F2-8500CL5-1GBPK x4 + GA-EP45-UD3L 1066 Finally!

[jh4db536]

F2-8500CL5-2GBPK (4 sticks) = 4gb ram dual channel mode
GA-EP45-UD3L v1.0 // F9 bios
e7500 @ 3.66ghz


yeah it's hybrid air & watercooled.

When i first got them, i kept getting BSOD maybe once a day, and i didn't know why. After a little bit of reading, i downclocked them down to 800 and they were fine with everything on auto/default settings. I forgot about it for a month then i started itching for 1066 again. I tried many things and failed. Then i found this forum and did some brief skimming of recommended settings. I would still fail memtest on the recommended settings, but it made the BSOD during actual use much more infrequent.

I was about to throw them out the window and buy some Kxxxn sticks in hopes of attaining real stability out of the box, but then i realized DDR2 1066 is inherently difficult regardless of brand.

I started with these settings recommended by Gskill:
CAS latency =5
tRCD=5
tRP=5
tRAS=15
tRRD=4
tWTR=4
tWR=8
tRFC=68 or 80
tRTP=4
CMD Rate=2
333MCH x 3.20 ratio = ~1066
XMP disabled
MCH Core = 1.40v
DRAM = 2.100v
Line load calibrate disabled

It didnt really work for me although it gave me more time before i failed memtest86 or OCCT. Every system must be different, i just need to tweak it a little to get it to pass. Here's what i ended up changing.



CPU-z

[jh4db536]

tRFC for me needed to be 70.

Yeah that's 2.180. It would fail memtest86 within 3 minutes at 2.1.

[GSKILL TECH]

Score! So it is stable now? Typically tRFC 80 is most stable for 8GB.

Other than that, all looks good! =)

Thank you
GSKILL TECH

[jh4db536]

I would call it 85% stable (enough). I think i can get it to drop errors if i test long enough, however, i haven't seen a BSOD or reboot during actual use that i believe is memorys fault yet. my OCed CPU is less stable than the memory right now so that is my current focus in regards to tweaking.

btw, i only have 4gb of ram (4 sticks x 1gb each = 4gb) or TWO kits of 2gb dual channel. Maybe that's why tRFC is little different.

the RAM should be able to take 2.20v right? mine are only warm at that voltage with all the fans i have going. the hottest part of my system right now is the northbridge . I'm contemplating what i should do about it. Intel's P45 max temp is pretty high on their specification writeup. The NB heatsink is pretty beefy on my MB and im not reading conclusive reviews for aftermarket aircoolers. I don't intend to watercool the NB just yet.

[GSKILL TECH]

Yes, the RAM should be fine with 2.20V. If NB temps are high, an easy solution is placing a fan over it. That's what I do, and it works great.

Thank you
GSKILL TECH

[jh4db536]

is there any particular reason that line load calibration herein referred to as "LLC" needs to be disabled? I think this is supposed to correct vdroop "vD"?

i am still getting some errors. Maybe 50 errors in 10 test runs for memtest86. That's still pretty good. I raised the DRAM voltage more last night another 0.04v with positive results. Should i continue raising the voltage?

I have a feeling that vD is causing the errors. I will post more test results when i get home from work. Actually it is distinctively obvious in my OCCT graph (above) from the CPU test. If i raise the voltage a lot that makes me fail tests earlier rather than later. That leads me to think that vD is the reason. The voltage is too high under no load causing early fail or the voltage can be acceptably high during no load and perfect during most of the testing/insufficient due to changing voltage.

I theorize that if i enable LLC i can run stably with lower voltage settings.

i have a thermaltake TPX-775M powersupply which i just put in on sunday. My first 2 posts were running with a Corsair 520HX. The reason i've been changing settings is because the powersupply has changed my voltages slightly based on graphs before and after using the same settings and also changes in test results. I was at the limit with the 520w so i can't go back.

[GSKILL TECH]

You can surely try LLC. With high overclocks, it has been known to cause some issues, but it may be different depending on how you configure your system.

Thank you
GSKILL TECH

[jh4db536]

notice the vD, cpu vcore are set at like ~1.31v in the bios of all session

1st session using HX520 Corsair w/ LLC; 520w isnt cutting it for me.


2nd session using TPX-775M w/o LLC; look at the vD, it falls from 1.28 no load to way 1.26 loaded.


3rd session using TPX-775 w/ LLC; it's trying much harder to correct the vD without spiking above the max.

[jh4db536]

I guess LLC might be dangerous for major overclocking based on an article i read. 3.66 is pretty mild. I think anything above 4.0 would be major. I'm still not seeing major vD overshoot after it was loaded and eased. I guess that means LLC really works?

I think i have acheived my target stability with this memory.
I bumped the northbridge up a little bit more, lowered the DRAM voltage a little, and increased tWR to 9. I will retest and see if it was not a "lucky" testing day. I have not seen a single BSOD or unexpected reboot so far.

[GSKILL TECH]

Yes, LLC does have a significant impact.

