pc slows down after ram upgrade

[wg]

Hello,

I have an HP Vectra VL 6/300 Series 7MT. It is a 300Mhz Pentium II with
a total of 3 memory slots. I have two 256 mb modules installed. When I
install a 3rd 256 mb module the machine boots up but it is *extremely*
slow. Booting up and shutting down the machine take forever. I believe
that it took 15-20 minutes (possibly longer, but at least that long, I
didn't time it exactly) to do a shutdown under Linux on the machine. The
only explanation that I can think of is that the motherboard is not able
to cache memory above a certain limit. Although I can't imagine this
causing such a massive slowdown. All the memory modules are identical.
And if I switch the modules to different slots the problem only occurs
when all three modules are installed.

Any one have any suggestions as to what the problem might be and what,
if any, fixes for the problem are available? Would a bios upgrade help?

 

[ihoc-a-attbi-d-com]

| I have an HP Vectra VL 6/300 Series 7MT. It is a 300Mhz Pentium II with
| a total of 3 memory slots. I have two 256 mb modules installed. When I
| install a 3rd 256 mb module the machine boots up but it is *extremely*
| slow. Booting up and shutting down the machine take forever. I believe
| that it took 15-20 minutes (possibly longer, but at least that long, I
| didn't time it exactly) to do a shutdown under Linux on the machine. The
| only explanation that I can think of is that the motherboard is not able
| to cache memory above a certain limit. Although I can't imagine this
| causing such a massive slowdown. All the memory modules are identical.
| And if I switch the modules to different slots the problem only occurs
| when all three modules are installed.

Intel made a rather bizarre design decision in the early versions
of the Pentium II CPU: the L2 cache can only map to the first
512 MB of main memory. If you must have more than 512 MB of
memory, your only option is to replace the CPU with a later
version that can cache unlimited main memory. The 350-450 MHz
versions should work fine, and the "Katmai" Pentium III (450-
600 MHz) will probably work fine too, but the "Coppermine"
Pentium III probably will not work.

Unless you can find a really cheap 350-400 MHz Pentium II
somewhere (Ebay?) or absolutely must have 768 MB, you're
probably better off just staying at 512 MB.

 

[Tony Hill]

| I have an HP Vectra VL 6/300 Series 7MT. It is a 300Mhz Pentium II with
| a total of 3 memory slots. I have two 256 mb modules installed. When I
| install a 3rd 256 mb module the machine boots up but it is *extremely*
| slow. Booting up and shutting down the machine take forever. I believe
| that it took 15-20 minutes (possibly longer, but at least that long, I
| didn't time it exactly) to do a shutdown under Linux on the machine. The
| only explanation that I can think of is that the motherboard is not able
| to cache memory above a certain limit. Although I can't imagine this
| causing such a massive slowdown. All the memory modules are identical.
| And if I switch the modules to different slots the problem only occurs
| when all three modules are installed.

Intel made a rather bizarre design decision in the early versions
of the Pentium II CPU: the L2 cache can only map to the first
512 MB of main memory.

My understanding is that going over 512MB of memory on these systems
actually causes the processor to disable its L1 cache as well as its
L2 cache! That would explain the HUGE slowdown that the original
poster was seeing, a modern processor with neither L1 or L2 cache is
barely faster than an old 386!

Why Intel castrated these early PIIs is completely beyond me, it was
plain stupidity, as was pointed out by several people at the time.
Fortunately they fixed the problem with the next spin of the chip, so
even a late-model 300MHz PII will work fine, while all faster chips
never had the problem. However for those few who got the first batch
of PII (233 through to 300MHz) processors, 512MB is the max.

Unless you can find a really cheap 350-400 MHz Pentium II
somewhere (Ebay?) or absolutely must have 768 MB, you're
probably better off just staying at 512 MB.

Finding an old PII or PIII on Ebay shouldn't be too though. The
problem is that the motherboard used for this processor almost
certainly does not support 100MHz bus speeds. Most likely the
original poster will need either a 333MHz PII or a Celeron processor.
the latter is probably the best bet, Celerons of up to ~466MHz should
work fine on the board (though it may need a BIOS update). Slot 1
Celerons are a bit rare, bit they do exist. A quick search on Ebay
found a 400MHz slot 1 Celeron selling for $5 with only 3 hours to go
before the end of the auction. Even came with a heatsink (which are
different between the PII and Celeron chips).

Kind of scary that even the lowly Celeron never had this deficiency
that the first PII chips had!

 

[wg]

So it is only a CPU issue and not affected by the motherboard design?

My understanding is that going over 512MB of memory on these systems
actually causes the processor to disable its L1 cache as well as its
L2 cache! That would explain the HUGE slowdown that the original
poster was seeing, a modern processor with neither L1 or L2 cache is
barely faster than an old 386!

I hate it when companies intentionally cripple their products. PIIs are
actually still very useful as non-gui linux servers. Using a slower
machine is often ok as long as you have enough memory.

Why Intel castrated these early PIIs is completely beyond me, it was
plain stupidity, as was pointed out by several people at the time.

