RAM to AGP mem share: Bad?

[PaulFXH]

Hi everybody
My computer has a PCChips M925LR mobo with VIA VT8751 Apollo P4M266
(Nth)/VIA VT8233C (Sth) chipset and an Intel Pentium 4 CPU running at
1.8GHZ and 100MHz frequency.
Two RAM sticks (128MB PC133 SDRAM but running at 100MHz) are in place
and the mobo has an integrated S3 ProSavageDDR video card which shares
memory with the RAM. It also has an external AGP slot which is
currently empty.

The computer runs well on WinME with startup and shutdown being
impressively rapid and apps, in general, appear without delays.

However, when I recently ran some benchmark checks on the memory
(Everest), the RAM read/write speeds as well as the CAS latency turned
out to be considerably worse than expected for similarly configured
machines.
Subsequent checks using SiSoft Sandra suggested that shared memory (RAM
to AGP) can "seriously reduce system performance" and recommends
installing an external AGP card.
For this reason, I reduced (in BIOS) the amount of RAM memory shared
with the onboard AGP from 32MB to 8MB. Surprisingly, this made no
difference to the RAM speeds measured by Everest. Neither was the video
quality adversely affected although both Everest and Sandra reported
the available RAM as having truly increased from 224MB to 248MB.

The following very interesting article from Tom's Hardware Guide

http://www.tomshardware.com/2003/09/03/integrated_graphics_performance/index.html

suggests that integrated AGP cards sharing memory with RAM can very
significantly reduce video quality and gaming speeds but is unlikely to
reduce the performance of machines running office tasks.
I'm in the latter category and therefore should not be adversely
affected. Then why is Sandra saying the opposite and how do I explain
the apparently poor results in the benchmark tests?

My apologies for the long-winded lead-up but I'd would appreciate
hearing from anybody who can clarify this subject for me, particularly
anybody who has direct experience plugging in an external AGP and
switching off the integrated card.

TIA
Paul

 

[kony]

Hi everybody
My computer has a PCChips M925LR mobo with VIA VT8751 Apollo P4M266
(Nth)/VIA VT8233C (Sth) chipset and an Intel Pentium 4 CPU running at
1.8GHZ and 100MHz frequency.
Two RAM sticks (128MB PC133 SDRAM but running at 100MHz) are in place
and the mobo has an integrated S3 ProSavageDDR video card which shares
memory with the RAM. It also has an external AGP slot which is
currently empty.

The computer runs well on WinME with startup and shutdown being
impressively rapid and apps, in general, appear without delays.

However, when I recently ran some benchmark checks on the memory
(Everest), the RAM read/write speeds as well as the CAS latency turned
out to be considerably worse than expected for similarly configured
machines.

Integrated video does that, but also Everest might be
considering best-case scenarios which your memory or board
bios won't do. PC100 (speed) memory just doesn't have a
lot of bandwidth to start with then taking away some for the
video makes it worse. If on the other hand you had a
dual-channel DDR400 (PC3200) memory based system, there
wouldn't be any perceived diference in performance for most
uses, mainly gaming would be impacted by the integrated
video.

Subsequent checks using SiSoft Sandra suggested that shared memory (RAM
to AGP) can "seriously reduce system performance" and recommends
installing an external AGP card.

Shared-memory integrated video does reduce performance in
some tasks, but then the tradeoff was that it's cheaper. As
always, if you want best performance the cost goes up and up
and up and ...

For this reason, I reduced (in BIOS) the amount of RAM memory shared
with the onboard AGP from 32MB to 8MB. Surprisingly, this made no
difference to the RAM speeds measured by Everest.

It wasn't expected to, the issue is the bandwidth being
used, not the amount. The amount is more relevant to what
resolutions and bit depths are supported on your monitor,
and to a very limited extent (since it's relatively slow
video) how well it can play older 3D games.

Neither was the video
quality adversely affected although both Everest and Sandra reported
the available RAM as having truly increased from 224MB to 248MB.

What operating system are you using? If WinXP or Win2k with
a lot of multitasking, the best performance increase for
typical (internet, email, office, etc) uses would be to add
some memory, increase to at least 512MB total if not more.
However, the system is aging now and at some point you'd
have to decide if pouring more money into it is appropriate
instead of replacing whole thing.

