Buying Kingston RAM chips...

[Wald]

Hi group,

The present situation here at work is that almost half the people
work on newer, powerful machines, and the others have less powerful,
older machines. To be more precise, this mostly concerns Compaq
Deskpro's with Pentium II/III processors ranging from 300 to 700 Mhz.

Since it seems not very likely that there will be a total upgrade
soon, I was thinking of buying extra memory chips for these older
machines, since they all still run on 64MB.

I was planning to order chips from Kingston
(http://www.kingston.com), since the majority of the machines already
have Kingston chips, but I'm a bit concerned for compatibility.
Some questions:

- will these guys deliver me the right kind of chip for each machine?
Anyone experiences with Kingston regarding this? I understand Compaq
is quite sensitive for incompatibilities, so I guess I should watch
out.

- some machines have non-Kingston chips (usually 1x64MB). Will it
work if I put an extra Kingston chip in those? Or should I go for one
brand per machine?

Thanks for any answers,

Wald

 

[Pen]

I'd recommend crucial.com, since they will guarantee
compatibility with your machines.

 

[Phrederick]

The BEST solution would be...

- Take half the PC's and pull the memory out of them and put it in the other
half.
- Install new memory in the empty PC's.

....if PC's have three slots, you could empty 2/3's of the PC's to bring the
first third up to 192meg and fill the empty PC's with new stuff.

 

[larrymoencurly]

I was planning to order chips from Kingston (www.kingston.com),
since the majority of the machines already have Kingston chips,
but I'm a bit concerned for compatibility.

- will these guys deliver me the right kind of chip for each
machine? Anyone experiences with Kingston regarding this?

- some machines have non-Kingston chips (usually 1x64MB). Will it
work if I put an extra Kingston chip in those? Or should I go for
one brand per machine?

I've had very few problems using different brand memory modules and
modules with different brand chips on them, but I wouldn't buy
Kingston modules except locally from a dealer that gives 100% refunds
because I've had compatibility problems more recent Kingston ValueRAM,
and Kingston hasn't been able to explain why, and they've provided
contradictory information.

About six months ago, I bought some Kingston ValueRAM from OfficeMax.
It had two different labels on it, KVR-PC2100DDR/256 and
266X64C25/256, assembled in USA, and it worked fine when I tested it
with both www.memtest86.com and www.goldmemory.cz. But a month later
I bought another module, also from OfficeMax, labelled
KVR-PC2100DDR/256 and PC2100DDR/256 and assembled in China, and it
failed GoldMemory in a few minutes and MemTest86 in 3-5 hours, the
same bit in every bad location. I tried another one of these modules,
and it also failed, but in different locations and with different
bits. I tried all sorts of BIOS settings with both modules, but the
only thing that eliminated all the errors was slowing the memory bus
speed from 266MHz down to 200 MHz. Kingston told me that the latter
part was a) exactly the same as the former, or b)slightly different
than the former and not made for my mobos (ECS K7S5A and P4S5A), and
they offered to exchange it for the former. I didn't feel like paying
to ship something that was free after rebate, so I just got a refund
from OfficeMax. A week ago, I bought yet more Kingston ValueRAM from
OfficeMax, again labelled KVR-PC2100DDR/256 and PC2100DDR/256 but
assembled in USA, and it worked OK with the K7S5A, but with the P4S5A
the default BIOS memory speed setting of "FAST" had to be changed to
"NORMAL" or "SLOW" to prevent errors. Again, Kingston offered to
replace the memory, but their tech support spoke only in generalities
and couldn't explain the problem.

With all three types of these memory modules, the circuit boards
seemed to be 100% identical, but I don't know about the chips because
they had no manufacturer markings, only "32M x 8 DDR 7-1", and I think
they come from different manufacturers since the print looks slightly
different for each.

It's important to run any memory diagnostic for several hours because
the second Kingston module from the second purchase tested fine for
the first few hours. Also run more than one diagnostic because
GoldMemory found errors very quickly with my earlier modules while
MemTest86 took forever to find any, but for my last memory module,
GoldMemory passed it, even with faster BIOS settings, while MemTest86
consistently found bad bits at even the default settings.

