[DFINF4Ultra-D/CorsairXMS4400C25] Overclocking,memtweaking:experiences,tips,question s

[Mastakilla]

After loooots of rebooting, trial and erroring, benching, etc i FINALLY got my pc as far as it wanne go with everything tweaked to the max...

as most, i did the testing with multiple bioses...

from the default bios i allways changed the following for testing:

* disable errate 94
* disable cpy throttling
* pci pay load to 256
* disable serial ata controllers (im testing with ata disc)
* disable 1394
* cpu startup vid 1.475v
* cpu vid 1.55v x 104% (more doesnt help and less lowers my max oc)
* ltd vid 1.3 (didnt really matter though)
* chipset vid 1.6 (didnt really matter though)
* ltd/fsb divider 4x
* dram vid 2.6v (seemed to work the best with all bioses and types of cooling)
* multiplier 10x (i tried all other multipliers in all bioses i tried, and none got near the speed i got with 10x )

first i used 1/25 for a long time, it actually worked ok, it just wouldnt let me oc nice and stable...
at stock speed the defaults worked really nice

since memtest failed quite soon in 10x250 and didnt fail in 10x246, i used this for testing most of the memory options...

if it would fail 246 it was worse, if it would pass 250 it was way better (none did pass without extra cooling since i only added that later i wont talk about the 250mhz results here)

for the memory options, this page helped me out ALOT:
http://www.ocforums.com/showthread.php?t=329627

these are the results i got: (bolt i am using now)

DRAM FREQ (200 = auto)
--> 200 and 180 passed
--> 166 did NOT pass (strange uh??)

CPC (1T = auto)
--> 1T and 2T passed (1T provides 200MB/sec extra in memtest compared to 2T)

Tcl (2.5 = auto)
--> 2.5 and 3 passed (didnt write down speed difference, but who the is intrested in 3 anyway?)
--> 2 did not pass 246mhz

Trcd (4 = auto)
--> 3, 4 and 5 passed (3 provides me 60MB/sec extra in memtest compared to 4)
--> 2 not tested yet? (damn i forgot... but 99% it fails anyway...)

Tras (8 = auto)
--> 7 and 8 passed (no speed or noticable stability difference)

Trp (4 = auto)
--> 3 and 4 passed (3 provides me another 60MB/sec extra in memtest compared to 4)
--> 2 failed 246mhz

Trc (? = auto)
--> 7 and 22 passed (7 provides me a 70MB/sec extra in memtest compared to auto, BUT ONLY when having the main mem settings at 2.5 3 3 7 instead of 2.5 4 4 8!!)

Trfc (? = auto)
--> 11, 12 and 24 passed
--> 9 and 10 failed

Trrd (? = auto)
--> 7, 3, 2, 1, 0 all passed

Twr (? = auto)
--> 2 and 3 passed (BUT 2 was pretty close without dram strength set optimal)

Twtr (? = auto)
--> 1 and 2 passed (1 provides me 70MB/sec extra in memtest compared to 2)

Trwt (? = auto)
--> 3, 4 and 7 passed (BUT 3 was pretty close without dram strength set optimal) (3 provides me 150MB/sec extra in memtest compared to auto, BUT ONLY when having Trc set to 7!!)
--> 1 and 2 failed (even needed a bios reset to recover!!)

Tref (? = auto)
--> not really tested good

Twcl (? = auto)
--> 1 passed
--> 8, 4, 2 failed (all even needed a bios reset to recover!!)

interleaving (on = auto)
--> on and off passed (on provides me 150MB/sec extra in memtest compared to off)

then i tried 2/17 bios for a short while,
it didnt really help, actually i found out it wasnt 100% stable above 10x243

i didnt test the other memsettings with this bios

then the 3/10 bios finally came out!!

it made my ram alot more stable, while earlier memtest would fail at about the same speed as occt did, now my mem went alot futher...

i also discovered aiming a SUPER loud/strong fan at the mem and area right to the mem improved the stability ALOT...
10x250 was the speed i could just get now for about 30mins occt... if a setting failed sooner its worse, else its better or as good...

so i did the other memtestings with that loud fan @ 12v and occt...

here are the results: (bolt i am using now)

DQS Skew Control and DQS Skew Value (? = auto)
--> not really tested yet...

32 byte granulation (disable = auto)
--> disabled and enabled passed

DRAM drive strength (8? = auto)
--> 7 passed (improved mem stability ALOT, try it!!)
--> 1, 2, 3, 4, 5, 6, 8 failed

