ATT Paul --->Re: memory experts...

[ñíñjà¤têç]

I received my replacement memory(4x256mb balistix 2226) from crucial
and all seems to be running fine now in my P4P800e-deluxe mobo.

I've done memtest86 overnight and also been running P95's torture test
in the background while doing some basic file work and surfing, boy,
the ole' system sure heats up. But all tests are passing...now.

So I've been benchmarking with PCMark2k4 and trying to set my optimal
settings for system and ram and I was wondering if you do the same and
what your bios settings are for memory speed, voltage and timmings.
Also, do you manually set these:
-Dram idle time (auto)
-Dram refresh rate (auto)

My settings are as follows:
-full manual mode, pat on
-fsb @ 210 (northwood 2800 @ 2940) 1.55v
-memory @400( 2-2-2-8) 2.75v and Dram burst length @ 4 (not sure about
that one either...)

It's a bit conservative, but I'm trying lower votages to see what is
acceptable and performance hasn't been better with these settings.

How high can I go? How high are your settings ? I've also installed a
side mount 80mm fan to blow directly over the northbridge and those
"hot" Balistix modules...Thanks for any feedback!

Regards,

ñíñjà¤têç

---END---PGP---SIGNATURE---

 

[Paul]

I received my replacement memory(4x256mb balistix 2226) from crucial
and all seems to be running fine now in my P4P800e-deluxe mobo.

I've done memtest86 overnight and also been running P95's torture test
in the background while doing some basic file work and surfing, boy,
the ole' system sure heats up. But all tests are passing...now.

So I've been benchmarking with PCMark2k4 and trying to set my optimal
settings for system and ram and I was wondering if you do the same and
what your bios settings are for memory speed, voltage and timmings.
Also, do you manually set these:
-Dram idle time (auto)
-Dram refresh rate (auto)

My settings are as follows:
-full manual mode, pat on
-fsb @ 210 (northwood 2800 @ 2940) 1.55v
-memory @400( 2-2-2-8) 2.75v and Dram burst length @ 4 (not sure about
that one either...)

It's a bit conservative, but I'm trying lower votages to see what is
acceptable and performance hasn't been better with these settings.

How high can I go? How high are your settings ? I've also installed a
side mount 80mm fan to blow directly over the northbridge and those
"hot" Balistix modules...Thanks for any feedback!

Regards,

ñíñjà¤têç

I'm not really that adventurous FSB and Memory are both
stock with the Ballistix I've got. I turned Pat on, but that
is about it. (My lack of overclocking experiments on the
P4C800-E has a lot to do with having to reenter the BIOS
settings, if I have an overclocking failure. Takes a lot of
the fun out of it. Hell, I even lost my BIOS settings one
day, when I plugged the keyboard and mouse into the wrong
ports by accident.)

DRAM burst length should have something to do with the cache
line size. The burst should be large enough to fill a cache
line. Dual channel is 16 bytes per cycle, so a burst length
of four totals 64 bytes. Perhaps if a person was running
single channel, then you set the burst size to 8 to
get 64 bytes ? If the burst size is too short, then a
second burst would be required to finish the cache line
operation, but I don't know what happens if the burst
is larger than a cache line.

You can play with refresh by extending the time between
refresh operations, and on some of the private forums,
they like to push that option. But what that does, is test
how much better the memory is than its spec, in terms of
the ability to retain the charge on each DRAM cell. I
don't really see the profit in playing with refresh myself,
as the benefit might be a small one. It is possible there
is a test in memtest86 that will be good at verifying
retention (fills the memory, then waits a long time
before verifying it), if you wanted to spend the time
testing it.

There is some BIOS settings info here:
http://www.rojakpot.com/freeBOG.aspx

I think I invested more time on the A7N8X-E, because it was
so picky about RAM. That generation of Intel boards is
hardly a challenge by comparison, and other than just
cranking the clocks on it, I don't think it really needs
a lot of tweaking. Since my processor would only give
me about 3.5GHz, I don't have a lot of reason to fiddle with
it. And the box runs warm enough as it is.

With 256MB modules, I guess those are single sided modules ?
That should allow a higher memory clock, if you wanted to try
it. Since there is a Ballistix bin for DDR500, they should
get up fairly close to that. Maybe you could get to 240Mhz
memory clock (DDR480) with them, but with the CAS set to a
higher value. (I just checked the Crucial site, and it is
interesting to see that the PC4000 256MB Ballistix module
is rated CAS2.5, while the PC4000 512MB module is rated CAS3.
They would both be using the same part number of chips. I
think I'd start at a setting of CAS3, as most of the advantage
of overclocking the memory, is the clock rate. Shaving
0.5 from CAS really isn't going to help a lot.)