The only other DDR memory I've bought recently has been K-Byte/RK-Byte
and cheap Fry's Electronics no-name stuff, and except for an early
K-Byte module containing Spectec chips (Micron's brand for second-rate
and used chips), it's all worked perfectly. So I'd say that Kingston
can't complain about my mobos being very particular about their
memory. BTW, the K-byte modules with Elixir chips (Nanya's
second-rate brand) seemed to be underrated the most and tested fine
except when I set all the BIOS timings at their fastest settings (BIOS
limited to clock to 266 MHz max.). With all the memory I've bought,
the only modules that have never given me problems were those with
chips labelled with the chip maker's primary brand (i.e., Micron, not
Spectec for Micron).

It's important to buy from a good dealer. OfficeMax was always very
accommodating with my purchases, even when they gave all my money
back. But if you buy from CompUSA, expect to pay a 15% restocking fee
for opened merchandise, even if every module you get from them is
defective, and also expect them to claim that nothing is wrong because
it passes their cheap memory tester. I've found that CompUSA will try
to charge the 15% even if you didn't open the box -- they are not a
very honest company.

 

[kony]

On 4 Dec 2003 08:37:51 -0800, (e-mail address removed)
(larrymoencurly) wrote:

...it
failed GoldMemory in a few minutes and MemTest86 in 3-5 hours, the
same bit in every bad location.

Sounds like a motherboard problem, not a memory problem.
A memory problem should have the same location, different bits, not
same bit, different locations.

Did you test only the new module in same slot old module was in or all
of them?

I tried another one of these modules,
and it also failed, but in different locations and with different
bits. I tried all sorts of BIOS settings with both modules, but the
only thing that eliminated all the errors was slowing the memory bus
speed from 266MHz down to 200 MHz. Kingston told me that the latter
part was a) exactly the same as the former, or b)slightly different
than the former and not made for my mobos (ECS K7S5A and P4S5A), and
they offered to exchange it for the former. I didn't feel like paying
to ship something that was free after rebate, so I just got a refund
from OfficeMax. A week ago, I bought yet more Kingston ValueRAM from
OfficeMax, again labelled KVR-PC2100DDR/256 and PC2100DDR/256 but
assembled in USA, and it worked OK with the K7S5A, but with the P4S5A
the default BIOS memory speed setting of "FAST" had to be changed to
"NORMAL" or "SLOW" to prevent errors. Again, Kingston offered to
replace the memory, but their tech support spoke only in generalities
and couldn't explain the problem.

This is very significant. EXTREMELY important. You should have never
set the BIOS memory to "FAST" manually unless you know the exact
chipset and memory timing specs, have previously compared them. If it
defaulted to "FAST", it shouldn't have. This is clearly a motherboard
or BIOS problem, not a memory problem. Budget memory is not supposed
to be able to run the FAST timings, that's WHY it's budget memory.
It is incredible to me that you were trying to get Kingston to deal
with this. There is nothing wrong yet you're trying to get them to
upgrade your memory. Changing the timings is essentially
overclocking, something which no memory manufacturer guarantees. When
you contacted Kingston did you specifically mention that you didn't
want to run the memory at it's spec'd speed, that it was only
producing errors when the timings were changed? What do you think
they would say to that?

There is only one situation where you should ever be changing those
settings from their slowest default/SPD values, when you have read
the SPD info from the memory and deciphered it yourself, confirmed
that when the motherboard uses the new memory timings, they are still
within the guaranteed spec of the memory.

Suppose the following:

You have memory that runs perfectly at it's spec'd speed and timings,
but the motherboard is instable with memory that doesn't have a large
margin of stability beyond that spec, that motherboard actually needs
to run memory at a lower spec to stay stable. Which motherboards
might have these issues? Any could, but it's a whole lot more likely
on low-end boards like you mentioned above. In particular many ECS
boards have questionable G-Luxon capacitors, which are among those
considered defective by many people... I've replaced a few dozen
myself and the boards worked better afterwards.

A cheap motherboard may run OK with good memory, or a good motherboard
with cheap memory, or an average board with average memory, but when
you cut too many corners you're asking for trouble.

Millions of Kingston memory modules are sold, but only a few people
have problems... if it was the memory then you'd be hearing of a LOT
more reports of problems. What's the varible? Motherboard, number of
modules, BIOS settings.


Dave

 

[Guest]

Sounds like a motherboard problem, not a memory problem. A
memory problem should have the same location, different bits,
not same bit, different locations.