The following very interesting article from Tom's Hardware Guide

http://www.tomshardware.com/2003/09/03/integrated_graphics_performance/index.html

suggests that integrated AGP cards sharing memory with RAM can very
significantly reduce video quality and gaming speeds but is unlikely to
reduce the performance of machines running office tasks.
I'm in the latter category and therefore should not be adversely
affected. Then why is Sandra saying the opposite and how do I explain
the apparently poor results in the benchmark tests?

Sandra runs a synthetic benchmark. It's not applicable to
all real-world uses because some things don't stress memory
as much as CPU or hard drive, other parts.

My apologies for the long-winded lead-up but I'd would appreciate
hearing from anybody who can clarify this subject for me, particularly
anybody who has direct experience plugging in an external AGP and
switching off the integrated card.

Before you ran the benchmarks for the memory did you feel
the system was too slow? If so, did it seem slower than it
previuosly did? If so, have you defragged your hard drive
and scanned for spyware, virus, etc?

Your motherbaord bios might be running the memory at 133MHz,
an asynchronous mode to the front side bus. "CPU-Z" (google
will find it) will show the memory's SPD rating and the
speed it's actually running at currently. If the running
speed is below it's rated (SPD) speed, you might be able to
change some bios settings to speed it up some... but be
careful, it is very easy to cause instability doing this and
at a minimum you should run memtest86+ for several hours to
check stabiltiy if you do change the bios, _before_ ever
booting windows to avoid possible file corruption.

 

[PaulFXH]

kony escreveu:

On 16 Apr 2006 07:50:00 -0700, "PaulFXH"

Hi Kony,
Thanks a lot for your reply.
My comments on your suggestions are interspersed below.

Integrated video does that, but also Everest might be
considering best-case scenarios which your memory or board
bios won't do. PC100 (speed) memory just doesn't have a
lot of bandwidth to start with then taking away some for the
video makes it worse. If on the other hand you had a
dual-channel DDR400 (PC3200) memory based system, there
wouldn't be any perceived diference in performance for most
uses, mainly gaming would be impacted by the integrated
video.

My intention was, and still is, to see what cheap (or preferably zero
cost) options are available for optimizing the machine's performance.
Therefore, I do not intend to upgrade the RAM sticks for the moment.

Shared-memory integrated video does reduce performance in
some tasks, but then the tradeoff was that it's cheaper. As
always, if you want best performance the cost goes up and up
and up and ...

I had thought of installing a second-hand (and therefore very cheap)
32MB (or 64MB) AGP 4x card and switching off the integrated chip.
However, I was using this thread to see if the expected improvement was
likely to be significant. The impression I'm getting from your comments
is that this is improbable particularly as I'm not a gamer.

It wasn't expected to, the issue is the bandwidth being
used, not the amount. The amount is more relevant to what
resolutions and bit depths are supported on your monitor,
and to a very limited extent (since it's relatively slow
video) how well it can play older 3D games.



What operating system are you using? If WinXP or Win2k with
a lot of multitasking, the best performance increase for
typical (internet, email, office, etc) uses would be to add
some memory, increase to at least 512MB total if not more.
However, the system is aging now and at some point you'd
have to decide if pouring more money into it is appropriate
instead of replacing whole thing.

The OS is WinME which I understand needs no more than 256MB of RAM.
Indeed, I am told it can operate well with as little as 64MB.

Sandra runs a synthetic benchmark. It's not applicable to
all real-world uses because some things don't stress memory
as much as CPU or hard drive, other parts.




Before you ran the benchmarks for the memory did you feel
the system was too slow? If so, did it seem slower than it
previuosly did? If so, have you defragged your hard drive
and scanned for spyware, virus, etc?

I have only been using this computer for about 5 weeks while
temporarily away from home. My home computer runs on WinXP and has
512MB RAM and 2.53GHz CPU with an external AGP. However, I was very
surprised to find that the computer I'm referring to in this thread is
by no means slow in comparison. Indeed, it does many things faster than
the machine I have at home.
So my attempts to speed things up are not because of intolerably poor
performance but rather a desire to "leave no stone unturned" in my
quest for optimum performance.
This computer is defragged at least on a weekly basis (Diskeeper Lite)
and is scanned for viruses and other malware at least once per day with
a range of software including AVG, BitDefender and Ad-Aware.

