Hi members 😁
Meljor thanks for you commitment 😁

Ok so the motherboard to every ones surprise is..... Drum role.....

Drum role....

Jetway 542C 🤣

To my knowledge these were the last boards made with AGP slot
Does not look like it was intended for a premium gaming board as its Baby AT
However the chipset is ALI Rev G with no external tag ram on the board.

I am still testing settings but I am close to say I am done.

Geforce 2 Ti Max fps in 3dmark2000 in game 2 low detail is 186fps vs Geforce 3 Ti 200 152fps
Does look like the driver 7.76 and Gf3 Ti is slightly faster with minimum fps

The awesome part is that when I run 3dmark with 512mb ram vs 128mb the scores are almost the same!! 😁 And this is what I have been looking for!
100Mhz X 5.5 550Mhz gives me 4350+- points with 512mb ram. 100Mhz X 6 gives me 4663

GTA 3 is smoother than ever at only 550Mhz!! It still does have small stutters now and then but is way more playable!

Next I want to try Windows XP :p

Skalabala wrote:

Hi members :D Meljor thanks for you commitment :D […]
Show full quote

Hi members 😁
Meljor thanks for you commitment 😁

Ok so the motherboard to every ones surprise is..... Drum role.....

Drum role....

Jetway 524C 🤣

To my knowledge these were the last boards made with AGP slot
Does not look like it was intended for a premium gaming board as its Baby AT
However the chipset is ALI Rev G with no external tag ram on the board.

I am still testing settings but I am close to say I am done.

Geforce 2 Ti Max fps in 3dmark2000 in game 2 low detail is 186fps vs Geforce 3 Ti 200 152fps
Does look like the driver 7.76 and Gf3 Ti is slightly faster with minimum fps

The awesome part is that when I run 3dmark with 512mb ram vs 128mb the scores are almost the same!! 😁 And this is what I have been looking for!
100Mhz X 5.5 550Mhz gives me 4350+- points with 512mb ram. 100Mhz X 6 gives me 4663

GTA 3 is smoother than ever at only 550Mhz!! It still does have small stutters now and then but is way more playable!

Next I want to try Windows XP :p

Never did look at my J-542C to see what revision chipset it had, it's long gone (sold) now so too bad... The j-542B i still have is just a normal revision E.

Did you test on the p5a-b how big the influence of 512mb is? As far as i know the performance penalty is pretty small anyway on a k6-3 because of the 256kb on-die cache of the cpu?

BTW, i didn't know about the internal Tag chip. I checked with my p5a 1.06 (revision G) and indeed that board has no Tag chip.

I always wondered if Asus made a mistake or the revision G was wrong so it couldn't run the k6+ at full speed (it runs very slow on the 1.05 and 1.06 boards).

Since your J-542C is also a G revision and it runs the k6+ fine i guess that mystery is solved and it is just Asus that screwed up (big time).

meljor wrote:

BTW, i didn't know about the internal Tag chip. I checked with my p5a 1.06 (revision G) and indeed that board has no Tag chip.

I always wondered if Asus made a mistake or the revision G was wrong so it couldn't run the k6+ at full speed (it runs very slow on the 1.05 and 1.06 boards).

Since your J-542C is also a G revision and it runs the k6+ fine i guess that mystery is solved and it is just Asus that screwed up (big time).

Probably a BIOS bug.

meljor wrote:

BTW, i didn't know about the internal Tag chip. I checked with my p5a 1.06 (revision G) and indeed that board has no Tag chip.

I always wondered if Asus made a mistake or the revision G was wrong so it couldn't run the k6+ at full speed (it runs very slow on the 1.05 and 1.06 boards).

Since your J-542C is also a G revision and it runs the k6+ fine i guess that mystery is solved and it is just Asus that screwed up (big time).

I do not think Asus is the problem. I think both ALi and Asus was the problem.
Feast your eyes on the date code of the chipset my K6 friends 😁
The mosfets of this motherboard also runs cool 😀
I like it! Now to upgrade the Cache chip and then make turbo PLL for it. 😎

cyclone3d wrote:

meljor wrote:

BTW, i didn't know about the internal Tag chip. I checked with my p5a 1.06 (revision G) and indeed that board has no Tag chip.

