Jump to content

Tearing my hair out over memory problems!


whocrazy

Recommended Posts


erpdude, I read of a couple of guys who said they were running 2gb, but they had 64 bit processors[1 AMD,1 Intel] & I ain't there yet :no: Time I get there they'll have 256 bit processors :whistle:  :D

ssmokesee, I'm currently running 1gb ddr 2700[Kingston x 2-512's], works great!

I know I have heard 1GB works ok. I think I remember reading on this forum about problems with anything greater than 1GB...could be mistaken though. Now that I remember back to my system.ini file from a 1.6.2 SP install, it did have "MaxPhysPage=40000", or something very close to that. The setting effectively limited the usable physical memory to 1GB. I guess I answered my own question.

Link to comment
Share on other sites

I know I have heard 1GB works ok.  I think I remember reading on this forum about problems with anything greater than 1GB...could be mistaken though.  Now that I remember back to my system.ini file from a 1.6.2 SP install, it did have "MaxPhysPage=40000", or something very close to that.  The setting effectively limited the usable physical memory to 1GB.  I guess I answered my own question.

Ya,we started out banging our head against 512 limit, then 768, now its 1gb. With fine people like Gape, MDGx & the rest of the people that contribute we may see that barrier fall, who knows :)

Link to comment
Share on other sites

i finaly have a 64 bit processor but i took 98 off completely and gone to xp.i was afraid 98 would not work on it sides before the end it gave me some troubles that i didn't wanna bother fixing.

great, soldier1st. a 64-bit cpu! awesome! now you can use the 64-bit edition of Windows XP. Win98 is totally a 32-bit OS and may or may not work on a 64-bit machine.

yeah, those 512 Mb+ memory barriers will be broken with those tweaks. hopefully those higher capacity memory chips will be cheaper and more affordable so we can put more RAM onto our machines.

Link to comment
Share on other sites

great, soldier1st.  a 64-bit cpu!  awesome!  now you can use the 64-bit edition of Windows XP.  Win98 is totally a 32-bit OS and may or may not work on a 64-bit machine.

yeah, those 512 Mb+ memory barriers will be broken with those tweaks.  hopefully those higher capacity memory chips will be cheaper and more affordable so we can put more RAM onto our machines.

epdude8, I personally know 1 person[lucky dog] who has been running 98se on a 64bit cpu[w/usp 2 rc1 & MDGx's 982me] for about a mo now. His opinion: "it rocks my socks". So it can be done. He's only running 1gb ram, trying to find the tweaks that'll allow a bump to 2gb :D

Addendum: This guy is also a little on weird side, went thru nasty divorce & turned him into Howard Huges type :blink:

Edited by randiroo76073
Link to comment
Share on other sites

epdude8, I personally know 1 person[lucky dog] who has been running 98se on a 64bit cpu[w/usp 2 rc1 & MDGx's 982me] for about a mo now. His opinion: "it rocks my socks". So it can be done. He's only running 1gb ram, trying to find the tweaks that'll allow a bump to 2gb :D

Addendum: This guy is also a little on weird side, went thru nasty divorce & turned him into Howard Huges type :blink:

w98se can run on a 64bit processor though it'll run it in 32-bit emulation mode.

performance should not be a problem with all that RAM installed.

hope that guy you talked about can recover and put his life back together.

Link to comment
Share on other sites

  • 1 year later...
ALL MEMORY AGES AND GETS SLOWER...

maybe. or maybe not. we shall see. I havent seen or experience any of the memory modules I've used on my computers get any slower. perhaps is US (computer users) are getting faster and have a perception that memory modules are moving a little slower.

we do live in a fast changing world today.

Edited by erpdude8
Link to comment
Share on other sites

ALL MEMORY AGES AND GETS SLOWER...

maybe. or maybe not. we shall see. I havent seen or experience any of the memory modules I've used on my computers get any slower. perhaps is US (computer users) are getting faster and have a perception that memory modules are moving a little slower.

we do live in a fast changing world today.

Glad your memory modules aren't too unreliable as they age ..... AND GET SLOWER.

erp:

You apparently don't get the point:

Your memory chips are getting slower; they all do. If you bought reputable modules, they were rated for the slower speed they become, not the fleeting-fast you get when they are new.

Your machine does NOT change the access speed to the memory at all! Your USAGE of the memory never changes [unless ...] but the chips are getting slower.

Your machine is likely one of the overwhelmingly large number of machines that have no manual memory timing capability. What happens instead is there is an SPD chip on the DRAM board that the CPU and BIOS POST program reads to dynamically determine what the speed of the memory [apparently] is.

If you have a reputable chip, then the SPD timing is consistent with where your chips are when they are old, and still some additional "headroom" to perhaps age further, all this harmlessly.

So, in this case [consider yourself fortunate] then no matter how old or slow your memory gets, it still meets or exceeds the requirement of the overall system as reported to same by the SPD chip. But clearly, it is getting closer to merely passing the requirement; when it was new it passed it with flying colors, but no more.

The problem is that there are sleezeballs in this world that sell memory that is NOT capable of long-term use [i could mention brand names, but let's just say that companies with three random letters in their name as a group have been the most problematic over the years]. The SPD chips on these modules are too ambitious; the chips fade and the module becomes the source of erratic behavior; the system cannot slow down to match the newer less-capable SDRAM because it is obligated to honor the timing offered up by the SPD.

In an honest world, these parts should never have been sold together. Sometimes you find chips honestly rated lousy, such as CL3 instead of CL2.5. Clearly these slow your machine down from day one as compared to the 2.5 ones, but perhaps it's an honest rating that will always be met in the long run. Clearly the same chips should not be soldered onto a board where the SPD says this is a CL2.5 overall device!

I had an encounter with a machine based on an AOPEN board which actually supported manual timing to override the SPD chip. It was getting really flakey, and I hadn't been to it since it became about 2.5 years old and quite unreliable. [Previously I was there when it was 6-8 months old and not a problem.] As a bandaid, I was able to retard the memory timing, over-riding the SPD timing. The machine was noticeably different - quite slow and quite reliable. A good trade-off temporarily. I came back later with good memory; it worked fine and just as slow; then I enabled the timing-by-SPD and it got faster again, but continued to work reliably, as expected. [All this happened about 8 months ago; apparently still working fine; the old memory was made by a three-letter-company, etc.]

With regard to laptops specifically PC-100 SDRAM: There are some problematic models of laptops that, while rated expressly for PC-100, do NOT like just any old PC-100! This is not an SPD issue [which is probably also present!] but an aberration of certain Intel chipsets.

Basically, PC-100 SODIMM laptop memory comes in two flavors: One is noted by the fact the board is filled with a bunch of chips that are clearly quite rectangular, generally standing vertically next to each other, and probably at most 8 on a side [assuming there is anything on the back side]. If these are 128 MB or larger, they may not reliably work! [Again, it depends on the precise implementation of the chipset on the mobo.]

The one that will ALWAYS work [assuming the SPD problem above is honestly dealt with] is the one where there are 16 little chips on a side, where each one is essentially a little square chip [which is pretty much the only way to get 16 on a side in that tiny SO-DIMM package!].

Generally, this version costs more; many places sell the former for less than half of the price of the latter! But if your system is one of those that requires the latter, you are wasting your money; they will fail quite quickly, perhaps even on day one, but more likely within 6 months to a year, NOT because they are fraudulently rated, but because the chipset requires "low-density" SDRAM and the 8-chip variety is "high-density". [slightly more technical detail: Chips are organized in a Zx1 and a zx4 configuration internally. You get to the same place, but in one arrangement, the physical chip is supplying 4 bits at once, containing 1/4 the address space, while in the other arrangement, each chip supplies 1 data bit, but there are 4 times as many addresses.]