It wouldn't surprise me if they crippled them to create an incentive to
purchase the more expensive Xeons for server machines. Or maybe they are
just stupid.

Fortunately they fixed the problem with the next spin of the chip, so
even a late-model 300MHz PII will work fine, while all faster chips
never had the problem. However for those few who got the first batch
of PII (233 through to 300MHz) processors, 512MB is the max.




Finding an old PII or PIII on Ebay shouldn't be too though. The
problem is that the motherboard used for this processor almost
certainly does not support 100MHz bus speeds. Most likely the
original poster will need either a 333MHz PII or a Celeron processor.
the latter is probably the best bet, Celerons of up to ~466MHz should
work fine on the board (though it may need a BIOS update).

Yes I ran into this problem when I tried to upgrade a different 300 Mhz
PII machine to a 400Mhz PII. The bus was too slow for the upgrade. I did
not know that you could use a celeron with the PII motherboards. Thanks
for the info.

Kind of scary that even the lowly Celeron never had this deficiency
that the first PII chips had!

IIRC the Celerons were also better because their L2 caches were built in
to the chip. So they had better performance even though their L2 cache
was 1/4 the size of the PII.

Thank you both (ihoc and Tony) for your help. Unless I happen to come
across a really cheap CPU, I will probably just live with 512MB. :-( The
machine is really too old to be worth blowing a lot of $$$ to upgrade
and I really don't have to have 768MB since this is for use at home. It
just would have been nice.

 

[Nate Edel]

It wouldn't surprise me if they crippled them to create an incentive to
purchase the more expensive Xeons for server machines. Or maybe they are
just stupid.

Did they have Xeons yet? I don't remember Xeon or Slot 2 before oh, 400mhz
or so.

IIRC the Celerons were also better because their L2 caches were built in
to the chip. So they had better performance even though their L2 cache
was 1/4 the size of the PII.

Not to mention many of them were OCable like crazy.

 

[Tony Hill]

So it is only a CPU issue and not affected by the motherboard design?

The BIOS may play a role, but I think it was primarily a limitation of
the CPU.

I hate it when companies intentionally cripple their products. PIIs are
actually still very useful as non-gui linux servers. Using a slower
machine is often ok as long as you have enough memory.

I'm not sure that it was exactly an *intentionally* crippling of the
product, I think it was more of a ****-up. Guess they just didn't
think it was an issue as most systems of the time shipped with only 32
or 64MB of memory.

It wouldn't surprise me if they crippled them to create an incentive to
purchase the more expensive Xeons for server machines. Or maybe they are
just stupid.

This one really just looked like a case of stupidity. It certainly
wasn't an attempt to push the Xeon since these chips all predate the
Xeon line by a year or so. By the time the Xeons hit the market the
issue had been fixed.

Yes I ran into this problem when I tried to upgrade a different 300 Mhz
PII machine to a 400Mhz PII. The bus was too slow for the upgrade. I did
not know that you could use a celeron with the PII motherboards. Thanks
for the info.

Yup, Celerons up to somewhere around 733MHz used a 66MHz bus speed,
though the chances of getting one of the late-model Celerons to work
on that board are pretty slim, even if the bus speeds do match.
However up to somewhere around the 400-466MHz range should work ok (ie
those produced on the older 250nm fab process, faster Celerons were
produced on a 180nm or smaller process and mostly came in Socket 370
form anyway). You might want check up with your motherboard
manufacturer for specifics though.

IIRC the Celerons were also better because their L2 caches were built in
to the chip. So they had better performance even though their L2 cache
was 1/4 the size of the PII.

In rare occasions the Celerons did perform better because of the
faster cache, though usually the larger cache more than made up for
the difference. The PIIs, as noted above, also typically had faster
bus speeds (100MHz vs. the 66MHz of the Celeron).

Thank you both (ihoc and Tony) for your help. Unless I happen to come
across a really cheap CPU, I will probably just live with 512MB. :-( The
machine is really too old to be worth blowing a lot of $$$ to upgrade
and I really don't have to have 768MB since this is for use at home. It
just would have been nice.

As mentioned above, Slot 1 Celerons are selling for pretty damn cheap
on EBay, it might be worthwhile to check it out.

 

[Nate Edel]

I'm not sure that it was exactly an *intentionally* crippling of the
product, I think it was more of a ****-up. Guess they just didn't
think it was an issue as most systems of the time shipped with only 32
or 64MB of memory.

On the Pentium classics, the HX chipsets had caches which worked with up to
512mb; some of the other ones of the same era only worked with 64mb.

In rare occasions the Celerons did perform better because of the
faster cache, though usually the larger cache more than made up for
the difference. The PIIs, as noted above, also typically had faster
bus speeds (100MHz vs. the 66MHz of the Celeron).

The Celeron 300A and 333A pretty well matched the PII 300 and 333mhz.
Upgrading the FSB to the 350 made a much bigger difference than either the
cache speed or cache size.

 

[George Macdonald]

The BIOS may play a role, but I think it was primarily a limitation of
the CPU.