I always wondered if Asus made a mistake or the revision G was wrong so it couldn't run the k6+ at full speed (it runs very slow on the 1.05 and 1.06 boards).

Since your J-542C is also a G revision and it runs the k6+ fine i guess that mystery is solved and it is just Asus that screwed up (big time).

Probably a BIOS bug.

It is a bug that is known for a very long time and it is not a simple bios problem but a hardware one. These revisions run perfectly fine with normal k6-2 and k6-3 cpu's. It's just that they become extremely slow when a mobile k6-2+ or k6-3+ is used. Older revisions 1.03 and 1.04 have no problems. All revisions run the same bios.

But 1.03 and 1.04 have the older Ali chipset revisions that can cache up to 128mb where the 1.05 and 1.06 use the later ones that can cache up to 512mb.

Skalabala has a Jetway J-542C Motherboard with the same late revision that can cache 512mb but also runs fine with k6-3+. Win-Win.

Thanks to the on-die cache of the k6-3 it by itself can cache up to 4gb so no real problem but there is a small speed difference.
Motherboard L2 becomes L3 cache with these cpu's and as long as the used amount of memory is within the cacheable area it is fine but otherwise you lose about 5% performance.

So, beeing able to use more than 128mb (up to 512mb) and not loosing any performance is very nice and boards that can do that with this chipset are pretty hard to find.

meljor wrote:

But 1.03 and 1.04 have the older Ali chipset revisions that can cache up to 128mb where the 1.05 and 1.06 use the later ones tha […]
Show full quote

But 1.03 and 1.04 have the older Ali chipset revisions that can cache up to 128mb where the 1.05 and 1.06 use the later ones that can cache up to 512mb.
...
Thanks to the on-die cache of the k6-3 it by itself can cache up to 4gb so no real problem but there is a small speed difference.
Motherboard L2 becomes L3 cache with these cpu's and as long as the used amount of memory is within the cacheable area it is fine but otherwise you lose about 5% performance.

So, beeing able to use more than 128mb (up to 512mb) and not loosing any performance is very nice and boards that can do that with this chipset are pretty hard to find.

I guess it's the same for P5A (ATX version)? I have rev 1.06 and considered K6-3+ to have more than 128 MB RAM cacheable, but as you say higher revs can cache more I wouldn't need a + anymore and can go for K6-2 570 for example.

What exactly make the slowdowns? Why disable CPU external cache isn't a solution?
And what is - besides the cache - the advantage of K6-3 or K6-2+/3+ over K6-2? The K6-3 has max. 450 MHz, K6-2 up to 570. Especially for this "aim for the stars".

Yes, 1.06 and 1.06 of both the p5a and p5a/b have the bug. It is a hardware bug and nothing can be done (lots of people tried).
What the actual problem is nobody seems to know. It only affects the k6-+ (mobile) cpu's. Disable cache does not help.

The later boards do have the later revision chipset that can cache up to 512mb. This works fine with a k6-2 and normal k6-3.

K6-2 is slow compared to a k6-3 or k6-2+ or/k6-3+. The on-die cache of the cpu makes it a fair bit faster, not only because of the cacheable memory range.
A k6-3 (256kb on-die cache) at 450mhz will be quicker in most scenario's compared to a k6-2 570 (no own cache).

The k6-2+ does have cache (128kb) and is also a lot faster compared to normal k6-2.

The benefit of the ''+'' or mobile cpu's is the fact that the multiplier can be changed in software on the fly and the fact that they are made on a smaller process, because of that they can clock up to 600mhz without any problems in most cases. Some times even higher AND they run cooler.

For a ''+'' cpu you need bios support in most cases. At stock speeds they also require a lower voltage that a lot of boards cannot support like 1,6v-2.0v. Running these at a more common 2,1v is not a problem as long as there is a decent cpu cooler.

I've managed to squeeze 4230 from 3dmark 2000 with my k63+@550 and gf3ti200@240-430.
Motherboard is an Epox EP-MVP3G5.

