Speed hit for replacing PC2700 with PC2100?

[CharlesBlackstone]

I want to give my friend a gig of PC2100 from my Dell latitude to put
in his Dell inspiron. His takes PC2700, mine is PC2100. How much will
it slow his machine down?

Thanks....

 

[Derek Simmons]

I want to give my friend a gig of PC2100 from my Dell latitude to put
in his Dell inspiron. His takes PC2700, mine is PC2100. How much will
it slow his machine down?

Thanks....

If you are related to Scott Blackstone I'm not going to answer your
question. When he sobers up you can go ask him.

But, I'll chance it that you aren't related to him. Based on the
little bit of information you have provided I'll hazard a guess.
PC2100 and PC2700 are theoretical maximum transfer rates. Doing the
math you might see 78% (100 x 2100 / 2700) speed difference. But if
your friends laptop only had 256 Mb vs 1 Gb that you are installing he
then now has four times as much memory and for memory intensive
applications it only needs to go out to disk ¼ of the time. Accessing
something that has been swapped out to disk takes milliseconds vs
physical RAM that takes nanoseconds, 500 - 1000 times faster.

There are also memory architectures that can take advantage of
installing two 512 Mb DIMMs vs one 1 Gb DIMM but manufactures aren't
always fourth coming with information like that because there is a
sales tactic where they will try to sell you a PC today with two 256 Mb
DIMMs and lead you to believe that you can upgrade it tomorrow to 1 Gb
with minimal cost. But when you go to upgrade it they tell you they
have to replace both SIMMs to upgrade it and they replace the two 256
Mb DIMMs with one 1 Gb DIMM. You get the computer home and it seems
like it is running slower. In essence it is because before it could
split memory accesses across two DIMMs now it is limited to one. You
take your computer back to the store and they explain this little
hidden feature and because of their return policy they can't take
back the one 1 Gb DIMM for two 512 Mb DIMMs but they can fix the
problem by installing a second 1 Gb DIMM. So to summarize the computer
you originally bought with 512 Mb of RAM that you were under the false
impression that you could upgrade to 1 Gb by adding another 512 Mb of
RAM you have been manipulated into buying two 1 Gb DIMMs.

Also keep in mind that additional memory can effect battery runtime.

So before I can give you my opinion of your proposed memory upgrade,
the 1 Gb you are going to install is it one 1 Gb DIMM or two 512 Mb
DIMM and how much memory are replacing and in what size DIMMs is it?

Derek

 

[CharlesBlackstone]

If you are related to Scott Blackstone I'm not going to answer your
question. When he sobers up you can go ask him.

But, I'll chance it that you aren't related to him. Based on the
little bit of information you have provided I'll hazard a guess.
PC2100 and PC2700 are theoretical maximum transfer rates. Doing the
math you might see 78% (100 x 2100 / 2700) speed difference. But if
your friends laptop only had 256 Mb vs 1 Gb that you are installing he
then now has four times as much memory and for memory intensive
applications it only needs to go out to disk ¼ of the time. Accessing
something that has been swapped out to disk takes milliseconds vs
physical RAM that takes nanoseconds, 500 - 1000 times faster.

There are also memory architectures that can take advantage of
installing two 512 Mb DIMMs vs one 1 Gb DIMM but manufactures aren't
always fourth coming with information like that because there is a
sales tactic where they will try to sell you a PC today with two 256 Mb
DIMMs and lead you to believe that you can upgrade it tomorrow to 1 Gb
with minimal cost. But when you go to upgrade it they tell you they
have to replace both SIMMs to upgrade it and they replace the two 256
Mb DIMMs with one 1 Gb DIMM. You get the computer home and it seems
like it is running slower. In essence it is because before it could
split memory accesses across two DIMMs now it is limited to one. You
take your computer back to the store and they explain this little
hidden feature and because of their return policy they can't take
back the one 1 Gb DIMM for two 512 Mb DIMMs but they can fix the
problem by installing a second 1 Gb DIMM. So to summarize the computer
you originally bought with 512 Mb of RAM that you were under the false
impression that you could upgrade to 1 Gb by adding another 512 Mb of
RAM you have been manipulated into buying two 1 Gb DIMMs.

Also keep in mind that additional memory can effect battery runtime.

So before I can give you my opinion of your proposed memory upgrade,
the 1 Gb you are going to install is it one 1 Gb DIMM or two 512 Mb
DIMM and how much memory are replacing and in what size DIMMs is it?

Derek

Hi Derek, thanks for the excellent answer.

Would be replacing two 256 dimms with two 512 dimms, so sounds like we
won't run into that problem.

The 78% is what I was looking for. I didn't know if it just depended on
comparing 2100 to 2700, or might involve some other parameter like bus
speeds, etc.

Thanks....

 

[Derek Simmons]

Hi Derek, thanks for the excellent answer.

Would be replacing two 256 dimms with two 512 dimms, so sounds like we
won't run into that problem.

The 78% is what I was looking for. I didn't know if it just depended on
comparing 2100 to 2700, or might involve some other parameter like bus
speeds, etc.

Thanks....

If your frined is using memory intensive applications, (ie. MS Visual
Stduio, 3D Studio MAX, Xilinx ISE or Altera QuartusII) Then I would
expect him to see a difference. If he is running smaller apps that
don't require a lot of swapping to disk then I would expect him not to
see a change or feel the ill effects of slower RAM too much.

 

[krw]

If your frined is using memory intensive applications, (ie. MS Visual
Stduio, 3D Studio MAX, Xilinx ISE or Altera QuartusII) Then I would
expect him to see a difference. If he is running smaller apps that
don't require a lot of swapping to disk then I would expect him not to
see a change or feel the ill effects of slower RAM too much.

You got that backwards. If there is any swapping at all, I
wouldn't expect to see any difference. No matter how slow the RAM
is it's still 1000x faster than disk. Also, the larger memory
favors anything that might have to swap.

Everything said, the system may not accept the slower RAM.

 

[CharlesBlackstone]

If you are related to Scott Blackstone I'm not going to answer your
question. When he sobers up you can go ask him.

But, I'll chance it that you aren't related to him. Based on the
little bit of information you have provided I'll hazard a guess.
PC2100 and PC2700 are theoretical maximum transfer rates. Doing the
math you might see 78% (100 x 2100 / 2700) speed difference. But if
your friends laptop only had 256 Mb vs 1 Gb that you are installing he
then now has four times as much memory and for memory intensive
applications it only needs to go out to disk ¼ of the time. Accessing
something that has been swapped out to disk takes milliseconds vs
physical RAM that takes nanoseconds, 500 - 1000 times faster.

There are also memory architectures that can take advantage of
installing two 512 Mb DIMMs vs one 1 Gb DIMM but manufactures aren't
always fourth coming with information like that because there is a
sales tactic where they will try to sell you a PC today with two 256 Mb
DIMMs and lead you to believe that you can upgrade it tomorrow to 1 Gb
with minimal cost. But when you go to upgrade it they tell you they
have to replace both SIMMs to upgrade it and they replace the two 256
Mb DIMMs with one 1 Gb DIMM. You get the computer home and it seems
like it is running slower. In essence it is because before it could
split memory accesses across two DIMMs now it is limited to one. You
take your computer back to the store and they explain this little
hidden feature and because of their return policy they can't take
back the one 1 Gb DIMM for two 512 Mb DIMMs but they can fix the
problem by installing a second 1 Gb DIMM. So to summarize the computer
you originally bought with 512 Mb of RAM that you were under the false
impression that you could upgrade to 1 Gb by adding another 512 Mb of
RAM you have been manipulated into buying two 1 Gb DIMMs.

Also keep in mind that additional memory can effect battery runtime.

So before I can give you my opinion of your proposed memory upgrade,
the 1 Gb you are going to install is it one 1 Gb DIMM or two 512 Mb
DIMM and how much memory are replacing and in what size DIMMs is it?

Derek

Thanks Derek.

Any sense of, for a given time, how much of the battery is getting
eaten by RAM? So, say, 512 Megs uses 5% of power at any given time
(just a hypothetical number, with CPU, screen lighting, hard disk,
graphics card, wireless, ?? taking up the rest). If I double RAM is it
then taking 10 %?

I guess to decide if this is a significant use of power, we'd have to
know what percent of power goes to RAM....

Jim

 

[David Kanter]

Thanks Derek.

Any sense of, for a given time, how much of the battery is getting
eaten by RAM? So, say, 512 Megs uses 5% of power at any given time
(just a hypothetical number, with CPU, screen lighting, hard disk,
graphics card, wireless, ?? taking up the rest). If I double RAM is it
then taking 10 %?

I guess to decide if this is a significant use of power, we'd have to
know what percent of power goes to RAM....

More memory will almost always improve performance and battery life.
Consider the following cases:

1. Workload uses < initial memory capacity
No change in performance, minor increase in power consumption.

2. Workload uses > initial capacity and < upgraded capacity
Big win - you no longer go to disk, which means the HD, the HD
controller and all associated data paths (to southbridge, etc.) are now
idle, and the system will put them to sleep, saving loads of power.

3. Workload uses > upgraded capacity
Still a win - you go to disk less frequently, so you get some of the
power savings from the memory.

Generally the speed of memory is a 2nd order effect. It's always
better to have enough memory to hold your workload, rather than fast
memory that doesn't hold the entire workload.

DK

 

[kony]

More memory will almost always improve performance and battery life.
Consider the following cases:

1. Workload uses < initial memory capacity
No change in performance, minor increase in power consumption.

2. Workload uses > initial capacity and < upgraded capacity
Big win - you no longer go to disk, ...

False, it's windows... It will write to disk no matter how
much memory you have because MS didn't care to design it any
better than they did, without even the ability to force the
OS not to write to disk for various reasons even if pagefile
is disabled. There are hacks to get around that like the
EWF filter for XPe, but too involving for the average person
to undertake.

...which means the HD, the HD
controller and all associated data paths (to southbridge, etc.) are now
idle, and the system will put them to sleep, saving loads of power.

It saves a tiny bit of power not having the HDD seek, but it
may still be spinning, due to OS requests even if user sets
a short spin-down period in power management. IOW, the
power consumption is basically a wash, swamped by the
consumption of anything else like the screen or CPU & video
(If not integrated on chipset). The "data path" is not put
to sleep at all, it is integral to the chipset and always
consumes power except that on the HDD itself.

3. Workload uses > upgraded capacity
Still a win - you go to disk less frequently, so you get some of the
power savings from the memory.

Generally the speed of memory is a 2nd order effect. It's always
better to have enough memory to hold your workload, rather than fast
memory that doesn't hold the entire workload.

Absolutely, and especially so in a laptop with it's slower
HDD (than even budget grade desktop drive).

 

[Stephen Sprunk]

Would be replacing two 256 dimms with two 512 dimms, so sounds like
we won't run into that problem.

Nope. Power consumption won't change too much either; it'll be slightly
higher from there being more transistors, but that's offset by the lower
frequency. Net consumption might go up _or_ down, but either way it
won't be by enough to notice since it's the same number of DIMMs.

The 78% is what I was looking for. I didn't know if it just depended
on
comparing 2100 to 2700, or might involve some other parameter like bus
speeds, etc.

2700 and 2100 are the theoretical maximum transfer rates, which are
determined from the bus speeds, etc. The difference might be noticeable
in some apps (e.g. FPS games), but a 22% hit isn't so bad overall -- and
if your friend noticed/cared about memory speed, he'd be running PC3200
or above already.

However, any reduction in performance due to slower transfers will be
overshadowed by the huge performance improvement due to doubling the
available memory. Even idling, my WinXP laptop runs about 400MB
allocated; a single app like Outlook or IE7 will push that to the 512MB
installed size, and two apps will start the disk thrashing. 1GB and my
laptop would spend most of its day with the HD spun down -- and probably
double the battery life while improving performance. Too bad the 512MB
upgrade costs as much as a new laptop, and there's no money in the
budget for either right now.

S

 

[kony]

Nope. Power consumption won't change too much either; it'll be slightly
higher from there being more transistors, but that's offset by the lower
frequency. Net consumption might go up _or_ down, but either way it
won't be by enough to notice since it's the same number of DIMMs.

# of DIMMs does not matter but # of chips does. Frequency
might matter but not much, it's only a slight difference.

However, any reduction in performance due to slower transfers will be
overshadowed by the huge performance improvement due to doubling the
available memory. Even idling, my WinXP laptop runs about 400MB
allocated;

Really? That's a HUGE amount of memory usage for an idling
XP system, closer to 150MB is typical.

a single app like Outlook or IE7 will push that to the 512MB
installed size, and two apps will start the disk thrashing.

I would scan for spyware/etc, merely running IE and OE
should be fine on a system with 256MB memory unless you're
talking about multiple instances of IE.

1GB and my
laptop would spend most of its day with the HD spun down

You're dreaming, Windows itself still writes to the HDD
constantly.

-- and probably
double the battery life while improving performance.

It's easily a performance increase, but not remotely close
to doubling battery life. Might gain 5%

Too bad the 512MB
upgrade costs as much as a new laptop, and there's no money in the
budget for either right now.

??

512MB not even close to 1/4 the cost of a laptop.

 

[Stephen Sprunk]

Really? That's a HUGE amount of memory usage for an idling
XP system, closer to 150MB is typical.

I just killed off everything I safely could and looked in TaskMan and
it's closer to 350MB; 400MB is with all the crap that loads on
startup -- my mistake.

I would scan for spyware/etc, merely running IE and OE
should be fine on a system with 256MB memory unless you're
talking about multiple instances of IE.

No spyware other than the corporate stuff. There's a lot of background
tasks that eat up a lot of memory and that can't be killed or some apps
just won't work -- or IT will start yelling at me for trying to "hide"
my activities.

??

512MB not even close to 1/4 the cost of a laptop.

It is when you figure that a 512MB upgrade has to be expensed this
quarter, but a new laptop is amortized (and thus hits my budget) over 12
quarters. Ah, the wonders of corporate finance...

S

 

[CharlesBlackstone]

More memory will almost always improve performance and battery life.
Consider the following cases:

1. Workload uses < initial memory capacity
No change in performance, minor increase in power consumption.

2. Workload uses > initial capacity and < upgraded capacity
Big win - you no longer go to disk, which means the HD, the HD
controller and all associated data paths (to southbridge, etc.) are now
idle, and the system will put them to sleep, saving loads of power.

3. Workload uses > upgraded capacity
Still a win - you go to disk less frequently, so you get some of the
power savings from the memory.

Generally the speed of memory is a 2nd order effect. It's always
better to have enough memory to hold your workload, rather than fast
memory that doesn't hold the entire workload.

DK

Thanks much for the great answer.

 

[CharlesBlackstone]

False, it's windows... It will write to disk no matter how
much memory you have because MS didn't care to design it any
better than they did, without even the ability to force the
OS not to write to disk for various reasons even if pagefile
is disabled. There are hacks to get around that like the
EWF filter for XPe, but too involving for the average person
to undertake.



It saves a tiny bit of power not having the HDD seek, but it
may still be spinning, due to OS requests even if user sets
a short spin-down period in power management. IOW, the
power consumption is basically a wash, swamped by the
consumption of anything else like the screen or CPU & video
(If not integrated on chipset). The "data path" is not put
to sleep at all, it is integral to the chipset and always
consumes power except that on the HDD itself.





Absolutely, and especially so in a laptop with it's slower
HDD (than even budget grade desktop drive).

Wow, really? You mean Bill designed an operating system that writes to
disk even when physical ram is not full? I know that's what you just
said, but it flabbergasts me so much I had to make sure. Why in the
world would they do this, forever limiting your ability to build up a
computationally fast machine?

 

[David Kanter]

On 31 Dec 2006 13:56:40 -0800, "David Kanter"
[snip]

2. Workload uses > initial capacity and < upgraded capacity
Big win - you no longer go to disk, ...

False, it's windows... It will write to disk no matter how
much memory you have because MS didn't care to design it any
better than they did, without even the ability to force the
OS not to write to disk for various reasons even if pagefile
is disabled. There are hacks to get around that like the
EWF filter for XPe, but too involving for the average person
to undertake.