I'm not sure that it was exactly an *intentionally* crippling of the
product, I think it was more of a ****-up. Guess they just didn't
think it was an issue as most systems of the time shipped with only 32
or 64MB of memory.

If it was, it was a monumental ****-up. There were several cache
controllers which were limited back then of course, but to actually go to
the trouble of disabling caching, instead of just trapping addresses beyond
the watermark and bypassing the cache lookup seems off the wall.

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

 

[Tony Hill]

On the Pentium classics, the HX chipsets had caches which worked with up to
512mb; some of the other ones of the same era only worked with 64mb.

Yup, that was another poor choice IMO. That may have been a bit of
intentional crippling, particularly in the case of the Intel TX
chipset (released after the HX chipset). Intel has had a bit of a
history of underestimating memory needs, something that they have been
repeating recently with their insistence that 64-bit chips won't be
needed on the desktop until the end of the decade.

The Celeron 300A and 333A pretty well matched the PII 300 and 333mhz.
Upgrading the FSB to the 350 made a much bigger difference than either the
cache speed or cache size.

Agreed. Dollar for dollar though, the Celerons continued to offer
some pretty respectable performance in those days, unlike the Celerons
of today.

 

[Nate Edel]

Agreed. Dollar for dollar though, the Celerons continued to offer
some pretty respectable performance in those days, unlike the Celerons
of today.

The late Tualatin Celerons offered pretty respectable performance well into
the P4 era... especially sad was comparing a Tualatin C/1.4 to a Northwood
C/2.0 and realizing that we'd taken a step backwards.

Actually, that's not always true; some of the cheaper Northwood Celerons are
quite a deal for very specific workloads -- if you have tight, P4-optimized
code that lives entirely on a working set in cache, they can work pretty
nicely.

Outside of a few friends' number-crunching for class projects, I can't think
of anything like that in real life.

 

[KR Williams]

Yup, that was another poor choice IMO. That may have been a bit of
intentional crippling, particularly in the case of the Intel TX
chipset (released after the HX chipset). Intel has had a bit of a
history of underestimating memory needs, something that they have been
repeating recently with their insistence that 64-bit chips won't be
needed on the desktop until the end of the decade.

Remember, the TX came out right at the end of Intel's socket-7
market. They were about to switch to slot-1 so limiting memory
to 64MB was seen as more "Market Segmentation by Intel" [TM]. ;-)

To be completely fair though, to support more than 64MB the HX
had to have two external tag SRAMs. Most boards shipped with
only one and few had provisions for both, effectively limiting
them to 64MB also. The TX tags were internal to the processor so
allowing more than 64MB had costs associated with it. The TX had
other "issues" as well.

Agreed. Dollar for dollar though, the Celerons continued to offer
some pretty respectable performance in those days, unlike the Celerons
of today.

You mean "respectable performance" like the 300? ;-) (BTW, I
didn't think there was a 333A.)

 

[Nate Edel]

(e-mail address removed) says...

You mean "respectable performance" like the 300? ;-) (BTW, I
didn't think there was a 333A.)

I remembered there being a cacheless 333, but I've double-checked the spec
finder and if there was, Intel isn't admitting it, so you're correct there.

That said, I'm pretty sure that a fair number of the retailers I dealt with
at the time _called_ it a 333A, probably just to make clear "yes, this one
has cache."

 

[KR Williams]

I remembered there being a cacheless 333, but I've double-checked the spec
finder and if there was, Intel isn't admitting it, so you're correct there.

That said, I'm pretty sure that a fair number of the retailers I dealt with
at the time _called_ it a 333A, probably just to make clear "yes, this one
has cache."

Kinda like "Tiger Direct"? Yeah, I remember the same. Nothing
has changed. Pond-scum will be forever pond-scum.

On the absolute *opposite* end, I've found newegg to be great
(and my entrance hall is full of boxes from them .

 

[Rob Stow]

Kinda like "Tiger Direct"? Yeah, I remember the same. Nothing
has changed. Pond-scum will be forever pond-scum.

On the absolute *opposite* end, I've found newegg to be great
(and my entrance hall is full of boxes from them .

And did the parts all play together nicely when you put
together your new Opty dualie ?

 

[KR Williams]

And did the parts all play together nicely when you put
together your new Opty dualie ?

No dualie here (S2875S). I'm a simple working man and cannot
afford such luxuries. ...and no that's why all the boxes are
still in the entrance hall. I bought when I did so I'd have toyz
to play with on my week off (next week, but shipping was
amazingly fast).

I also have painting to do before I can play. The CFO speaketh!
;-)

 

[Rob Stow]

No dualie here (S2875S).

Ah, I thought you were getting the S2875.

 

[KR Williams]

Ah, I thought you were getting the S2875.

Well, I ran out of money. I didn't even buy a new disk drive
(perhaps next month). The CFO upped the budget twice already and
I didn't want to *PAY* more. As it was I had to buy flowers that
Sunday (and she's not even *my* mother! ;-).

In any case a single Opteron (144 is what I settled on) was all I
could swing if I wanted the second monitor (I did). The
processor and memory went up about $100 during the month I was
procrastinating. I had no time to play anyway (mandatory OT, and
all that).