I used Rivatuner to turn on sideband addressing and to get the overclocks.
The 7.76 driver is definitely a bit faster than 7.58 with this setup.
I also downgraded dx9 to dx8a which helped a bit too.

Unfortunately this mobo cannot exceed 100fsb unless the L3 cache is turned off, then 124fsb
seems to work fine but lower scores. Dos Quake does run faster though in that case.

Thanks for your contribution! What is the score if you do not use riva tuner at all?
And did ou try 600Mhz?

Is there any settings that could be adjusted in regedit on the MVP3G5?

Imperious wrote:

I've managed to squeeze 4230 from 3dmark 2000 with my k63+@550 and gf3ti200@240-430. Motherboard is an Epox EP-MVP3G5. […]
Show full quote

I've managed to squeeze 4230 from 3dmark 2000 with my k63+@550 and gf3ti200@240-430.
Motherboard is an Epox EP-MVP3G5.

I used Rivatuner to turn on sideband addressing and to get the overclocks.
The 7.76 driver is definitely a bit faster than 7.58 with this setup.
I also downgraded dx9 to dx8a which helped a bit too.

Unfortunately this mobo cannot exceed 100fsb unless the L3 cache is turned off, then 124fsb
seems to work fine but lower scores. Dos Quake does run faster though in that case.

Thanks for that score, for 5,5 x 100 that is a good score. Is that a 2mb cache board? Directx 7.0a should be your best bet, i don't know still what the best Via agp driver is.

I recenly got a Fic 503+ ! The one board that was left missing in my ss7 collection 😎 When i have time i will set it up and see if it is the beast some people claim it to be. It is a 1mb board.
I don't think it will be as fast as the Ali V boards but i would love to be surprised.

@ Skala : Any progress on your fast cache mod?

To answer a few questions.

1. This motherboard is a 2mb cache model
2. I will downgrade to dx7.0a and post that score a bit later
3. I will also try without rivatuner and hence no overclock,
although coolbits can be used but only allows for 220mhz core
4 My k63+ is not stable at 600mhz, but I can complete the game tests,
extra voltage does not always help with stability.
5. I looked but couldn't find Via Agp drivers older than what is in the
4-in-1 4.35.

Good luck with that cache upgrade Skalabala, my Epox mobo has the exact same
chips, but apparently the Tag ram chip is not good enough to allow for 100+fsb,
or the timings are too tight.

I wonder if anyone has tried Windows 95?

OK, I installed dx7.0a and got 4271. Unless someone knows where in the registry via hides agp settings then I
suspect this is as good as it will get. I'll try 600mhz tomorrow sometime.

600mhz score

Nice job, you're getting up there! But it will get harder and harder to sqeeze the last points out of the system 🤣

You're still 300 points off of my score and a 550 points of skalabala (amazing score).

Feel free ofcourse to try win95, i don't think it will be faster but i might be wrong. Very nice board for having the 2mb cache, i have just never seen one, still hoping to find one some day.
The oldest driver might not be the fastest driver, maybe try a bit later ones and see how they go? My highest scores for example were with other Ali agp driver than the one Skalabala used.
Some boards might behave differently with newer/older drivers.

It will not matter an awful lot but for the highest score every point counts 🤣

Edit: I now see your score is still with the stock 100fsb (6x100) and that is impressive! Try 2,3v-2,4v and 5,5x110fsb. If you get it stable enough it will be a big boost!

That's as good as it gets with this motherboard, as higher than 100fsb with the L3 cache on is not possible.
Even at 124fsb with l3 cache off the scores are lower.
It's been a fun time but I'm moving on to other builds now. My other 2 SS7 boards only go to 100fsb so no point
messing with them.

No progress on my cache chips 🙁 And windows 95 does not work with 3dmark 2000 🙁
Maybe it could be hacked :p

Looks like ALi driver 1.90 works good on GA5AX stable and you can enable the fast options in regedit and still be stable.
Got 4500+ with GA5AX with cpu 110x5.5 not bad.

Try agp driver from Ali / Uli integrated driver package 2.13 , worked the absolute best for me. I used it together with some ali utility to set it to ''turbo mode''

Hello,

this is what my K6 can do: