Need help with new memory!

Do you have a question? Post it now! No Registration Necessary.  Now with pictures!

Threaded View
Alright..I'm a gamer, so I was looking to beef up the 1gb in my system
and bought a 2gb kit. Before I get into the problem, here are some
relevant specs:

CPU is P4, 3.2 Ghz.
Motherboard is Asus P4P800-E Deluxe, recently flashed to latest stable

400MHZ 184PIN". Runs at 2.5-3-3-8, if that matters at all.

New RAM is: "OCZ EL Platinum PC3200 2GB 2X1GB DDR400 CL2-3-2-5 184PIN
Pin Dual Channel Memory Kit /W Ramsink". Runs at 2.5-3-2-5. Should be
capable of Cas 2, but I can't seem to get the machine to post when I
set it to 2.

So, both old and new are dual channel, ddr400, pc3200. There are 4
slots on the motherboard.

The problem:

The new 2gb RAM seems to be performing the same/worse than my old 1gb.
Using Sisoft Sandra's memory bandwidth (int/float) test:

1gb kit averages ~4800/4800
2gb kit averages ~4320/4320
Putting in all 3gb -> Half the results come to around 4600/4600, the
other half ridiculously low (examples: 1712/4663, 2943/2158, 4669/2375)

Then I tried testing loading up a certain saved game in Doom3.
1gb kit: consistently 47 seconds
2gb kit: consistently 47 seconds
All 3gb: 60 seconds

Is there a problem with the new ram (unlikely?), or am I missing
something here? Any suggestions? Any way to make the system like the
full 3gb more (or even just the 2gb alone if it comes to that)? I'd
hate for this to have been a useless purchase. Thanks in advance for
any help.

Re: Need help with new memory!

On 29 Jul 2006 20:07:24 -0700, wrote:

Quoted text here. Click to load it

Keep in mind that the primary function of increasing memory
is to be able to do larger jobs, or to reduce swapfile
paging.  In itself, there should not be a large change in
performance for any tasks that fit within your old memory
space, particularly so when the timings both sets used
aren't so different.

Quoted text here. Click to load it

This also measures the CPU cache, IIRC, so replacing one
memory module with another isn't expected to make much
difference unless again, the timings were vastly different.

What you could do (though a risk of instability) is to see
if your system stays stable with a higher FSB & memory bus.
Yes it's overclocking but if you insist on more performance
doing same things, it may be the main benefit this newer
memory provides (again, when the tasks already fit in 1GB
you previously had installed).

Quoted text here. Click to load it

It would seem from the above that your bios is automatically
setting a parameter slower with the 1+1GB pair.  You could
go into your bios and see if there's any applicable settings
to tweak but I dont' know what it would be nor if it will
remain stable.  Overall, the hit to go to 2GB isn't so bad,
you might be content now if it's stable (memtest86+ will run
indefinitely without errors).

Quoted text here. Click to load it

Level loads should mostly be a function of HDD speed.  Even
so, it's been a while since I played Doom3 but 47 seconds
seems quite slow.  I don't recall exactly what the gaming
system was like when I was playing Doom3 but possibly an o'c
Athlon Barton and 120GB HDD.  If your HDD is older you might
upgrade the HDD if game level load times are important...
actually I think I could load Doom3 saved games across a GbE
network share in 47 seconds, so there should be a lot of
room to improve that aspect of gaming, but in general I
don't think the memory is the answer, except that it seems
having ALL of it installed is defaulting you to horribly
slow memory timings and thus you should avoid doing that,
except if you can tweak the bios settings to regain some of
that, but you'll never be abe to tweak all 4 modules to the
same (highest) performance possible as with only two of

Quoted text here. Click to load it

Do you *need* 3GB installed?
1GB is going to be a little less than optimal for next-gen
games and Windows Vista, so 2GB seems a nice amount to have
though we can't know how long you'd keep the system instead
of going to next-gen, DDR2 based platform.