Your motherbaord bios might be running the memory at 133MHz,
an asynchronous mode to the front side bus. "CPU-Z" (google
will find it) will show the memory's SPD rating and the
speed it's actually running at currently. If the running
speed is below it's rated (SPD) speed, you might be able to
change some bios settings to speed it up some... but be
careful, it is very easy to cause instability doing this and
at a minimum you should run memtest86+ for several hours to
check stabiltiy if you do change the bios, _before_ ever
booting windows to avoid possible file corruption.

CPU-Z confirms that the RAM SPD max bandwidth is 133MHz while it runs
at 100MHz.
The mobo does have a jumper which allows the CPU frequency to be
changed from 100MHz to 133MHz.
The mobo manual, in its BIOS section, indicates that "automatic
adjustments" would be made to the CPU settings after the board had
detected the "type of installed CPU".
I took it from this that, were I to increase the CPU frequency from
100MHz to 133MHz, the CPU ratio selection would automatically be
reduced to avoid overclocking of the CPU but allowing, at the same time
the RAM to run at the higher speed.
However, after I changed the jumpers, the machine simply refused to
boot. Happily, it rebooted without problems when I readjusted the
jumpers.
My explanation for this is that the CPU just won't handle the higher
frequency.

In summary, can I conclude that you believe it unlikely that any
significant performance gain will result from the installation of an
external AGP card?

TIA
Paul

 

[CBFalconer]

.... snip ...
This computer is defragged at least on a weekly basis (Diskeeper
Lite) and is scanned for viruses and other malware at least once
per day with a range of software including AVG, BitDefender and
Ad-Aware.

If you don't have ECC memory frequent defragging can be very
harmful. Each file move passes through a memory buffer, and there
is NO check for dropped or altered bits in that buffer. So you can
be gradually destroying your files with no warning. The actual
destruction can be due to such things as cosmic rays, over which
you have no control whatsoever. The damage will be in all your
backups also.

Always insist on ECC memory.

--
"If you want to post a followup via groups.google.com, don't use
the broken "Reply" link at the bottom of the article. Click on
"show options" at the top of the article, then click on the
"Reply" at the bottom of the article headers." - Keith Thompson
More details at: <http://cfaj.freeshell.org/google/>
Also see <http://www.safalra.com/special/googlegroupsreply/>

 

[Paul]

Hi everybody
My computer has a PCChips M925LR mobo with VIA VT8751 Apollo P4M266
(Nth)/VIA VT8233C (Sth) chipset and an Intel Pentium 4 CPU running at
1.8GHZ and 100MHz frequency.
Two RAM sticks (128MB PC133 SDRAM but running at 100MHz) are in place
and the mobo has an integrated S3 ProSavageDDR video card which shares
memory with the RAM. It also has an external AGP slot which is
currently empty.

The computer runs well on WinME with startup and shutdown being
impressively rapid and apps, in general, appear without delays.

However, when I recently ran some benchmark checks on the memory
(Everest), the RAM read/write speeds as well as the CAS latency turned
out to be considerably worse than expected for similarly configured
machines.
Subsequent checks using SiSoft Sandra suggested that shared memory (RAM
to AGP) can "seriously reduce system performance" and recommends
installing an external AGP card.
For this reason, I reduced (in BIOS) the amount of RAM memory shared
with the onboard AGP from 32MB to 8MB. Surprisingly, this made no
difference to the RAM speeds measured by Everest. Neither was the video
quality adversely affected although both Everest and Sandra reported
the available RAM as having truly increased from 224MB to 248MB.

The following very interesting article from Tom's Hardware Guide

http://www.tomshardware.com/2003/09/03/integrated_graphics_performance/index.html

suggests that integrated AGP cards sharing memory with RAM can very
significantly reduce video quality and gaming speeds but is unlikely to
reduce the performance of machines running office tasks.
I'm in the latter category and therefore should not be adversely
affected. Then why is Sandra saying the opposite and how do I explain
the apparently poor results in the benchmark tests?

My apologies for the long-winded lead-up but I'd would appreciate
hearing from anybody who can clarify this subject for me, particularly
anybody who has direct experience plugging in an external AGP and
switching off the integrated card.