The downfall of the x4 [high-density] chips is that they are inherently slower to access, and this is NOT accounted for in the SPD. If the memory socket and support chips want the low-density only, then the high-density will be either flakey, short-lived until flakiness, or downright just won't work even when new depending on circumstances.

In the [now somewhat obsolescent] world of PC-100 SDRAM DIMM modules, the same problems happen, but generally there isn't a price gouge assocatiated with specifying low-density. Another watchword could also be that the module is PC66 or PC100 compatible, which is NEVER the case with the high-density [assuming the info is accurate!] I have a system based on a PII-333 that loves PC-66 and low-density PC-100 but totally hates high-density PC-100.

cjl

Link to comment
Share on other sites

I don't believe RAM can age and become slower. They are, after all, only solid-state devices. There is nothing to "wear out", neglecting the effects of electromigration.

I have an XT 286 from the mid 1980s with original RAM, and it still works as it always did (the hard drive, however, is another story...)

Link to comment
Share on other sites

Yeah, the little mice running in the ram don't get tired. :D My Amigas still run top notch the same as when I first purchased them. One still has the orig. HD that I installed in it. Maxtor 540 meg scsi 10+ years. :thumbup:thumbup:thumbup

jd

Link to comment
Share on other sites

I don't believe RAM can age and become slower. They are, after all, only solid-state devices. There is nothing to "wear out", neglecting the effects of electromigration.

I have an XT 286 from the mid 1980s with original RAM, and it still works as it always did (the hard drive, however, is another story...)

You can keep your beliefs. I'll stick with the information from the chip manufacturers. Actually, you do have it somewhat right.

This is not a matter of "wearing out" as much as migration. The net effect is that undesirable capacitance builds up and the absolute maximum speed the chip could theoretically be accessed at becomes lower. Most of the effect occurs in the early years and mostly while the chip is powered on [as opposed to just lying around or being ambiently heated; however higher temperatures while powered on can exacerbate the situation, this is the basis of accellerated testing.]

Of course, if the SPD chip is limiting required access speeds to far slower than that, you never have a problem!

Back when my IBM PC II was new in the early '80's with original RAM, each motherboard was "burned in" with the entire board subjected to testing. As machines got newer, less testing was done. I suspect your XT 286 board was still using burned-in RAM that was thoroughly tested for ultimate [and lowered] speed, and then timed to match.

Back then, there was always some joker who thought that it was cool to set less wait states than the chips were marked for because "it worked". Of course it only ACTUALLY worked for some months! [The fix, of course, was to honor the RATED speed, not the casual observation speed!]

The sleezeballs from back then sold reject memory and people were duped into using it as if the markings could be trusted. These machines failed because the real speed of the memory was such that when new it barely passed, and as it aged, it went into unreliable territory at the claimed speed. In some machines, you could have wide lattitude on setting wait states to compensate, thus you got a slower machine but at least it worked. If the memory was that bad, then you were SOL for buying crap quality, etc.

Ever since we all switched to SDRAM with the little SPD chips on the memory board, all of this should be moot. But this doesn't stop Mr. three-letters and his brethren from making crap memory boards that claim the eventual speed to be the same as the when-new speed, and without a mobo that can compensate [and slow down] you are SOL with them when they age, etc.

cjl

Link to comment
Share on other sites

Yeah, the little mice running in the ram don't get tired. :D My Amigas still run top notch the same as when I first purchased them. One still has the orig. HD that I installed in it. Maxtor 540 meg scsi 10+ years. :thumbup:thumbup:thumbup

jd

Expected value. When chips aren't asked to do the ultimate impossible, they continue for centuries to do the possible.

Today memory isn't even unit-tested. All is done by statistics. Pull out a few random samples, subject them to accellerated testing [probably to the point of destruction!] and in essence find out the ultimate slowed-down access speed that can be reliably depended on when the chips are old.