I'd have thought about that already though, that buying 2GB
of memory today for a system already having 1GB is a tough
call... when it's DDR(1) memory that'll be abandoned the
next time the system is rebuilt/upgraded/etc, if it will be.

I'd pull out the  2 x 512MB and either use in another system
or sell it while it still has some value.

Re: Need help with new memory!

Quoted text here. Click to load it

Have you upped the voltage on the memory? To get the full benefit you need
to move the memory up to 2.8v. Check the OCZ website to see what the top
voltage that they recommend is.

Also, overclocking the CPU will give a bigger bang that tweaking those
memory timings.

My P4C800E-Dlx has a 2.6 Northwood running at 3.15Ghz and doing it cool,
with memory at 1:1 (2x256meg PC4200 OCZ memory)

Re: Need help with new memory!

In response to the DDR vs DDR2 issue:
My motherboard doesn't support DDR2. If I change mobo, I'm gonna have
to change CPU, PSU, video the very least. After this RAM
update (provided it goes well), I only intend on sticking in a good AGP
video card (which will be cheap soon) and then riding the wave for 2-3
years til I change computers.

In response to the voltages issue:
I have now begun messing around. I'm currently running the 2x1gb kit
alone. Voltage is at 2.75, timings tightened to 2-2-3-5. Memory
bandwidth test comes to.....exactly the same. Approximately 4350/4350,
still much slower than the 2 year old Kingston memory. Bandwidth
efficiency is ~68%.

Is this a defective unit?

Re: Need help with new memory!

Quoted text here. Click to load it

When installing RAM in a motherboard, it pays to use a utility
as an "impartial observer". For example, the BIOS setup screen
on my P4C800-E doesn't display the same timing numbers as are
seen in CPUZ. As a result of that, I don't trust the BIOS
screen any more, to display what is going on. (shows timing while in Windows)

If the memory timing numbers were exactly the same for the
2x512MB test case and the 2x1GB test case, the measured bandwidth
should be the same. You can check that with CPUZ and see if both
tests are applying the same conditions.

Everyone has a different test they like to use, and my favorite
for comparative bandwidth numbers, is the bandwidth display
in memtest86+ ( ). That is the third number down on
the screen, when the program is running. The reason I like this
test, is source code is available for the program. I have even
made a three line modification, so I could measure memory bandwidth
in different areas of the RAM (for testing how dual channel works
on 3 slot Nforce2 systems). Basically, the test is as simple as
can be, and involves a cache flush by reading more data than the
processor cache can hold, followed by repeated timed reads of
large blocks of sequential RAM. The computed bandwidth is
displayed on the screen.

(As an added bonus, memtest86+ will also test that your
memory is error free :-) )

Sandra could give different results, because it might be
using a stride greater than 1, which could involve factors
related to the page size of the RAM. At least with the simple
minded test in memtest86+, I feed more confident that I am
looking at timing related effects.

Now, in the case of mixing 2x512 and 2x1GB, I would think that
interleaving might be disabled. If you look at this document,
your mixed config is called "normal mode", and is listed as
the fifth entry down in the table on the last page of the
document. I don't understand everything that is in this
table, such as why 4 DIMMs with single rank has the same
performance as dual rank (dual rank should be able to keep
more pages open and should be better). But it does make
sense that a "normal mode" config will be slightly less
than the "dynamic mode" cases. Note that the steps in the
table are not the same size, and 4 DIMM normal mode is
a lot better than any of the single channel modes. It
would have helped if Intel had put performance numbers
in the table.

"865PE Chipset Memory Configuration Guide"


Re: Need help with new memory!

For the record, I have been using CPU-Z to discern about timings and
such. My previous post still holds true, I'm currently running the
2x1gb kit at 2-2-3-5 (and Sandra scores still disappointing).

Now, I had run memtest for stability and it seemed OK. As for the third
number referring to this line?:

Memory      2047M      2611 MB/S

Seems low, though I have nothing to compare it with. Either way, you
seem to know a hell of a lot about this...what would you do in this
situation, or what do you suggest I test next?