TIA
Paul

But the office productivity benchmarks here, show little difference
between plugging in a Radeon 9200 video card, and using the
integrated video, in office applications.

http://www.tomshardware.com/2003/09/03/integrated_graphics_performance/page19.html

A reason for moving to an AGP video card, would be to increase the
display performance. For most people, this would be for gaming. In
office applications, maybe you could scroll faster with the AGP
card, but it might be a tie between the two graphics systems.

In terms of the chipset itself, it is recent enough to have a
266MB/sec hub bus between Northbridge and Southbridge. Previous
generations of chipsets used a PCI bus starting at the Northbridge,
to connect to the Southbridge and the PCI slots on the motherboard.
That sharing accounted for some of the slowness of the previous
chipsets. So your chipset is "modern enough" in that sense.

http://www.via.com.tw/en/products/chipsets/p4-series/p4m266

To improve performance, from most improvement to least improvement,
the order would be:

1) Increase the CPU core clock. On many processors which have
locked multipliers, increasing the FSB is the only method
available to do this. On a certain percentage of AMD Athlon
processors, the multiplier may also be adjusted. To enhance
the CPU core clock, sometimes the Vcore to the processor must
be boosted, and not all motherboards have an adjustment for
this. "Wire mods" or switches have been used, to adjust the
Vcore voltage on motherboards without a BIOS adjustment. The
way I would do this, is disconnect the VID pins on the Vcore
regulator chip, and control them with switches. Then the
motherboard has no say in the matter

2) Increasing the memory clock can sometimes help. But there is
a caveat. If the CPU FSB clock and the memory clock are the
same, the Northbridge designers remove resync flip-flops,
which gives reduced latency whenever the clocks to the
two subsystems are the same. When you run the CPU clock at
100MHz, and the memory at 133MHz, the resync flip-flops (a
longer logic path), subtracts from the performance. On some
chipsets, that means a significant increase in memory clock
is needed, to pay for the added cost of resync. So, increasing
memory speed pays, if you raise it high enough (without errors).
Some chipsets just don't have the headroom for the necessary
amount of memory clock speed increase.

3) Reducing memory CAS will give a slight improvement. Moving
from CAS3 memory to CAS2 memory would help a bit. But this is
not a cost effective upgrade, unless you can sell the original
memory for what you paid for it.

4) Changing the motherboard might help a little bit. Some chipsets
have better memory controllers than others. The biggest payoff
would be moving your P4 to a dual channel memory controller,
but as far as I know, that architecture change happened when
DDR memory was more popular, so you'd need new RAM.

So increasing the CPU core clock is the answer, and that is made
easier if the BIOS has a jumperfree clock adjustment. This program
can adjust some clock generators, while you are in Windows, but
a very limited number of clock generators are supported. There are
other programs available like this, where a specific version of
program is needed for each model of clock generator. You must be
"born lucky" to find the necessary clock adjustment program
And the amount of overclock is limited by not being able to
increase Vcore, if that becomes necessary.

http://www.cpuid.com/clockgen.php

HTH,
Paul

 

[PaulFXH]

CBFalconer escreveu:

If you don't have ECC memory frequent defragging can be very
harmful. Each file move passes through a memory buffer, and there
is NO check for dropped or altered bits in that buffer. So you can
be gradually destroying your files with no warning. The actual
destruction can be due to such things as cosmic rays, over which
you have no control whatsoever. The damage will be in all your
backups also.

Always insist on ECC memory.

CBF
Thanks for pointing this out to me. I had never heard of this before.
Nevertheless, in three years of frequent defragging on two computers I
have never had the Windows Memory Diagnostic indicate any problems
whatsoever with any of my RAM sticks none of which have ECC.
Just lucky?
Paul

 

[kony]

CBF
Thanks for pointing this out to me. I had never heard of this before.
Nevertheless, in three years of frequent defragging on two computers I
have never had the Windows Memory Diagnostic indicate any problems
whatsoever with any of my RAM sticks none of which have ECC.
Just lucky?
Paul

Don't even bother with the Windows memory diagnostic, a
proper memory tester must run without the OS because the OS
occupies a significant, even substantial amount of memory on
your system. Memtest86+ is preferred.

