HELP NEEDED WITH CORSAIR DOMINATOR GT!!

[ZeroFatal1ty]

I'm kind of a noob when it comes to tightening timings. I just bought this kit: CMT8GX3M2A2000C9, and I want to keep the 2000Mhz Freq. I just want to tighten the timings as much as possible. Please also let me know what my safe max voltage is for these. I've heard they can do 1.7v just fine. My specs are as follows:

ASUS Crosshair V Formula (Bios 1301)
AMD Phenom II X6 1090T
Corsair Dominator GT 2x4GB
2x GTX 470's in SLI @ 825/1650/2000
2x Corsair Force GT 120GB in RAID-0 128k
WD VelociRaptor 600GB Sata 6GB/s (Data Drive)
ASUS BD-RW
Cooler Master 932 Advanced
Thermaltake FRIO with Cooler Master SickleFlow Push/Pull
OCZ ZX-1250W PSU
Windows 7 Ultimate

Current OC Settings:

FSB: 250
CPU: 4000Mhz @ 1.4875v
CPU/NB: 3000Mhz @1.35v
HT-Link: Stock
CPU VDDA: 2.6v
NBv: 1.175
DRAM: 2000Mhz @1.65v
9-10-9-27-41-2T

LLC: Extreme
CPU/NB LLC: Extreme

 

[King107s]

I used to have this RAM in tri-channel on an LGA1366 board so voltages for your AMD set up I can't help you with but what I can tell you is that thoes timmings are about as low as youre going to get with it still being 100% stable at 2000MHz... getting one of the CAS timmings to drop an increment would provide next to zero performance gain anyway.

On the LGA1366 setup I set the mainboard to run the xmp profile for the RAM and set the DRAM voltage to 1.65V and Bumped QPI/DRAM Core voltage (Vtt) to 1.55V to get that RAM running right.

 

[Woomack]

@ZeroFatal1ty
Most RAM kits will work fine with voltage +/- 0.05V so 1.70V is ok.
I would set this memory to 1866 and try to lower timings to 8-10-8-27 or 9-9-9-27 or 8-9-8-24 keeping 1.60-1.70V. On this setup 2000+ won't help in performance and I doubt that you will be able to drop timings much @2000.
Other thing is that Phenoms don't like anything above 1866 and even if you set it higher then you will get almost no performance improvement. For higher memory bandwidth I would play some more with CPU-NB clock ( just question if it's worth it as you already have 3000 what is quite nice ).
If you really want to have your memory @2000 then try the same as above so 8-10-8-27 2T, 9-9-9-27 2T, 8-9-8-24 ( or 27 ) 2T 1.6-1.7V.
1.75V should be still safe but if I'm right then going above 1.7V won't help much with this kit.

 

[ZeroFatal1ty]

Thank you both for your advice! Woomack, your reply was very insightful and you seem to be an expert in this field The reason I want to keep 2000Mhz, is because the FSB of 250Mhz is a nice round value which works nicely with the CPU & CPU/NB Freqs. If I go by Multiplier, than the ram sits at 1600Mhz, and if I go with DOCP @1866Mhz it changes my FSB to around 234Mhz which throws everything else off. I've worked very hard at getting this CPU to 4Ghz with 3Ghz CPU/NB and getting it stable, (24Hrs Prime95 Blend, 20 passes Linx max memory, and 4 passes 32Mb HyperPi). That's why I'm sticking with 2000Mhz.

I will try those timings, one at a time @1.70v and then maybe 1.75v. Is there any program i can use for ram temps and how hot is too hot. Also, memtest on phenom X6 only registers 1600Mhz; what else can I use for Ram stability test. Thanks in advance!

 

[King107s]

Don't go beyond 1.70V.... really all it needs and should get is 1.65V

 

[ZeroFatal1ty]

Thanks King107s... I'll stick with 1.70v, but I do know that there is a little more room for voltage adjustment on AMD platforms, as compared to intels and their limitations to memory voltage. I personally like to stick with stability/longevity rather than extreme performance with the loss of lifespan.

Do I need to adjust tRC, or just leave it at auto? And how about advanced timings, or should I leave everything else at auto as well?

I found ICH MemTest for stability testing. Is there anything else you can recommend?

**I did read on another forum somewhere to run 4 passes 32MB HyperPi for memory stability. Can anyone validate this?

 

[moocow]

Thanks King107s... I'll stick with 1.70v, but I do know that there is a little more room for voltage adjustment on AMD platforms, as compared to intels and their limitations to memory voltage. I personally like to stick with stability/longevity rather than extreme performance with the loss of lifespan.

Do I need to adjust tRC, or just leave it at auto? And how about advanced timings, or should I leave everything else at auto as well?

I found ICH MemTest for stability testing. Is there anything else you can recommend?

**I did read on another forum somewhere to run 4 passes 32MB HyperPi for memory stability. Can anyone validate this?

I think even one pass of hyperpi is more than enough

 

[ZeroFatal1ty]

Thanks Moocow for the reply... Really, you think just one is good enough for 100% DRAM stability? I'm not being sarcastic.. I'm just super OCD about stability because I'm running my OS on SSD Raid-0 and can't afford BSODs. I'll go ahead and run 4 passes anyway.

I also heard about memtest, a program with windows GUI. Says you have to run an instance for each core though. I guess that means setting the affinity in task manager.

**Update: Could not post at 2000Mhz 8-9-8-24 2T 1.7v. Posted with 2000Mhz 8-9-9-24 1T 1.7v. Finished one run 32MB HyperPi.

Should i put tRC at 33. tRP + tRAS = tRC right? Right now its 41(Auto)

 

[Woomack]

tRC depends from some other things. It's more like higher tRC = lower performance but higher stability. Minimum tRC you can count like tRC = tRAS + tRP but you will achieve higher stability ( at least in theory ) setting it higher.
About adding timings there is also CL+tRCD+tRP = tRAS ( or tRAS -1 ) but again it depends from memory and clock so can be lower or higher. Hynix IC has it higher while Elpida/PSC usually lower.

If I'm right then your memory is on Hynix IC so pumping more voltage won't help much. If you made it work @ 8-9-9-24 1.7V then I would try something like:
8-9-9-27 2T ,42-48 tRC 1.65V or just higher tRFC like 160
also try 8-9-8-27 2T up to 1.70V using higher tRFC maybe even 300

About soft, Intel Burn Test is ok , just set custom test / memory to ~90-95% and let it pass. 5 passes that are in standard test will take about 1h.
Memtest for windows is also ok. In this case you have to run as many instances as you have cores to load memory controller. About 300% pass is enough.
As moocow said , 1 pass of hyperpi should be enough to test memory ( at least for daily usage like games etc ).

 

[King107s]

Memtest86 was just updated as well and has always been my first choice for memory testing.... Can also run Prime95 Blend

 

[ZeroFatal1ty]

Hey, just wanted to thank you guys again. I ended with 2000Mhz 7-9-8-24-32-1t 1.70v. Passed multiple runs HyperPi, I left memtest GUI on overnight for the last 3 nights, Prime95 blend 24 hrs, LinX 20 passes @ 99% ram, and IBT @ 95% ram. All is stable. Amazing chips.

Thanks again guys.. I learned a lot.