The problem occurred at only a few addresses and never when the oldest
Kingston module (KVR-266X64C25/256) or the K-byte module with
Nanya/Elixir chips were tested.

Did you test only the new module in same slot old module was
in or all of them?

I always test in every slot, if possible (my Intel i810 chipset didn't
allow that for single modules), and if the module passes in all of
them I then test with other modules installed as well, trying each
module in every slot.

This is very significant. EXTREMELY important. You should have
never set the BIOS memory to "FAST" manually
If it defaulted to "FAST", it shouldn't have.

It did -- two BIOSes, defaults after CMOS battery was discharged, and
I have at least one FIC or Soyo mobo that uses the same setting for
its BIOS (PC100 or PC66, not PC2100).

Budget memory is not supposed to be able to run the FAST timings,
that's WHY it's budget memory.

Then why did even cheaper K-byte and no-name Fry's memory work fine
that way? Some of the Fry's memory I've gotten has been really
questionable, but so far, all the DDR modules have tested OK.

It is incredible to me that you were trying to get Kingston to deal
with this. There is nothing wrong yet you're trying to get them to
upgrade your memory. Changing the timings is essentially
overclocking, something which no memory manufacturer guarantees.

It's incredible that you ignored what you quoted of my own message:

IOW I tried the _slowest_ settings as well, but the memory still
failed. I did try faster settings because I once had some PC133 CAS3
memory that wouldn't work in a VIA PC100 mobo except at CAS2, but this
also didn't help with the DDR.

I'd never expect a manufacturer to be responsible for performance
beyond specs, but memory should work fine even at the BIOS' defaults,
and I'd insist on that because BIOS parameters are sometimes reset
automatically. I'm not an overclocker -- the fact that my fastest
computer is just an XP1800+ with a TNT2M64 card should indicate that,
and the only times I've even tried overclocking was while testing
memory, and that's only for curiousity (IOW if memory fails only while
overclocked, I don't consider it a real failure). My concern is only
about reliability.

you contacted Kingston did you specifically mention that you didn't
want to run the memory at it's spec'd speed, that it was only
producing errors when the timings were changed? What do you think
they would say to that?

I said that their memory failed even at the slowest settings unless
the bus was slowed to 200 MHz. They couldn't explain anything but
just gave the usual sales type excuses.

Suppose the following:

You have memory that runs perfectly at it's spec'd speed and timings,
but the motherboard is instable with memory that doesn't have a large
margin of stability beyond that spec, that motherboard actually needs
to run memory at a lower spec to stay stable. Which motherboards
might have these issues?

Any could, but it's a whole lot more likely on low-end boards like
you mentioned above. In particular many ECS boards have questionable
G-Luxon capacitors, which are among those considered defective by
many people... I've replaced a few dozen myself and the boards worked
better afterwards.

The only memory modules that showed errors with these mobos were
K-byte with Spectec chips and later Kingstons. These are obviously
not prime memory modules because their chips are not labelled with the
chip manufacturer's standard part numbers but are either used or
second grade.

On my K7S5A Pro and P4S5A, the Luxons are only for bypass (green with
silver printing), and I don't think they're the low-ESR types, the
only ones affected by the Taiwan capacitor scandal. The low-ESR ones
in at least one of my ECS mobos are "Ost" brand, purple with gold
printing on them, but I don't know if they're involved in the scandal.
I took ESR measurements and found nothing, but I can't be sure the
measurements were valid because the capacitors may be in parallel.

Millions of Kingston memory modules are sold, but only a few people
have problems... if it was the memory then you'd be hearing of a LOT
more reports of problems. What's the varible? Motherboard, number of
modules, BIOS settings.

Kingston is selling junk with unlabelled chips and can't even tell me
who's making their chips. Does that seem like a quality operation to
you? Most people don't test their memory thoroughly before putting it
to use, and in the case of one of the Kingston modules the errors
occurred only above the 220MB-240MB area, so it's possible that this
would rarely be used by Windows (but I don't know how Windows
allocates memory).

 

[kony]

Then why did even cheaper K-byte and no-name Fry's memory work fine
that way? Some of the Fry's memory I've gotten has been really
questionable, but so far, all the DDR modules have tested OK.