Glad to hear you were able to set everything up. Seems like you're doing everything correctly, so I'm sure it's not a fluke. ;-)

Thank you
GSKILL TECH

[jh4db536]

not a fluke

confirmed it's no fluke



so the setting changes from my above posts are tWR = 9 (from 8)
DRAM Voltage = 2.260v (from 2.180)
MCH [Northbridge] Core = 1.440v (from 1.40)
Driving strength at 1066 for channel A & B
LLC enabled




my northbridge heatsink at 45celcius average
gskill sticks at 36 celcius average according to my IR thermometer

these settings are minimum (for me) with respect to mch and dram voltages. If i lower 1 increment with any of them, i will get errors with memtest.
getting 4 sticks to work right is a lot of work however not impossible.

[jh4db536]

Why did i buy this ram?

I have learned a lot by playing with this RAM and i also forgot the most important thing. Why did i buy DDR2-1066 when i couldve bought DDR2-800? I only remembered once i started overclocking my CPU. Okay, ill admit that i didnt know why i bought this ram when i did, i just bought it because 1066 was higher than 800 without knowing what that really meant. I was too focused on getting this "1066"

The above posts shows my noobness. The employees of gskill & advanced users on this forum are probably laughing at my stupidity.

Everything i have accomplished in my above posts is retarded and unnecessary. I was pumping almost 2.3v into the DRAM and almost 1.50v into the NB. It's nice to know this ram is good enough to handle abuse like this; you probably have to screw up really bad to kill the ram.

assuming 266mhz was the native bus speed of my e7500 (266 x 4.0b = ~1066; 1:2 divider), (333 x 3.20c would be a ~25% O/C on the FSB)
now that i was handed down a q9550 last week the native bus speed is 333 (333 x 3.20 on a 5:8 divider *should be easy but i havent actually tried it) (400 x 2.66d on a 3:4 divider would be a 20% O/C on the FSB - this didn't work out the way i wanted)

The reason achieving 1066 (333 x 3.20c) is so difficult is because i'm not supposed to except for the sake of doing it. With a whacked memory divider (5:8; asynchronous mode is not ideal for the memory controller) like that, 4 sticks of this stuff, and me trying to overclock the cpu at the same time, no wonder i had to pump so much juice into the components to keep it stable (and gskill and gigabyte mbs are awesome components because they actually were able to do it).

Now that i am currently pushing my CPU towards 4+ghz /FSB bus towards 533mhz (1066/2=533) have i realized what i was really doing. This would probably be less desireable with DDR2-800 because it can only go up to rated (800/2=400 bus speed; 400x8.5 max multiplier = 3.4ghz pretty much the ceiling limited by ram). 3.4ghz@400fsb with 4-4-4-12 timings is not necessarily a bad thing either. The stability with a 1:1 divider even with 4 sticks and overclocking is great. I was able to reduce the voltages greatly on most components.

Running 400 x 2.66 = ~1066 was dumb as i learned once i attempted to go past 400. APPROACHING (while clamping timing down to 4's if possible) 533 x 2.00 is smarter (much easier to do with a 1:1 divider). So right now im happy around 450-475 with minimal voltage tweaking and that's getting me very close to 4ghz. So yes that means i am only around DDR-900 to 1000, but that's okay with the rated 5-5-5-15 timings because there's not a noticeable difference anyway except in synthetic benchmarks. The ram itself is probably still slower than DDR2-800 @ 4-4-4-12, but i am overclocking way more than those could achieve.

cliff notes -
i bought this ram for the 400 to 533 headroom for o/cing.

now if gskill could direct me on how to get the CAS latency down to 4 while approaching 1066 on a 1:1 divider that would be really cool. is it even possible with this model of ram? Running ddr2-1066 CL5 downclocked to ddr2-800 4-4-4-12

[GSKILL TECH]

Very informative post from a user perspective. You went from noobie to novice now. ;-)

Glad to hear you are very familiar with your computer now and you can essentially adjust everything without a problem.

But as far as lowering timings at DDR2-1066, it will require more memory and memory controller voltage. Simply reduce timings one at a time to see what the memory is capable of.

Thank you
GSKILL TECH

[jh4db536]

new approach, new results

Compared to 5-5-5-15 DDR2-1066 (could not achieve near the same o/c with massive voltages hikes)

my FPS in very demanding games increased by like 15.
i also feel that my gtx470 is no longer bottlenecked
my temps are much lower
my MCH and DRAM voltages decreased drastically. CPU Cvore is up due to the 40% o/c
This is where i wanted to be.
I chose 466FSB because it's the next 66 multiple from 400. Anything above this required a massive hike in vcore that my cooling system cant handle. I can't fit a bigger heatsink in this case unfortuntely, although i thought about adding my CPU to the GPU waterloop.

Timings clamped down to 4-4-4-12, DDR2-932, 1:1 divider
my Windows7-WEI RAM score increased by like 0.6


I had to up the DRAM voltage to 2.20 (bios); 2.10 is stable with 5-5-5-15 timings.
Check out the vdroop (LLC ON) on the ram when the system is being hammered (droops like 0.02)
Large data set to stress the ram to the max; it also passes 10reps of memtest86

[GSKILL TECH]

Sweet! That is VERY good fine tuning. Should be a great gaming system.

Thank you
GSKILL TECH