Re: Need help with new memory!

Quoted text here. Click to load it

Well, first thing you do, is plug in the 2x512MB config and
repeat memtest. Does the screen report the same bandwidth as
your 2611MB/sec for the 2x1GB config ? Are the timing numbers
used, exactly the same, as verified with CPUZ ? The value of using
memtest86+ is not the absolute number (which usually is depressingly
low). The value is in comparing different RAMs or different timing
choices, to see which ones are better or worse.

As for your running config, I would choose 2x1GB, assuming
you can figure out what is going wrong.

Oh, and by the way, try experimenting with Tras (the "5" in your
timing numbers). Try a value of 8 or so, and see if things
change. If you follow some of the testing on Anandtech, the
policy on Tras changes from situation to situation, and the
lowest number is not always the best choice. (The lowest number
is the best choice, for the first three digits.) Mind you, I've
never wasted the time trying all the Tras values on any of
my systems...

Another factor to check, is MAM (also known as PAT on the
875 chipset). That is good for a few percent performance, and
changing the memory sticks might have throwh off that setting.
Now, I don't know what the best util is today for checking that.
CPUZ and CTIAW are two possible utils you could try.

(CTIAW is one util for checking PAT. It runs in a DOS window.)

On my P4C800-E, I get 2732MB/sec in memtest86+, with 2-2-2-6
memory 2x512MB PC3200 with PAT enabled. With PAT disabled I
get 2667MB/sec. On my board, PAT is enabled by setting
memory performance mode to "Standard". Your board has a
separate "Memory Acceleration Mode" setting, and as long
as you are at FSB800/DDR400 with two sticks, I think it should
work. You can force it to [Enabled] and see what happens.

When I get a new board, I spend a considerable amount of
time fiddling with the memory :-) Especially when it isn't
running right. And I always use memtest86+.


Re: Need help with new memory!

Quoted text here. Click to load it

I have run more extensive tests. Here are the memtest results:
1024 MB Kingston ram at 2.5-3-3-8 (stock) = 2557 MB/S
2048 MB OCZ ram at 2.5-3-2-5 and no tweaks = 2557 MB/S
2048 MB OCZ ram at 2.0-3-2-5 and tweaks = 2666 MB/S
(more tweaks = 2725 but system was unstable)

This makes a little more sense, but it's hardly impressive. Big
question: Is this a qualitative rating alone, or does it take into
account the total amount of RAM? In other words, is the only
performance boost the boost from 2557 to 2666, or is it the fact that
the memory itself is that much faster, AND there's now twice as much of
it? I'm certainly hoping it's the latter, it'll make me feel better.

Quoted text here. Click to load it

I played around and there was no difference in memtest or Sandra
testing, so I've left it at 5 for now.

Quoted text here. Click to load it

I have played with the Performance Mode and Acceleration Mode settings
to obtain the optimal results for each set of tests presented.

Quoted text here. Click to load it

PAT Enabled makes my system hang, but setting to Automatic seems to
enable performance mode without any problems. Memory Accel mode is
enabled in all my test results.

You can see the memtest results higher up in this message.

Sisoft Sandra Memory Bandwidth test shows the following:
2x1024 OCZ RAM with no tweaks and 2.5-3-2-5: 4319/4343 (67% bw
2x1024 OCZ RAM with full tweaks and 2.0-3-2-5: 4854/4854 (75% bw
2x512 KINGSTON RAM with no tweaks (not made to be tweaked..): 4849/4852

Still makes little to no sense on this front, as the 2 year old
Kingston is still kicking the OCZ's ass.

Here are some screenshots from CTIAW and CPU-Z, in case you need some
info: CTIAW
(Kingston) CPU-Z
(Kingston) CTIAW (OCZ,
pre-tweaking) CTIAW
(OCZ, post-tweaking) CPU-Z (OCZ,
pre-tweaking) CPU-Z
(OCZ, post-tweaking)

Big note here, CTIAW incorrectly reports my FSB and DDR MHz to have
lowered after tweaking. It seems wrong, as neither the BIOS nor CPU-Z
seem to think so.