Notice how I said nothing about writing to disk, I said 'no longer go
to disk'. Most applications have 2x (or more) the number of reads as
writes...

Any sane operating system has to write dirty pages to disk from time to
time. If the system were to experience a non-recoverable fault and
reboot, then any dirty pages will be lost as memory is cycled. Keeping
a large number of dirty pages in memory is really pretty silly,
especially since the cost of a combined write is relatively small.
Moreover, even if you had a UPS and could keep dirty pages around
indefinitely, you really are much better off with consistent write
backs. If you keep too many dirty pages around, you could lock the
system up when you try to write them all back; this situation is
analogous to garbage collection. The best user experience is when the
GC occurs every X time periods in the background, rather than pausing
all execution to do GC. No interruptions or usability degradation that
way.

It saves a tiny bit of power not having the HDD seek, but it
may still be spinning, due to OS requests even if user sets
a short spin-down period in power management.

An active hard drive consumes 2-3W, the infrequent disk activity due to
windows uses 0.2W, that's a factor of 10-15x difference. See this
paper for the PCMark disk benchmarks:
http://www.crhc.uiuc.edu/~mahesri/classes/project_report_cs497yyz.pdf

IOW, the
power consumption is basically a wash, swamped by the
consumption of anything else like the screen or CPU & video
(If not integrated on chipset). The "data path" is not put
to sleep at all, it is integral to the chipset and always
consumes power except that on the HDD itself.

No. Any modern laptop can drop the bus between the northbridge and
memory, the northbridge and south bridge, or the south bridge and
memory into a sleep state where almost no power is used. It's really
not that hard, and for Intel chipsets it's standard practice.

Absolutely, and especially so in a laptop with it's slower
HDD (than even budget grade desktop drive).

The bottom line is that more memory is always good for performance and
power efficiency. You should add more memory to the system and not
sweat about losing bandwidth.

DK

 

[David Kanter]

Thanks much for the great answer.

I'm glad you found it enlightening and useful.

Regards,


DK

 

[David Kanter]

Wow, really? You mean Bill designed an operating system that writes to

disk even when physical ram is not full? I know that's what you just
said, but it flabbergasts me so much I had to make sure. Why in the
world would they do this, forever limiting your ability to build up a
computationally fast machine?

One reason is that it's risky to keep a lot of dirty pages in memory.
The contents of memory are lost when the system is reset, so if you
keep say, 100MB of dirty memory, then you are going to be in rather
serious trouble were you to suddenly lose power. While a laptop should
not suddenly lose power, it's quite possible to experience a
non-recoverable failure that forces a restart.

Also, note that this "problem" is not isolated to a particular OS, but
is rather a result of using RAM memory with write back to permanent
storage in conjunction with virtual memory.

DK

 

[Bill Todd]

CharlesBlackstone wrote:

....

You mean Bill designed an operating system that writes to

disk even when physical ram is not full? I know that's what you just
said, but it flabbergasts me so much I had to make sure. Why in the
world would they do this,

David has already answered that question. The coming use of flash
memory as a disk-assist should, however, keep the disk from having to
spin up nearly as often when only light write activity is occurring.

forever limiting your ability to build up a

computationally fast machine?

I'm pretty sure that Windows prioritizes demand-reads and -writes over
lazy write-backs (i.e., writes back when the disk is not otherwise busy,
at least unless the amount of cached dirty data starts to become
dangerously large), so no significant performance is compromised in the
pursuit of data persistence.

- bill

 

[krw]

I just killed off everything I safely could and looked in TaskMan and
it's closer to 350MB; 400MB is with all the crap that loads on
startup -- my mistake.

Win is a mess, that's for sure.

No spyware other than the corporate stuff. There's a lot of background
tasks that eat up a lot of memory and that can't be killed or some apps
just won't work -- or IT will start yelling at me for trying to "hide"
my activities.

I always plead not-guilty by reason of Windows. "Who am I to know
what all that shite does?"

It is when you figure that a 512MB upgrade has to be expensed this
quarter, but a new laptop is amortized (and thus hits my budget) over 12
quarters. Ah, the wonders of corporate finance...