Then test the entire lot merely for current access speed and basic functionality. Assume the entire lot will behave as compared to the samples subjected to the intense testing, perhaps with a tiny derating safety factor to hedge against the possibility that the samples weren't actually representative, but were perhaps a little "hotter" than the rest, etc.

Given the amount of powered-on hours on average we put computers to, an honestly rated system that has worked fine for three years should never have a RAM problem due to aging.

That said, the occasional cheater still comes up where the memory ages below the point of reliability. [Please note: I have already changed the RAM on several box-makers machines due to proven failure; none of them were as much as three years old. I guess the memory is rated to get past the warranty period and then crap out!]

In the world of laptops, there is a pervasive conventional wisdom about a quality point of "new memory" which of course begs the question as to how the original memory came to be known as "old memory". I had to help someone with a ThinkPad that was failing. The memory turned out to be from mr alphabet-soup instead of the original IBM [and it could have been and should have been an upgrade, but instead the original was removed and replaced with the third-party twice as large; this smells because if the original memory was present, the cost to upgrade to double memory should be less than replacing with the "new" one, etc.]. I suspect the eBay seller of this machine was creating a dichotomy of machines with IBM RAM culled off of such as this one, doubling up on IBM RAM on reliable models, and selling off the "new" memory machines for more memory because they were "better" etc.

cjl

Link to comment
Share on other sites

IMHO:

To my knowledge memory chips [any generation] do not "get old".

RAM components [even cheapo ones] are just electronic chipsets + copper/similar alloy PCB embedded wiring, they do not change configuration as time goes by.

Except for [as LLXX put it so well]...

EMI [electromagnetic interference]/ESD [electrostatic discharge]/EMP [electromagnetic pulse]

interference/discharges/pulses/fields which may sometimes damage/"kill" electronic components due to electromigration, but that is highly unlikely.

Such fields/discharges must be really strong and in close enough range to do real damage.

I've been building computers for over 10 years, and have seen ESD doing damage only once: killed a hard drive. :(

Although I've lost a few VHS tapes because I didn't pay attention, and placed them too near [< 3 feet (1 meter ~ 1 yard)] to my 18 inch (46 centimeters) woofer speakers. :lol:

Another example...

when you overclock your CPU/FSB/RAM/mobo/GPU/video/etc, in time overheating/improper cooling may "kill" certain silicon microcomponents due to electromigration.

But this process is also highly unlikely to happen in a short amount of time. Electromigration effects are usually seen after many years of usage.

Most of the time the user is unaware of such "micro-damage", and he/she may go ahead and upgrade/build/buy a new computer way before this happens at such extent, that is when the computer becomes unusable/unstable/unreliable.

HTH

Edited by MDGx
Link to comment
Share on other sites

Not a p***ing contest, but I have been working on the hardware of computers since 1968, and PC's in particular since 1983.

Much of what MGDX says is true; he should listen to his own post.

The electromigration problem is exactly what "ages" the chips. It happens gradually over a period of months and years, assuming the ambient temperature is maintained in the usual manner within typical computer cases.

Moreover, it is totally predictable for any temperature and this principle is what is used to test RAM chips as they are made.

To ensure reliability, sample chips are tested using accelerated methods, which is just another way of saying they HEAT UP the chips, very similar to what overclocking does, as MGDX describes. The net effect is the same, namely a chip whose absolute access time is now longer/slower. The heat can be obtained by putting it into a temp-controlled oven, over-volting/over-clocking, or a combination of techniques. No matter how you get there, the chips are now "aged" essentially simulating the same effect that would happen over a larger course of time at more moderate temperatures.