 

[kony]

I had thought of installing a second-hand (and therefore very cheap)
32MB (or 64MB) AGP 4x card and switching off the integrated chip.
However, I was using this thread to see if the expected improvement was
likely to be significant. The impression I'm getting from your comments
is that this is improbable particularly as I'm not a gamer.

It would be a minor improvement, worth doing if you had the
spare video card just lying around unused. It wouldn't be
worth buying one though, unless you could get it for
peanuts. Similarly, some memory can be had cheaply and for
your purposes the memory could be as useful or even moreso
if you leave the system on for awhile between reboots, as
this results in more files from the HDD being cached into
main memory for faster access... IF you had enough memory to
do so, otherwise the cache gets flushed as much as is
necessary to run the applications.

The OS is WinME which I understand needs no more than 256MB of RAM.
Indeed, I am told it can operate well with as little as 64MB.

It can, but just like any other OS your particular uses
might exceed this amount of memory and then it would be
useful to have more to vastly reduce access to the swapfile.
Further the extra memory beyond that needed for the
applications is used as a file cache (as mentioned above)
which is much faster than rereading files from the hard
drive each time.

The issue isn't what the bare minimum is that'll work, since
you aren't looking to barely be able to run the system but
rather, the performance benefits of certain upgrades. Take
for example those who have more than 256MB running WinXP...
WinXP itself will run from 128MB, a lot better from 256MB,
but many use their systems in ways that make 512MB, 1GB or
more useful. WinME similarly benefits from more memory, but
due to it's underlying Win9x architecture, it's realistic
limit on the amount of memory it can use effectively is
about 1GB, maybe 1.5GB depending on tweaks to various
settings.

I have only been using this computer for about 5 weeks while
temporarily away from home. My home computer runs on WinXP and has
512MB RAM and 2.53GHz CPU with an external AGP. However, I was very
surprised to find that the computer I'm referring to in this thread is
by no means slow in comparison. Indeed, it does many things faster than
the machine I have at home.

Is the one at home an OEM setup? That, and misc. things
that accumulate over time can slow a system down some.
Plus, regardless of what some MS shills claim, Win9x IS
faster than XP on any and all systems when it comes to doing
basic tasks. More demanding uses will make them more even,
when the OS becomes a less significant amount of the
processing or % of memory utilization, it becomes the less
significant factor in performance too.

So my attempts to speed things up are not because of intolerably poor
performance but rather a desire to "leave no stone unturned" in my
quest for optimum performance.

Then you want a modern hard drive, to increase the memory,
and add the video card. Not necessarily in this order but
the former 2 will tend to make more of a difference for your
described uses than the latter will.

CPU-Z confirms that the RAM SPD max bandwidth is 133MHz while it runs
at 100MHz.

What about the memory timings? Those can account for a few
(single-digit) % performance difference as well...
especially when something is continually dependant on memory
throughput as with integrated video.

The mobo does have a jumper which allows the CPU frequency to be
changed from 100MHz to 133MHz.
The mobo manual, in its BIOS section, indicates that "automatic
adjustments" would be made to the CPU settings after the board had
detected the "type of installed CPU".
I took it from this that, were I to increase the CPU frequency from
100MHz to 133MHz, the CPU ratio selection would automatically be
reduced to avoid overclocking of the CPU but allowing, at the same time
the RAM to run at the higher speed.

No, the CPU multiplier (ratio selection) is locked,
unchangable on Intel CPUs. You do not want to, should not
change the front side bus. You want the same FSB but if
possible, an asynchronous (+33MHz higher) memory bus. Some
boards may not provide a means for the user to change this
setting. Other boards may hide the setting in the bios menu
until you change another setting from "auto" (or another
wording) to a manual mode at which point you pick either a
frequency or a ratio for the memory like 4:3 (or 3:4 if it
were FSB:Memory).

However, after I changed the jumpers, the machine simply refused to
boot. Happily, it rebooted without problems when I readjusted the
jumpers.
My explanation for this is that the CPU just won't handle the higher
frequency.

Yes, you'd be overclocking the CPU by 33%, which it "might"
do successfully if you rasied the CPU voltage but you
expressed no desire to do this and it may not be stable even
with higher voltage or might need more cooling than possible
with the stock heatsink.