Some may work fine because that's the situation with budget memory,
some of the chips are actually capable of higher, tigher timings than
others. It is only guaranteed to work at default SPD timings in a
motherboard conforming to spec as well. If the motherboard isn't
within spec, it may require higher quality memory... this is also a
known problem with ALL nForce2 boards.

It's incredible that you ignored what you quoted of my own message:

Sorry, I didn't draw a conclusion because "all sorts of" didn't tell
us anything specific. It could be that you just happened upon a bad
module... it happens, obviously it's not cost-effective for a
manufacturer to do as much testing on very cheap memory. On the other
hand, if the situation is reproducible with several modules it would
suggest a problem with the motherboard, since many many other people
use these modules without issue.

IOW I tried the _slowest_ settings as well, but the memory still
failed. I did try faster settings because I once had some PC133 CAS3
memory that wouldn't work in a VIA PC100 mobo except at CAS2, but this
also didn't help with the DDR.

Fair enough, it does sound like you did all the testing possible on
that board, but it doesn't rule out the board itself.

I'd never expect a manufacturer to be responsible for performance
beyond specs, but memory should work fine even at the BIOS' defaults,
and I'd insist on that because BIOS parameters are sometimes reset
automatically. I'm not an overclocker -- the fact that my fastest
computer is just an XP1800+ with a TNT2M64 card should indicate that,
and the only times I've even tried overclocking was while testing
memory, and that's only for curiousity (IOW if memory fails only while
overclocked, I don't consider it a real failure). My concern is only
about reliability.

Of the low-end modules, the ones I see with the least headroom for MHz
or timing increase are those with the cheaper PCB including film
resistors, not surface-mount parts, like this (thought it's of a PNY
PC2100 module):
http://69.36.189.159/usr_1034/pny-resistors.jpg

All the other Kingston PC2100 modules I've tried will do at least
166MHz @ 2.5,3,3,6 except for one that can only do ~ 145 or so,
providing only 1 or two modules are used. That is in multiple
motherboards of Via, Sis, and nForce chipsets.

I said that their memory failed even at the slowest settings unless
the bus was slowed to 200 MHz. They couldn't explain anything but
just gave the usual sales type excuses.

Unfortunately it's unlikely that you were speaking to anyone with
enough technical background to help unless they forwarded the call a
couple times, most likely you know a lot more about the issues than
whoever you were talking to.

The only memory modules that showed errors with these mobos were
K-byte with Spectec chips and later Kingstons. These are obviously
not prime memory modules because their chips are not labelled with the
chip manufacturer's standard part numbers but are either used or
second grade.

What were the numbers on the chips? I may have some, or a different
lot... all of those I have/had will do up to ~147Mhz, usually quite a
bit higher. I'm not surprised that they're not "prime" though, given
the low price. On the other hand I still think the motherboard is
playing a crucial role in the instability.

On my K7S5A Pro and P4S5A, the Luxons are only for bypass (green with
silver printing), and I don't think they're the low-ESR types, the
only ones affected by the Taiwan capacitor scandal. The low-ESR ones
in at least one of my ECS mobos are "Ost" brand, purple with gold
printing on them, but I don't know if they're involved in the scandal.
I took ESR measurements and found nothing, but I can't be sure the
measurements were valid because the capacitors may be in parallel.

The green-with-silver Luxons are "supposed" to be Low-ESR, they're
what's used in circuits clearly needing low-ESR caps, like for the
switching regulation of the CPU. I'm sure that at least one K7S5A
I've had, did have those green/silver Luxons there as well as around
the memory. I've seen Ost caps fail too, but not in large enough
numbers to draw a conclusion about 'em.

Well, actually I do recall having to replace about 9 of the
green/silver Luxons around the regulators on a K7S5A because they
vented... dead board... works fine now at 150MHz FSB with ??? (I
forget, some nearly FAR memory).

Kingston is selling junk with unlabelled chips and can't even tell me
who's making their chips. Does that seem like a quality operation to
you?

The chips that say "Kingston" on them? I USED to know who made
those... time takes it's toll on my memory... might be Toshiba chips,
which have two circular mold marks near the center like this:
http://www.biwa.ne.jp/~yok/DDR-KIBAN1/TOSHIBA75-OMO1.jpg

Most people don't test their memory thoroughly before putting it
to use, and in the case of one of the Kingston modules the errors
occurred only above the 220MB-240MB area, so it's possible that this
would rarely be used by Windows (but I don't know how Windows
allocates memory).