On the private forums, bumping the voltage doesn't seem to
help Ballistix that much. One guy managed to fry his sticks,
the first day he got them. They are not tolerant of voltage
like BH-5 would be.

Paul

 

[Mercury]

. It is possible there

is a test in memtest86 that will be good at verifying
retention (fills the memory, then waits a long time
before verifying it), if you wanted to spend the time
testing it.

there is a test that moves a pattern a to b to c to ..... and eventually
tests the pattern (#5), however I don't know if that would be a good test as
any reference to DRAM always results in a refresh (to a 'width' ?) and the
speed at which memtest86 clobbers DRAM may be much too fast to let refresh
come into it.

I have always been happy that modern systems handle refresh automatically
and AFAIK are extremely mature on that design front.

See:
http://www.memtest86.com/#details

the new bitfade test may also contribute, but the test obviously is not for
refresh problems (the wait is long)...

 

[ñíñjà¤têç]

While doing my benchmarking, I must report an interesting
observation...

I bumped up my fsb to 220 (3080) @ 1.575v and ran P95 for almost 2hrs
as well as PCMark2k4, all without incident. During the Prime run, my
cpu temp hovered around 50c. I ran the tests again, but with a lower
cpu voltage and my cpu temp increased a few degree's. This was
verified with PC Probe and speedfan!

As I type this, PCMark has ran successfully ( P95 will run shortly),
but my fsb is @ 225 (3150), votage @ 1.575 and cpu temp is @ 36c idle
and 40c during PCMark! Seems I've found a sweet spot ands it's pretty
darn cool.

All ram setings are stock DDR @3338, (except for the 2.75v), but my
scores are going through the roof! My test chart thus far is proving
beyond a doubt that cas latency plays second fiddle to clock speed or
fsb gains.

It's nice to finally have some good ram to play with and yes, they are
all single sided dimms.
Regards,

ñíñjà¤têç

---END---PGP---SIGNATURE---

 

[ñíñjà¤têç]

While doing my benchmarking, I must report an interesting
observation...

I bumped up my fsb to 220 (3080) @ 1.575v and ran P95 for almost 2hrs
as well as PCMark2k4, all without incident. During the Prime run, my
cpu temp hovered around 50c. I ran the tests again, but with a lower
cpu voltage and my cpu temp increased a few degree's. This was
verified with PC Probe and speedfan!

As I type this, PCMark has ran successfully ( P95 will run shortly),
but my fsb is @ 225 (3150), votage @ 1.575 and cpu temp is @ 36c idle
and 40c during PCMark! Seems I've found a sweet spot ands it's pretty
darn cool.

Ran P95 for 1+1/2hrs, A-O-K!

All ram setings are stock DDR @3338, (except for the 2.75v), but my
scores are going through the roof! My test chart thus far is proving
beyond a doubt that cas latency plays second fiddle to clock speed or
fsb gains.

Forgot to mention here that @ just 5% increase in fsb, I was able to
match memory scores. So it looks like this: (PCMark2k4)

-cpu @2800(200fsb)+balistix ddr400(2-2-2-6)=memscore 4599
vrs
-cpu @2940(210fsb)+stock ddr400(3-3-3-8)=memscore 4591 +(pcmark
score+150, cpu score+215, grafix score unchanged)

FYI

It's nice to finally have some good ram to play with and yes, they are
all single sided dimms.
Regards,

ñíñjà¤têç

---END---PGP---SIGNATURE---

ñíñjà¤têç

---END---PGP---SIGNATURE---

 

[Paul]

Ran P95 for 1+1/2hrs, A-O-K!


Forgot to mention here that @ just 5% increase in fsb, I was able to
match memory scores. So it looks like this: (PCMark2k4)

-cpu @2800(200fsb)+balistix ddr400(2-2-2-6)=memscore 4599
vrs
-cpu @2940(210fsb)+stock ddr400(3-3-3-8)=memscore 4591 +(pcmark
score+150, cpu score+215, grafix score unchanged)

FYI

Yes, "Clock rate is king" is the rule. There might still be room
to tune the memory timings a bit. For example, you might be able
to squeeze CAS2.5 out of the memory - that is assuming the chipset
doesn't have any restrictions. On my A7N8X-E, CAS2.5 and CAS3 have
exactly the same memory benchmark, implying that is one chipset
that cannot handle the fractional CAS values (it means possibly
the data phase on the memory, cannot start on both rising and
falling clock edges, but must be aligned with one of them). I'm
not aware of that problem with Intel.

I cannot explain CPU temp rising with a lowered Vcore. There are
devices that have non-linear responses to voltage, but I've never
heard of something like that with processors.

Paul