DRAM data drive strenght (? = auto)
--> 1, 2, 3, 4 passed

max assync latency (7 = auto)
--> 8, 9 , 10 passed
--> 7, 6, ... failed (7 was close)

read preamble time (? = auto)
--> 5 passed
--> 4, 3, 2, ... failed (4 was close)

idle cycle limit (256 = auto)
--> 256, 128, ..., 0 passed (put on 32 since 32-64 was adviced somewhere)

dynamic counter (disabled = auto)
--> disabled and enabled passed

RW queue bypass (16 = auto)
--> 16, 8, 4, ... passed

bypass max (7 = auto)
--> 7, 6, 5, ..., 0 passed

with these settings and the loud fan i got 10x250 stable

but since i hate noise, this was just for testing what became too hot

as you can read in this thread:
http://www.xtremesystems.org/forums/...ad.php?t=56214

it seems my ram and/or that resistor in the upper right of the board are getting quite hot without quite some airflow...
i also noticed my pwm ic rised to much higher temps when looping memtest 8 compared to prime or occt
(41°c max in occt... 45°C after memtest AND after bios disable memtest AND after reboot, so i guess it was like 48°C in memtest?)

so actually i dont know yet what cause is...

since i have no suitable fan yet for cooling that area, i started to look for a stable speed without that loud fan
i did some testing with the fan on 7v too, but not enough to know stability since i have to leave my case open for the construction as it is now (and that aint an option in long term)

the fastest stable speed i could find is 10x244
still gotta do some torture tests tomorrow to be sure

so my conclusion:
* i have a sucky winchester that only does 10x multiplier "good" and cant get high speeds
* i shouldnt have bought the expensive corsair mem, not that the mem is bad,
* but according to tests on madshrimps and anandtech it only gets fast from 275mhz or higher, lower its even slower then cheap mem
* i finally found a maximum stable speed of my configuration (at least i hope so) now i can start switching hds, turning sound on etc etc
* i need better active cooling on the mobo
* i still want watercooling and also on the chipset/gpu, in noise, all fans at 7v too (hope this wont become too far future plans )
* find ur optimal dram data drive strength!

i hope this can help some of you out there
if any of you has any tips or explanations for me, plz lemme know

some quick benches with videocard @ stock:
memtest ram speed: 2805MB/sec
sisoft: 6800MB/sec - 6720MB/sec
everest: 6813MB/sec - 2429MB/sec - 42.6ns
Aquamark 3 57278
3dmark 2005: 3308

[charlie]

good post

[Zedex]

Try tRAS of 9 if you haven't yet, seemed to give me nearly a 200MB/s boost in memtest and seemed more stable for me also.

[Mastakilla]

YIHAAA

10x244 really seems stable without active cooling of the pwm area!!

it just finished 8 hours (26 passes) of memtest
and previous night it ran 8 hours of prime
and a lil before that it did 1 hour occt
it doesnt crash during sisoft, everest, aquamark, 3dmark03 and 05

so i guess i can consider this as stable?

[dragao]

Man....its a professional work.......GREAT!!!!
Im DFI -D with 310 bios and Corsair 4400 2,5-4-4-8 and I try this settings....You use 10X multiplier for best performance???? How program you use for benchmark the memories and find the best settings???

Im in glue with you!!!!

[Mastakilla]

thx

dont get me wrong here, the 10x multi aint the best at all, just aim for the highest total mhz u can get with any multiplier... 10x is just the only multiplier of my sucky winchester that got me somewhere decent... actually i would have preferred something like 8x305 or 9x270 if my cpu could handle it, above my current 10x244...

allthough all this testing took me some evenings, I all did it quite quickly and rough...

fe:
with the 25/1 bios, my mem was stable for a couple of memtest test 8 loops at 246mhz,
but at 247mhz it failed quite quickly allready...

so i knew that when trying a "worse then default" memsetting at 246Mhz it would fail quickly

so i just started trial and erroring all settings... i checked the settings by running (dont be suprised) only 30-40% of 1 pass test 8 in memtest!! (so not even a full pass) (test 8 loop really seems the heaviest)

bad settings would fail, possibly good settings would pass

i only tried one each time

and once that was done i started trying em all at the same time and memtesting a lil longer... offcourse that didnt work right away, so i loosened some settings i suspected for causing the errors and started with actively cooling the pwm area, until i figured most settings out...

its a step by step job, looots of work... in the beginning just do quicktesting for finding approxamite good settings, dont start testing each setting with hours of memtest loops or u will be busy for years

[Mastakilla]

Try tRAS of 9 if you haven't yet, seemed to give me nearly a 200MB/s boost in memtest and seemed more stable for me also.

i tried it,
and for me it didnt really matter :/

so it prolly depends on some other settings too, or on the ram itself perhaps?

these were my results:

tras 9
memtest: 2805MB/sec
sisoft: 6795MB/sec - 6724MB/sec
everest: 6906MB/sec - 2425MB/sec - 43.1ns

tras 7
memtest: 2805MB/sec
sisoft: 6773MB/sec - 6727MB/sec (and the 2th try it was 6794MB/sec - 6725MB/sec, so sisoft isnt that precise...)
everest: 6901MB/sec - 2436MB/sec - 43.0ns

as u can see, not much difference

[JuanFlaiter]

They UPS guy touched my door today with some of these. I took of the heatspreaders and had a great surprise, they have TCCD.

I made it to 300x9 stable. Still working on more clocks since I can run Memtest at about 310.





Some results still at SPD in the attachments.

These are PressKits, I hope all of these come with Samsung TCCD. BTW I will post more screenies in a while.