i digged my old test results up:
on mvp3 board with 2mb cache, k6-2-550(not k6-2+) took 6:50 to run superpi, which is still a bit slower than pmmx-300. with onboard cache disabled, it took 10:15, a terrible 50% impact!
3dmark2000 is also very dependent on cache. 1600pts with onboard cache on, and 1070pts with disabled , thats nearly 50% impact aswell.
quake2 is a bit less sensitive. 57fps vs 42fps, thats about 36% difference.

My internet connection have only been working intermittently latly. For some reason this made me spend less time with the retro boxes which isnt connected to internet in the first place.
I have switched the Soyo board for a Pc-Chips m577 with Via mvp3, only 1mb cache though.

The superpi score is down to 8 min 34s with a k6-2@500
Still slow but at least better.

The board can only cache 255mb not 256mb
Is there a way to disable 1mb memory so all memory get cached?
I guess 1mb of uncached memory at least have some performace inpact in windows since it allocates memory top down?

Here are some pics

I came to a P5A-B rev 1.06 board lately, it booted fine. So I tested the cacheable limit, put 512 MB of RAM into it and a classic K6-2 450 MHz (with no L2 cache). And really this chipset revision (M1542 Rev. G) caches all the memory, despite having only 512 kB of L2 on board.

Anyway judging from the northbridge label, it is an Ali Aladdin V+ instead of the classic non-plus Aladdin V (M1541) version, right? What are other differences between M1541 and M1542 then? Besides the fact, that in Asus's case this board performs worse with K6-2+/III+ CPU (yes I read about modding it, but I do not plan running any of these CPUs in it...).

From your pic the Tag RAM IC is missing and CE2.

Yes there is no tag chip, but it is somewhat embedded in the ali northbridge isn't it?

G revision - integrated TAG
Earlier revisions - external TAG (same as Intel chipsets)

there isn't much difference between m1541 and m1542, except that the m1541 pinouts are arranged for atx layout while m1542 pinouts for at layout.
what matters is the revision as already mentioned.
however, i've become uninterested in ali5 boards as its performance simply cannot match that of the mvp3 with 2mb cache.

noshutdown wrote:

there isn't much difference between m1541 and m1542, except that the m1541 pinouts are arranged for atx layout while m1542 pinouts for at layout.
what matters is the revision as already mentioned.
however, i've become uninterested in ali5 boards as its performance simply cannot match that of the mvp3 with 2mb cache.

Is the difference that big? As i think the Ali boards are better performers against the Mvp3 boards hands down, and I've tested a lot of them (mostly with 3dmark and Everest).

But most Mvp3 boards I've seen only had 512kb cache and only one had 1MB (DFI board, didn't seem much faster as the 512kb boards?).

So, no experience with a 2MB board but i always doubted the performance would be much higher. Real world performance there isn't a big difference anyway imho but using benchmarks the memory speed of the Ali boards seem much better. I also had less problems with agp on the Ali boards.

I thought increasing the L2 cache on an MVP3 board was for the purpose of increasing how much RAM could be cached? I too don't fancy ALi SS7 boards because of the confusion around K6-2/3+ support. I use a 2 MB MVP3 board by Tyan in one of my builds. I know Epox and Tyan make 2 MB revisions that, anecdotally [and visually], seem to be of quality construction.

meljor wrote:

Is the difference that big? As i think the Ali boards are better performers against the Mvp3 boards hands down, and I've tested a lot of them (mostly with 3dmark and Everest).

But most Mvp3 boards I've seen only had 512kb cache and only one had 1MB (DFI board, didn't seem much faster as the 512kb boards?).

So, no experience with a 2MB board but i always doubted the performance would be much higher. Real world performance there isn't a big difference anyway imho but using benchmarks the memory speed of the Ali boards seem much better. I also had less problems with agp on the Ali boards.

i think the difference is quite noticeable, and i hardly had any agp problems with my mvp3 in win2000.

feipoa wrote:

I thought increasing the L2 cache on an MVP3 board was for the purpose of increasing how much RAM could be cached? I too don't fancy ALi SS7 boards because of the confusion around K6-2/3+ support. I use a 2 MB MVP3 board by Tyan in one of my builds. I know Epox and Tyan make 2 MB revisions that, anecdotally [and visually], seem to be of quality construction.

larger cache always help because the more you hit cache, the less you need to access ram, and cache is somehow faster than ram. although sometimes that advantage is not so revelant as the caches ram size.
this is particularly important for mvp3 because its sdram performance is quite slow, so it needs to be accessing cache for all time and avoid sdram access at any cost.

Are you able to provide some benchmark results showing configurations with 512 KB vs. 1 MB or 1 MB vs. 2 MB? I cannot imagine there being more than a few percent performance benefit, especially from 1 MB to 2 MB.

feipoa wrote:

Are you able to provide some benchmark results showing configurations with 512 KB vs. 1 MB or 1 MB vs. 2 MB? I cannot imagine there being more than a few percent performance benefit, especially from 1 MB to 2 MB.

yeah i think its a just few percents difference between 1mb and 2mb version, but thats noticeable and better than nothing.

but thats noticeable and better than nothing.

Few percent is not noticeable and can be easily negated by manufacturer performance tweaks to a smaller cache motherboard.

The Serpent Rider wrote:

Few percent is not noticeable and can be easily negated by manufacturer performance tweaks to a smaller cache motherboard.

well, maybe its just a problem of believe or not, but i surely do.

meljor wrote:

Is the difference that big? As i think the Ali boards are better performers against the Mvp3 boards hands down, and I've tested a lot of them (mostly with 3dmark and Everest).

But most Mvp3 boards I've seen only had 512kb cache and only one had 1MB (DFI board, didn't seem much faster as the 512kb boards?).

So, no experience with a 2MB board but i always doubted the performance would be much higher. Real world performance there isn't a big difference anyway imho but using benchmarks the memory speed of the Ali boards seem much better. I also had less problems with agp on the Ali boards.

Yep, my MPV3 is only 512kB (when I had a mvp3 board back as new in 1999 it also had only 512kB). I've never met a 1MB L2 board around (two weeks back, there was one on the local auction supposedly with 1MB, but it was pcchips 😊 and the price went quite high) let alone 2MB L2 mobo 🙁

Whenever someone resurrects these threads I end up lamenting not having a K6-2+ or a K6-III to use in my TX boards..

mv_cz wrote:

Yep, my MPV3 is only 512kB (when I had a mvp3 board back as new in 1999 it also had only 512kB). I've never met a 1MB L2 board around (two weeks back, there was one on the local auction supposedly with 1MB, but it was pcchips 😊 and the price went quite high) let alone 2MB L2 mobo 🙁

The PC-Chips M577 isn't a bad board.

I wouldn't pay insane money for one but something like 50-60 euro wouldn't be unreasonable if the board is in good working order.

The M577s cache and memory performance is nothing to sneeze at.

Oh, it ended at 40 EUR 😀 for the pcchips mobo+CPU (K6-2@300), some RAM and ATcase+PSU setup. And it seemed kind expensive to me - maybe just because I got ma VA503+ PC for free and the P5A-B cost next to nothing (about 10 EUR bundle with another working nforce2 board, some NOS socket7 coolers and SDRAM sticks).

appiah4 wrote:

Whenever someone resurrects these threads I end up lamenting not having a K6-2+ or a K6-III to use in my TX boards..

I regret almost accidentally having sold my TX board so I can't compare my P6-3+ against my Aladdin V and MVP3 😉