Geez, long post. Alright, to the point: What do you make of these
results, and do you have any further counsel for me? Should I try
getting the Kingston and OCZ to like each other, or keep OCZ and sell
Kingston? (or keep Kingston and sell OCZ...heh)

Thanks again for your continued help.

Re: Need help with new memory!

Quoted text here. Click to load it

OK, you are probably ahead on the timing. (CAS2 versus the other
memory at CAS2.5) You have twice the memory, and it is a little
bit faster. Twice the memory will help you if you were previously
running into problems while having only 1GB of memory (swapping to

Since many OSes use spare memory as a file cache, certain
file system operations can be speeded up as well (my favorite
was running a checksum program on a 1GB file, and the hard drive
light didn't even blink when I ran the program a second time against
the same file).

When you install the 3GB config in the machine, you will probably
end up at CAS2.5, PC3200, normal mode instead of dynamic, and
MAM may disable itself (not sure, but MAM may be limited to CAS2).
So in that case, you have even more memory to play with, if for
example you had managed to exhaust 2GB. If you were a Photoshop
professional who works all day with large high res images, you
might choose the 3GB configuration, even if the benchmarks were

There might be one game that gets a benefit from 2GB of memory.
My two most current PCs here have 1GB each, and I haven't really
run into a situation where I felt motivated to move the memory
out of one machine into the other. And in terms of memory
benchmarks, what is really weird, is running 4x512MB matched
RAM gives the same bench as 2x512MB. I figured the dynamic mode
differences would make a difference, but not in the memtest86
bandwidth test. I suppose Sandra might get a different number
for the two configs, but I really don't use Sandra that much.
(I did buy a copy.)

I would say your 2GB configuration is now better than your
1GB configuration, as it got a little boost in bandwidth, and
the latency is reduced by being at CAS2. And 2GB should be
plenty, even for video editing or Photoshop. There are a few
applications that can use "infinite" RAM, like large logic
simulations in chip design, but that is hardly something
you'd be doing at home :-)

If you need the cash, sell the Kingston. I'd wait a month
and make sure you are really happy with the OCZ, then put
the Kingston out for sale. (I'd game on the OCZ for a while
and see if it is perfectly stable.) Maybe in that month, you'll
think of a reason to use the whole 3GB :-)


Re: Need help with new memory!

Paul wrote:
Quoted text here. Click to load it

Awesome, thanks for all the help. I'm leaning toward keeping the new
2GB and selling the Kingstons to help reduce the upgrade cost. I
actually checked out some specs and it turns out they're far more
OCable than I'd thought, so I almost feel bad letting go of them, but
they simply won't "play nice" with the OCZ memory.

Oh, and I hear more voltage isn't necessarily better. If I can get all
my current settings running on a voltage like 2.65, should I perhaps
just keep it there to extend lifespan?

Can you suggest any additional methods for testing (I'll be running
memtest and prime for the next 2 nights) or extra tweaks to get more
out of these OCZs? Those simple BIOS tweaks helped a lot.

Re: Need help with new memory!

Quoted text here. Click to load it

Excess voltage is not necessary, as you note. The JEDEC spec for
PC3200 is 2.6V+/-0.1V , so the 2.65 is "normal" in a sense. I
would not feel the least bit nervous at 2.7V. In any case, if
Prime95 is passing the stuff for hours on end at 2.65V, then
leave it. If later in life, it gives you a problem, you'll have
room to crank it.

The overclockers have their favorite tools for establishing
stability. You could browse here, and find out all the latest
techniques: /

I feel memtest86+, Prime95, and 3DMark demo loop are a good start.
Also, gaming, using a game that is known not to be a buggy pig, is
pretty effective. If you can game for 12 hours straight and
not have a crash, that is generally a pretty good sign you've
done the job right.


Site Timeline