Once "aged to perfection" by these methods, the access time to reliably get data from the chips is clearly worse than when it was new, sans electromigration that hasn't happened [yet]. Reputable memory makers use this ultimate access time to be programmed into the SPD chip on the DIMM module [including hopefully a little safety factor in case the tested chip is NOT quite representative of its batch-mates, since only samples are taken, NOT every chip!]. The result should be a memory module that will not ever fail in normal use for the foreseeable future. Hopefully this is what most of you see in your computers. As such, while the absolute access time [when NEW] far exceeds what is being required of it, as the chips age, the absolute access time is longer, but never so long as to tax what is demanded of it by the SPD chip, set to require merely the ultimate longer time due to the aging process, etc.

The problem is there are sleazy manufacturers, some with cutesy names with three letters in them, that violate the entire process. They buy what amounts to reject quality or marginal quality stuff that will NOT hold up with time. Simply put, the ultimate absolute access time violates the value programmed into the SPD chip, so eventually these crap parts WILL fail due to aging. They use parts that POSSIBLY could have been fine, had the SPD chip been programmed to properly account for the predictable degradation, but then some people would notice the slower performance from day one because the timing is now much slower than you would want.

Back in the "old days" [in the '80's], every board had wait-state jumpers to accomodate almost anything you could get for the boards. [Apparently few of you remember the boards with the multiple rows of chips which had capacities like 256K x 1 and 64K x 1. Yes, these are socketed chips with a total capacity of 64K for the small ones and 256K for the large ones. Do the math and see how many sockets it takes to FULLY populate a board out to a mighty 640K [the most memory ANYONE would ever need! Or so said IBM back then!] using a bunch of 256K chips to get to 512K bytes, and the 64K chips to get to 640K, etc.] For motherboards like that, you had to conservatively set the wait-state jumpers for the RATED memory speed, but there always was a joker or two who thought they could get "something for nothing" and just hand-waved settings that would get a faster machine that worked for a few months, only to have to set it to what was recommended in the first place in order for the machine to even work at all! Again, normal migration aging at work, etc.

And of course, there were some unscrupulous memory suppliers as well [not necessarily even having names, alphabet soup or otherwise!] selling essentially reject chips that BARELY worked as marked, only to age into total unreliability or even inability to even boot up, etc. Some were so bad that setting the absolute maximum wait-states couldn't make it work properly, such was the brazenness of selling junk, etc.

We went through a period of CPU counterfitting as well [i believe recently with regard to some AMD CPU's being mis/re-marked] back in the Pentium days. Essentially, P-75 was being passed off as a P-133 by removing Intel's speed rating silk-screen and replacing it with the bogus over-rating.

Some of the chips work perfectly fine as P-133, simply because, as new, migration hadn't set in yet. As the chip ages, it becomes impossible to use at that much of an overclock, and will work perfectly fine, but only as a P-75 as it REALLY originally was rated. [i once had one of these fakes in my hand. The subtle giveaway is that REAL Intel speed marking is done LATER and as such, the silk-screen never quite lines up with the rest of the markings on the chip; go look at any of the grey ceramic P chips [before the green ones]. But the counterfitters used bogus marking machines that ALWAYS lined up, etc.]

Electromigration is normal, expected, and can be dealt with. For the most part it is so. As such, most CPU's and memory don't ever see reliability problems, because the aging is built into the design. For the rest, you get fore-shortened product life as you are cheated. In a world with automatic SPD timing, it's easier to rip people off, and it is happening to a few, but just enough that you need to be aware of it.

cjl (using knowledge, experience, and facts, not intuition)

Link to comment
Share on other sites

erpdude8... how do you rate MemTest86.... somewhere in between those 2?

I did get a chance to test out MemTest-86 horsecharles. Pretty good. A little better than the MS Windows Memory Diagnostic tool. Not sure with RAMExam. The makers of RAMExam have a demo but the demo version is nearly worthless.

as for Fredledingue's question, the SP upgrade may support up to 2 Gb of RAM. Win95, 98 & ME can theoretically access up to 2 gigs of RAM and no higher than that.

Ah, that MemTest86 memory diagnostic tool is a major keeper. I've tested this tool on all my computers whether they have old or new memory chips. Kudos to horsecharles for introducting me to it.

Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...