Overclocking DDR2-800

imported_wired247

Golden Member
Jan 18, 2008
1,184
0
0
My memory is corsair's 2x2GB TWIN2X4096-6400C5 G

stock timing is at 5-5-5-18, stock voltage (from SPD) is stable at 1.8V, although I have read that the memory is supposed to run at 1.9V.


Right now it is running at 400FSB so it is at a 1:1 ratio.


I am thinking about pushing the cpu a little further, and I want to make sure the memory can keep up.

Given the low voltage it runs at stock, what kind of overclock overhead do you think I have?

Or, would it be at all beneficial to try to tighten the timings? My first attempts at tightening timings resulted in mobo failing to post, but that was at stock 1.8V.

What are the odds that this ram can run at better timings than stock?

Thanks
 

QuixoticOne

Golden Member
Nov 4, 2005
1,855
0
0
Originally posted by: wired247
My memory is corsair's 2x2GB TWIN2X4096-6400C5 G

stock timing is at 5-5-5-18, stock voltage (from SPD) is stable at 1.8V, although I have read that the memory is supposed to run at 1.9V.


Right now it is running at 400FSB so it is at a 1:1 ratio.


I am thinking about pushing the cpu a little further, and I want to make sure the memory can keep up.

Given the low voltage it runs at stock, what kind of overclock overhead do you think I have?

Or, would it be at all beneficial to try to tighten the timings? My first attempts at tightening timings resulted in mobo failing to post, but that was at stock 1.8V.

What are the odds that this ram can run at better timings than stock?

Thanks


1) Several major types of current RAM chips *DO NOT* benefit significantly from
overvolting them, in fact, test indicate that some perform WORSE when overvolted.
Thus don't assume that that's the main method to improve your performance.
The best strategy is to get the maximum stable memory overclock frequency at the LOOSEST reasonable timings (e.g. 6-6-6-18 or whatever), then see what
(if any) improvement adding an extra 0.05 or 0.1V will get you in long term
stability with MEMTEST. I wouldn't feel bad about testing 1.8V SPD memory at
up to 1.9 or maybe 2.0V, but I'd have to be getting stellar, long term stable,
and consistent improvements with every extra 0.05V bump to take it up to 2.2V,
and even then....I'd be thinking it might burn them out prematurely.
Obviously SOME RAM thrives at 2.2 / 2.3, but usually it's designed / warranted for that,
and it's usually not the sort that runs WELL at 1.8 / 1.9 as yours seems to.

2) Use this to check stability. It's basically meaningless if you have "PASS" results
less than about 8-12 hours. I've seen it run for hours THEN fail, so if you're pushing
the RAM to its limits be prepared to test those limits for several overnight sessions
before you try running under a real OS. You'll just corrupt your data / crash randomly
otherwise. http://www.memtest.org/

3) Test one thing at a time. If you want to try to raise your base system clock from
400 to whatever the RAM's maximum is, fine, but don't run the CPU or other things
overclocked while you do it (as much as possible). That way if you crash you don't
know if it's the RAM, the CPU, the motherboard, etc. Set the CPU multiplier to the LOWEST
it'll go like x6 or whatever so that even at your highest possible testing clock
(e.g. 550MHz or whatever), the CPU itself is not running much or at all above stock
MHz, though your RAM may be overclocked. Similarly try the 4:5 CPU:RAM multiplier
so your RAM runs faster than the CPU etc. That way you can find the RAM's limit
without worrying about the CPU causing crashes while you memtest.

4) It's not uncommon that you could run 5-5-5-18 / 400 MHz rated RAM somewhere
up to maybe 450MHz, maybe even 500-525MHz if you loosen the timings to
something like 6-6-6-18, try to see if incremental volt bumping to 2.0V helps, putting
a fan to blow cooler air over the RAM, etc. It's just not very predictable. Some RAM
will barely get to 425MHz stable, others might get to 450 or 550. I would guess
440-460 might be where you'd start to get into problems since it starts OUT as
"not the fastest in the world" 5-5-5-18 vs say 4-4-4-12 which obviously would tend
to have more overclocking headroom by virtue of it being faster to begin with.

5) Once your base clock is so high that you could probably max out your CPU overclock
at the CPU's MAXIMUM multiplier and running 1:1 you might as well stop OCing the RAM
at that point; the performance benefit from faster RAM MHz with looser RAM timings
is not much, the main reason to OC the RAM is so you can OC the CPU up to a given
MHz at 1:1.

Good luck.
 

QuixoticOne

Golden Member
Nov 4, 2005
1,855
0
0
PS use StressPrime or Prime95 with "small ffts = stress CPU" mode
to test for CPU stability. Make sure you run one instance on EACH CPU core
you have simultaneously (I think the newer Prime95 does that more automatically
by default, but maybe I'm wrong -- check).

It should also be able to go for like 12 hours without error before you
conclude it's pretty stable.

If you find that the CPU's maximum stable MHz is much less than the RAM's
max. stable MHz (at the loosest possible RAM timings) then you can always
experiment with tightening the RAM timings or running 4:5 to get a little
more RAM performance at that point.

http://sp2004.fre3.com/
 

imported_wired247

Golden Member
Jan 18, 2008
1,184
0
0
thanks for the good info. I am still working on getting the CPU 100% stable. I haven't had it crash during anything except it did have a rounding error a few hours into prime95 / multicore

so I still have some tweaking to do before I will mess with the RAM. thanks again
 

imported_Scoop

Senior member
Dec 10, 2007
773
0
0
I'd say it's better to first check how well the RAM overclocks so you know it's not the problem when trying to go higher with the CPU FSB.
 

imported_wired247

Golden Member
Jan 18, 2008
1,184
0
0
update...... system is rock solid stable for over 12 h on prime 95 multicore small FFT

I'll keep it set like this for a while... then think about oc'ing the ram sometime in the next week.
 

Yellowbeard

Golden Member
Sep 9, 2003
1,542
2
0
I'd suggest staying at a 1:1 ratio when you do try to OC past 400 on CPU and memory. You can run up to 2.1v on that memory but, most people get the same speeds at about 1.9v-1.95v as they do at 2.0+. Put a fan on them and also I would not try to lower the latencies, go for more bus speed at CAS5. I have seen some get about 450 or so out of those modules.