Hmm, we could usually expense stuff, but capital $$ were like
pulling teeth. Depending on what it is, a 512MB stick should be
around $50-$60. With the limited upgrade possibility of a laptop,
I'd go for a 1GB stick (still $100-$130ish). If it's that much
trouble for your counters of beans, pull the green out of your
pocket (I've done that on more than one occasion).

 

[kony]

Any sane operating system has to write dirty pages to disk from time to
time.

Unfounded presumption.

If the system were to experience a non-recoverable fault and
reboot, then any dirty pages will be lost as memory is cycled.

.... Which we don't care about

Keeping
a large number of dirty pages in memory is really pretty silly,
especially since the cost of a combined write is relatively small.

Even worse is perpetual writies to a mechanical storage
device. "Cost" is cumulative, as well as consuming power,
putting wear on a drive.

Moreover, even if you had a UPS and could keep dirty pages around
indefinitely, you really are much better off with consistent write
backs.

UPS? You seem to be implying it's the AC power, not windows
or an app. fault, which isn't as likely.

You've also ignored the other reasons for disk writes,
presuming it's only a pagefile issue. The issue is that
windows writes to the HDD at all, ever. Virtual memory
implementation is but one problem.

If you keep too many dirty pages around, you could lock the
system up when you try to write them all back; this situation is
analogous to garbage collection.

So the whole problem is the HDD writes in the first place.

The best user experience is when the
GC occurs every X time periods in the background, rather than pausing
all execution to do GC. No interruptions or usability degradation that
way.

"User experience" is best when an OS doesn't build more and
more bloat on top of itself until there's no way to control
what happens. The pagefile logic tries to be a
one-size-fits-all, presuming a system won't have enough
memory to run the jobs, when these days it usually does
because most is allocated but never actually used.

An active hard drive consumes 2-3W, the infrequent disk activity due to
windows uses 0.2W, that's a factor of 10-15x difference.

Windows has FREQUENT disk activity, and it either keeps the
drive spinning or constantly spins it up and down again
(futher degradation in performance). What it does is
constantly cause the drive to not go into any deep state of
power management, so there's no 10-15X at all, Windows'
design itself is the CAUSE, not a savings of power.

See this
paper for the PCMark disk benchmarks:
http://www.crhc.uiuc.edu/~mahesri/classes/project_report_cs497yyz.pdf

So it describes what I had already, that windows writing
uses more power that if it didn't, and keeps the drive
spinning which uses even more (albeit a minor amount) power.

No. Any modern laptop can drop the bus between the northbridge and
memory, the northbridge and south bridge, or the south bridge and
memory into a sleep state where almost no power is used. It's really
not that hard, and for Intel chipsets it's standard practice.

You mean when the system isnt being used, which is not what
we were talking about... or at least not what most people
would be talking about, it is obvious enough that a laptop
uses less power when in a higher power management mode.

The bottom line is that more memory is always good for performance and
power efficiency. You should add more memory to the system and not
sweat about losing bandwidth.

Often more memory improves performance but if we are looking
at a goal of improving performance then more memory may not
be the answer. Blindly adding it may raise latency, and
overall it would be a bit too quick to assume that in
today's PC or laptop, the typical 1GB isn't enough for the
most common uses (web, office, email, etc). Certainly it's
easy to exceed 1GB *need*, but this is assessed on a
use/user by user basis, blindly beefing up a system is going
to have a practical diminishing return, and that often
before the system has max memory possible, installed. If
we could stop windows from wastefully writing to HDD, then
it would be better to have the extra memory but
unfortunately Windows doesn't give us that choice, it's
written to run on low spec systems at the expense of better
spec systems.

 

[chrisv]

(e-mail address removed) says...

Hmm, we could usually expense stuff, but capital $$ were like
pulling teeth. Depending on what it is, a 512MB stick should be
around $50-$60. With the limited upgrade possibility of a laptop,
I'd go for a 1GB stick (still $100-$130ish). If it's that much
trouble for your counters of beans, pull the green out of your
pocket (I've done that on more than one occasion).

Ha. I've got a corporate credit card and and an account at NewEgg. I
just order what I need and don't even tell the boss unless it's a big
(>$1k) item...