P4C800-E Deluxe - CPU speed vs Timming vs Memory speed?

[NEM]

Hi,

I've spend a bit of time going through and writing down all the various
setting combinations for OC'ing the system listed below. It boils down
to CPU speed vs memory speed. Which is more important in your opinion?

For example...

CPU |FSB |CAS |MB/s
------+--------------+-------------+------
3375 |225 (DDR450) |2-3-2-6-4 |2929

....vs...

CPU |FSB |CAS |MB/s
------+--------------+-------------+------
3405 |227 (DDR454) |2.5-3-2-6-4 |2893

....I know, I know, there isn't a lot of differences, but these are
rather extreme example settings. The differences are more apparent with
lower OC'ing, something of which I plan to eventually use.

System currently runs 40c under full load using stock heat sink (lapped
and AS5 used) and Motherboard at 27c using the above settings. All
memory test were run using Test#5 repeatedly (30+ passes) in Memtest86+
v1.0.

System:
~~~~~~~
Asus P4C800-E Deluxe
Mushkin PC3500 Level II 512MBx4
P4 3.0c
Antec PLUS 1080AMG case

Thanks for reading,

 

[Paul]

Hi,

I've spend a bit of time going through and writing down all the various
setting combinations for OC'ing the system listed below. It boils down
to CPU speed vs memory speed. Which is more important in your opinion?

For example...

CPU |FSB |CAS |MB/s
------+--------------+-------------+------
3375 |225 (DDR450) |2-3-2-6-4 |2929

...vs...

CPU |FSB |CAS |MB/s
------+--------------+-------------+------
3405 |227 (DDR454) |2.5-3-2-6-4 |2893

...I know, I know, there isn't a lot of differences, but these are
rather extreme example settings. The differences are more apparent with
lower OC'ing, something of which I plan to eventually use.

System currently runs 40c under full load using stock heat sink (lapped
and AS5 used) and Motherboard at 27c using the above settings. All
memory test were run using Test#5 repeatedly (30+ passes) in Memtest86+
v1.0.

System:
~~~~~~~
Asus P4C800-E Deluxe
Mushkin PC3500 Level II 512MBx4
P4 3.0c
Antec PLUS 1080AMG case

Thanks for reading,

Generally speaking, "clock rate is king". Whatever you lose via having
to increase the CAS setting, you can more than make up by the ability to
increase clock rate (at least until something else breaks).

So, for example, people buy PC4000 memory, because it runs at DDR500.
DDR500 is 25% more than DDR400, but the memory might have to run at
CAS 3 instead of CAS 2 . The CAS increase costs 10% of your performance
increase, leaving you with a net increase of 15%.

In your example, at the point you are forced to change from CAS2 to
CAS2.5, that is a bad point to leave the clock at. This is because
you've taken a hit on performance due to the CAS increasing, but you
haven't increased the clock enough to "pay back" what the CAS change
has cost you.

Clock speed alone should be giving you a "linear" performance ramp,
but every time CAS changes, that is a downward "stair-step". The
sum of those two functions gives you the memory bandwidth
performance curve.

Because of the fact that clock rate is king, that is why people
are happy to buy 3-4-4-8 memory for PC4000, because the higher clock
rate makes the memory timings irrelevant. (As long as you are harnessing
the extended clock rate range of course - buying PC4000 memory and
running it at DDR400 doesn't make too much sense. If a user plans on
staying at stock speed, the money would be better spent on a PC3200
low-CAS memory.)

HTH,
Paul

 

[NEM]

Hi Paul,

On Feb 08 2004, Paul wrote:

[...]

Generally speaking, "clock rate is king". Whatever you lose via having
to increase the CAS setting, you can more than make up by the ability to
increase clock rate (at least until something else breaks).

So, for example, people buy PC4000 memory, because it runs at DDR500.
DDR500 is 25% more than DDR400, but the memory might have to run at
CAS 3 instead of CAS 2 . The CAS increase costs 10% of your performance
increase, leaving you with a net increase of 15%.

But isn't true that at this time chips are not made specifically for
anything above PC3200. If that is the case, then PC4000 is PC3200
memory, but with lose SPD settings.

Speaking of the SPD, I found that the SPD was read wrong by the
motherboard, so manual timings were almost certainly needed.

The default SPD settings for the Mushkin PC3500 is 2-3-3-7. Mushkin
advertises the sticks at 2-2-2-?, I've read that the last digit is
suppose to be a 5, but I didn't have any luck with 2-2-2-5, so I bumped
it up to 6 and started to have error free results.

In my situation, I found one stick to be a little weaker then the other
three (even though they were purchased all at the same time as two
matched pairs). It wasn't until I took it off of auto SPD, and started
manually setting the timings that the weak stick fell in line with the
other three. I'm also told, that because of the amount of memory (2GB),
that I am fortunate to have all four working well together.

In your example, at the point you are forced to change from CAS2 to
CAS2.5, that is a bad point to leave the clock at. This is because
you've taken a hit on performance due to the CAS increasing, but you
haven't increased the clock enough to "pay back" what the CAS change
has cost you.

I stopped test at that point because of the increase in CPU speed/heat.
I felt that going further then 10 percent would be shortening the live
of the CPU unless I took further steps to cool it down.

So in my example, I would probably be better to leave it at 3.375GHz
since that is the last working combination with the faster 2-3-2-6
timings.

After 31 tests, so far these are the working results according to the
testing method mentioned above...

CPU |FSB |CAS |MB/s |Volts
------+--------------+-------------+------+-------
3060 |204 (DDR408) |2-2-2-6-8 |2781 |2.75
3255 |217 (DDR434) |2-2-2-6-8 |2825 |2.75
3270 |218 (DDR436) |2-2-2-6-4 |2838 |2.75
3284 |219 (DDR436) |2-2-2-6-4 |2850 |2.75
3314 |221 (DDR440) |2.5-2-2-6-4 |2815 |2.75
3329 |222 (DDR442) |2.5-2-2-6-4 |2828 |2.85
3345 |223 (DDR446) |2.5-2-2-6-4 |2842 |2.85
3345 |223 (DDR446) |2-2-2-6-4 |2903 |2.85
3359 |224 (DDR446) |2-2-2-6-4 |2916 |2.85
3375 |225 (DDR450) |2-2-2-6-4 |2929 |2.85
3390 |226 (DDR452) |2.5-2-2-6-4 |2880 |2.85
3405 |227 (DDR454) |2.5-2-2-6-4 |2893 |2.85

....I've also locked the AGP/PCI to 66.66/33.33 and left the divider to
1:1.

Clock speed alone should be giving you a "linear" performance ramp,
but every time CAS changes, that is a downward "stair-step". The
sum of those two functions gives you the memory bandwidth
performance curve.

Like a sawtooth wave. Going to 2.5 is like being dropped off the edge of

Because of the fact that clock rate is king, that is why people
are happy to buy 3-4-4-8 memory for PC4000, because the higher clock
rate makes the memory timings irrelevant. (As long as you are harnessing
the extended clock rate range of course - buying PC4000 memory and
running it at DDR400 doesn't make too much sense. If a user plans on
staying at stock speed, the money would be better spent on a PC3200
low-CAS memory.)

I didn't think there was much point doing that with the stock processor
speed of 3.0GHz. If I had a 2.4GHz that might be something to do because
of the mulitpler and divider, so I felt that I could squeese out the the
Mushkin PC3500 BH5s. Was that incorrect thinking?

Thanks Paul.