In summary, can I conclude that you believe it unlikely that any
significant performance gain will result from the installation of an
external AGP card?

Yes, unless there is some other use you haven't mentioned
that places a greater demand on either the video subsystem
or the memory throughput.

 

[CBFalconer]

CBFalconer escreveu:

Thanks for pointing this out to me. I had never heard of this before.
Nevertheless, in three years of frequent defragging on two computers I
have never had the Windows Memory Diagnostic indicate any problems
whatsoever with any of my RAM sticks none of which have ECC.
Just lucky?

Who knows? If you run MD5sum checks on ALL your files before and
after any defrag, you can be fairly certain nothing has bitten.
The troubles can easily take years after their occurance to show
up. I wouldn't expect any memory diagnostic to indicate any
problem.

--
"If you want to post a followup via groups.google.com, don't use
the broken "Reply" link at the bottom of the article. Click on
"show options" at the top of the article, then click on the
"Reply" at the bottom of the article headers." - Keith Thompson
More details at: <http://cfaj.freeshell.org/google/>
Also see <http://www.safalra.com/special/googlegroupsreply/>

 

[PaulFXH]

kony escreveu:

It would be a minor improvement, worth doing if you had the
spare video card just lying around unused. It wouldn't be
worth buying one though, unless you could get it for
peanuts. Similarly, some memory can be had cheaply and for
your purposes the memory could be as useful or even moreso
if you leave the system on for awhile between reboots, as
this results in more files from the HDD being cached into
main memory for faster access... IF you had enough memory to
do so, otherwise the cache gets flushed as much as is
necessary to run the applications.





It can, but just like any other OS your particular uses
might exceed this amount of memory and then it would be
useful to have more to vastly reduce access to the swapfile.
Further the extra memory beyond that needed for the
applications is used as a file cache (as mentioned above)
which is much faster than rereading files from the hard
drive each time.

The issue isn't what the bare minimum is that'll work, since
you aren't looking to barely be able to run the system but
rather, the performance benefits of certain upgrades. Take
for example those who have more than 256MB running WinXP...
WinXP itself will run from 128MB, a lot better from 256MB,
but many use their systems in ways that make 512MB, 1GB or
more useful. WinME similarly benefits from more memory, but
due to it's underlying Win9x architecture, it's realistic
limit on the amount of memory it can use effectively is
about 1GB, maybe 1.5GB depending on tweaks to various
settings.





Is the one at home an OEM setup? That, and misc. things
that accumulate over time can slow a system down some.
Plus, regardless of what some MS shills claim, Win9x IS
faster than XP on any and all systems when it comes to doing
basic tasks. More demanding uses will make them more even,
when the OS becomes a less significant amount of the
processing or % of memory utilization, it becomes the less
significant factor in performance too.



Then you want a modern hard drive, to increase the memory,
and add the video card. Not necessarily in this order but
the former 2 will tend to make more of a difference for your
described uses than the latter will.




What about the memory timings? Those can account for a few
(single-digit) % performance difference as well...
especially when something is continually dependant on memory
throughput as with integrated video.



No, the CPU multiplier (ratio selection) is locked,
unchangable on Intel CPUs. You do not want to, should not
change the front side bus. You want the same FSB but if
possible, an asynchronous (+33MHz higher) memory bus. Some
boards may not provide a means for the user to change this
setting. Other boards may hide the setting in the bios menu
until you change another setting from "auto" (or another
wording) to a manual mode at which point you pick either a
frequency or a ratio for the memory like 4:3 (or 3:4 if it
were FSB:Memory).



Yes, you'd be overclocking the CPU by 33%, which it "might"
do successfully if you rasied the CPU voltage but you
expressed no desire to do this and it may not be stable even
with higher voltage or might need more cooling than possible
with the stock heatsink.



Yes, unless there is some other use you haven't mentioned
that places a greater demand on either the video subsystem
or the memory throughput.

Kony
Thanks for your comments.
Basically, therefore the pain-free zero-cost option I was seeking to
significantly improve this machines performance (reflecting Sandra's
claim that shared memory will "significantly reduce system performance)
appears to be nothing more than a chimera in my case.
OK, so at least I now know and won't keep on chasing my tail.
Best wishes
Paul