Don't know, maybe it's just a regoinal thing... the Kingston PC2100
modules in your region might be different than those here, which are
(on average) able to do a little over 170MHz... maybe higher but I've
only been using them on older boards with chipsets that aren't very
stable over 180MHz anyway, so i can't draw any conclusions above
170MHz.


Dave

 

[larrymoencurly]

Budget memory is not supposed to be able to run the FAST timings,
that's WHY it's budget memory.

Some may work fine because that's the situation with budget memory,
some of the chips are actually capable of higher, tigher timings than
others. It is only guaranteed to work at default SPD timings in a
motherboard conforming to spec as well. If the motherboard isn't
within spec, it may require higher quality memory... this is also a
known problem with ALL nForce2 boards.

My boards are SiS 735 and 645, and I tried not only SPD timings but
slower ones, too. I think that some of the Kingston ValueRAM didn't
work because Kingston, in an effort to cut costs, lowered the margin
of safety of their memory.

It could be that you just happened upon a bad module... it happens,
obviously it's not cost-effective for a manufacturer to do as much
testing on very cheap memory.

I can understand one bad module (none seemed to have been returned by
other customers), but two in a row, separated by several months, one
soldered in China, the other in the US? I also had two bad
K-byte/Spectec modules in a row.

it does sound like you did all the testing possible on that
board, but it doesn't rule out the board itself.

True, but I also tested the first and second modules with two K7S5A
boards (not K7S5A Pro) that didnt' belong to me, and they acted the
same -- the oldest was fine, the newer one failed at the same address
and with the same bits. These boards use the same SiS 735 chipset but
had different BIOSes, and there have been hardware tweaks made with
the K7S5A.

Of the low-end modules, the ones I see with the least headroom for
MHz for timing increase are those with the cheaper PCB including
film resistors, not surface-mount parts, like this (thought it's of
a PNY PC2100 module): http://69.36.189.159/usr_1034/pny-resistors.jpg

All the other Kingston PC2100 modules I've tried will do at least
166MHz @ 2.5,3,3,6 except for one that can only do ~ 145 or so,
providing only 1 or two modules are used. That is in multiple
motherboards of Via, Sis, and nForce chipsets.

I have a "free" Fry's PC133 with film resistors, but all three
versions of the Kingston DDR modules had identical circuit boards with
surface mount resistors.

Unfortunately it's unlikely that you were speaking to anyone with
enough technical background to help unless they forwarded the call a
couple times, most likely you know a lot more about the issues than
whoever you were talking to.

They forwarded my call only once, to "second level tech support," but
a few months ago I was forwarded a couple of times, but nobody ever
gave me specifics.

What were the numbers on the chips? I may have some, or a different
lot... all of those I have/had will do up to ~147Mhz, usually quite a
bit higher. I'm not surprised that they're not "prime" though, given
the low price. On the other hand I still think the motherboard is
playing a crucial role in the instability.

All the Kingston DDR chips had only "32M x 8 - 7 - l" printed on them,
usually very faintly (invisible except at an angle), but they were
bright on the oldest module (the one that worked fine). The Spectec
and Elixir chips had numbers that at least looked like something put
on by chip manufacturers, the Elixers' being N2DS12H80AT-75B
201Y0XSY.

The green-with-silver Luxons are "supposed" to be Low-ESR, they're
what's used in circuits clearly needing low-ESR caps, like for the
switching regulation of the CPU. I'm sure that at least one K7S5A
I've had, did have those green/silver Luxons there as well as around
the memory. I've seen Ost caps fail too, but not in large enough
numbers to draw a conclusion about 'em.

Well, actually I do recall having to replace about 9 of the
green/silver Luxons around the regulators on a K7S5A because they
vented... dead board... works fine now at 150MHz FSB with ??? (I
forget, some nearly FAR memory).

On my K7S5A Pro and P4S5A, the 3-phase voltage regulator has several
purple/gold Ost brand capacitors around it, with only a couple of
green/silver Luxons. The other voltage regulators, which seem to be
the linear type because I didn't see any coils near them, have mostly
black/white and green/silver Luxons. I may just get replacements the
next time I order a bunch of capacitors
from Mouser or Digi-Key since it's easier and cheaper to replace caps
than transistors and diodes.