause a ruckus.

doogie wrote on 2019-10-08, 01:23:

As mentioned here and elsewhere on the forum, the ALi 2.13 drivers were key

Also, the Ali AGP Utility (v1.40) was necessary to make it work, in my case. There, I disabled Sideband addressing and set AGP speed to 1x, on my p5a rev 1.0.3 k6-2+ 550@500Mhz . Without these settings, Direct3D instantly crashed in my case, with three generations of NVIDIA cards (TNT2, Geforce2, FX).
The latest Nvidia drivers crashed for me. I used instead version 7.76 with success, as recommended on these forums. This version is available from vogonsdrivers.com.

It was helpful to start with bios defaults for the 'chipset features' page in bios settings as well as the 'safe' preset in Ali AGP Utility.
The 'dxdiag' which comes with Windows 98 SE (and DirectX) was helpful for a quick test of "does this work _at all_".

(Just providing some pointers for the next person, or for my future self:)

doogie wrote on 2019-10-08, 01:23:

Apologies for the thread necro, but wanted to say "thanks" for all the effort to document and test (and test and test). I am get […]
Show full quote

Apologies for the thread necro, but wanted to say "thanks" for all the effort to document and test (and test and test). I am getting awesome results on my new-to-me K6 system:

K6-III+ 450 ACZ (at stock speed 4.5x100)
Asus P5A 1.03 (1011 Beta 5 BIOS)
3x128MB Micron PC133 CL2
3dfx Voodoo3 3000 AGP
Sound Blaster AWE64 Gold
Intel PRO/1000 GT

As mentioned here and elsewhere on the forum, the ALi 2.13 drivers were key - the AGP tweaks provide additional gains with no apparent loss of stability. I'm curious to push the clock speed on this chip but want to get to a good baseline OS image before I cause a ruckus.

I have similar setup, same p5A 1.03, Voodoo 3 3000, Awe Gold, 3com nic, 128MB 100mhz and k6-2+ 570 @600 (5x100).
My score is 3922 in 3DMark 99 and 1917 in 3DMArk 2000. After install 3com nic drivers, 3638 and 1788¿?¿?
What values ​​do you get?
What values ​​do you configure in Bios, CTU and AGP tools?