You can't expect much from PC Chips. Have you tried using 32 and 64 mb modules?

The M571 and the M577 were exceptions to the PC Chips rule.

It works perfectly OK with 2x 32MB PC-100 SDRAM sticks. Guess would work OK with 3x 128MB, but I want 768MB if possible.

TELVM wrote:

However when trying to install Win 98SE I'm getting tons of problems 😢 . Sometimes errors/freezings during installation, sometimes errors/freezings/BSODs once I can get it installed.

Regardless of what kind of chipset you're using, Windows 98SE will not run properly with 512 MB of RAM or greater unless it is properly configured. Period.

I'm still not quite sure what the best solution is, but you can experiment with taking up a few hundred megs with a RAM disk before starting Windows.

TELVM wrote:

Guess would work OK with 3x 128MB, but I want 768MB if possible.

What could you possibly plan on running on such ancient hardware that would require that much RAM?

SS7 boards are, at least for me they always are, VERY picky about RAM. Finding large modules that are detected properly and work without errors can be quite a challenge.

Win98/SE has issues with more than 512MB RAM, so that's one reason for the instabilities.
However, if smaller modules work correctly, I'm guessing that the cause of the problem are incompatible 256MB modules.

Try different bios settings that don't stress the RAM as much. Maybe play around with Write-Through/Write-Back (if it's available).
You didn't mention the CPU you're using. If it's one without L2 cache you'll also run into performance issues, especially in Windows, since you're very likely to have large amounts of uncached RAM.

Edit: Too slow ^^

I'm aware of W98 problems over 512MB RAM and know how to deal with it, that's not the issue.

Mystery wrote:

... I'm guessing that the cause of the problem are incompatible 256MB modules.

I've read that the M577 can handle RAM sticks with 16MB per chip tops. I'm trying with PC-133 256MB sticks, double sided, 8 chips per side, 16 chips total, so 256/16 = 16MB per chip.

Mystery wrote:

Try different bios settings that don't stress the RAM as much. Maybe play around with Write-Through/Write-Back (if it's available).

I flashed the latest BIOS I could found, from this page. These are the BIOS options:

Mystery wrote:

You didn't mention the CPU you're using. If it's one without L2 cache you'll also run into performance issues, especially in Windows, since you're very likely to have large amounts of uncached RAM.

Currently a K6-2/500AFX. This M577 mobo has 1MB of onboard L2 cache, and seems it has tagram for caching only 256MB of RAM.

I'm after a K6-2+ or K6-3+ but haven't got one yet.

TELVM wrote:

I'm after a K6-2+ or K6-3+ but haven't got one yet.

aren't we all... 🤣

anyway. I never had any luck with socket 7 boards and pc133 ram. For some reason, they just got along like Chinese and Japanese people. Pc100? Stable as the day is long. If I tightened the ram timings of pc133 to 2-2-2 then usually things worked out but it was never 100%

There is something inherently incompatible with late (large) pc133 modules and Super 7 boards.

That said, my asus P5A-B is a LOT more tolerant of pc133 ram than any other board i've tried. It might be it defaults the ram .2V over spec, I don't know but it is the most forgiving board I have.

luckybob wrote:

aren't we all... 🤣

I have seen this on ebay:

http://www.ebay.com/itm/AMD-K6-2-Speed-up-Upg … =item4158322b92

But smells a bit weird (28000 in stock? 😳 )

luckybob wrote:

... If I tightened the ram timings of pc133 to 2-2-2 then usually things worked out but it was never 100% ...

Nice tip, thanks Luckybob. I've just pulled out the 2x 32mb PC-100, inserted 1x 256MB PC-133, set 'SDRAM cycle lenght' to 2 in BIOS, and W98SE starts fine 😀 . Torturing now with Prime95 in 'blend' to test RAM stability, so far so good:

TELVM wrote:

The M571 and the M577 were exceptions to the PC Chips rule.

I remember that page, they preferred something VIA over the i440BX: Insane!

Yeah well, I guess you always run memtest before installing any software?
So does memtest run through? 3 to 5 passes at least?

If not and you have only limited memory sticks available you can still try to use only one and then try each slot separate.

of course mvp3 chipset support 256*3 mb of ram, but for the board maybe you just need a bios update.
in my opinion the mvp3 needs no more than 256mb, because it can only read ~260mb in a second.

but, it can still store more data in the other ram at the same time possible increasing some reads as it doesn't have to write first as it still is in another stick.

I don't see a strict relation between what a chipset memory controller is able to read in 1s to the maximum senseful amount of RAM available. Usually a program doesn't reads/writes all RAM at the same time, so even if accessing all RAM requires more than a second the part a program effectively uses is typically much smaller. More RAM allows you to switch to different tables quickly, because the alternative would always mean to use paged out RAM from a storage device and this is more than 10.000 times slower than RAM.
With tremendous effort you could even visualize RAM access patterns, if you log the accessed memory locations (by sampling the address lines f.e.) you could plot this for a selectable time interval. An easier but less exact method is running the program in debug mode, where each access can be logged by software and translated to physical addresses.

As mystery already pointed out SS7 chipsets usually can use only one internal memory module layout at the maximum allowed capacity. As a rule of thumb, you need double sided modules.
And then always the memtest check follows, especially if it is a uncommon configuration and problems are to be expected. Pushing a SS7 chipset to maximum memory is a uncommon configuration.

If this continues to give you trouble, I have had a SS7 board which didn't like Write Cache Pipeline enabled when AMD CPUs were used. This particular motherboard had seperate options for both Read Cache Pipeline and Write Cache Pipeline, and it was the write one which caused issues in Windows.

Your motherboard's BIOS seems to have a similar, but combined, option, called cache Rd+CPU Wt Pipeline.

I should have done as Elianda says and pass memtest at the beginning, it turns out one of the three sticks was simply FUBAR. This system works flawlessly with the other two sticks for 512MB RAM:

And with a third non-FUBAR stick for 768MB RAM:

Time now to ged rid of this tagram-cacheable RAM dinosaur manure with some III / 2+ / 3+ 🤣 .

Yes as was suspected CTCM7 says this M577 mobo can't cache more than 256MB of RAM 😒 :

I seem to recall CTCM7 reports 508 or 510 MB cacheable when you have 2 MB of cache on a VIA MVP3 board.

Argh, I missed two K6-2+/550 today in ebay.de 😢 . Somebody paid 32€ (plus 6€ delivery) for one of them, that's fifty US bucks, this is Sparta!

And my K6-2/500 refuses to work stable at 6x100 even under serious electroshock 😵 .

Patience + CPU-World = cheap(er) CPUs.

Thanks for the tip Feipoa 😉 .