GSKILL 2400 Trident 32Gb. Would like some suggestions.

[bluezero5]

I just bought this kit a week ago.
and had been playing around with it a lot.

I am on R4E, so it is a x79 chip set with 3930k

first finding, is that on Primary 10-12-12-31-2
I can make it into 10-11-12-31-1 without problems.
however, anymore tuning will result in a Prime95 failure with large FFT.

secondaries.
I manage to tighten to this:

4-220-10-6-20-4-6

I cannot get the DRAM cycle time to go under 220.. which is a drag on SuperPi32M. (won't boot if below)
the rest I can't complain really. pretty good.

On Teritary. I can safely go:
0-0-0-0-3-3-1-1-5-0 without issue.

DRAM clock best at 6.
disabling enhancing training helps slightly on reset time.

NOW.

SuperPi32M time now at 7m02s which is about the same as my Corsair 2133 9-12-11-27-1 kit. in fact, the Corsair 2133 kit can be overclocked to 2400 for a 6m50s SuperPi 32M time.. (though system not too stable.. passes P95 large FFT, but will give low voltage crashes.. so I reverted 2400 back to 2133.) The Corsair Dominator GT has DRAM cycle time at 108, and I feel that is the main difference there. the other timings are fairly similar.


Now..
I am consider chaning my 2400 Trident from 2400 to 2133 for tighter timings. Anyone think that can be done? I was thinking if I can get it to go 8 on CAS that will be pretty good.. but I would like to see has anyone tried that, and learn from the experience before going further.

insights appreciated.

 

[Woomack]

I haven't been overclocking memory on X79 but I doubt that you will make much tighter timings on this kit. It's mainly because you are running 4 channels and there have to be longer delays to make it stable.
2x4GB are not making much tighter timings than you can set @2400 so I would keep it on higher clock. You can try to play with timings but I think that higher clock will give more in this case. Also you can try to set it @2600 11-13-13.

 

[bluezero5]

yea, i have yet to try to use CPU strap at 125, and see how it does.
my 2133 were able to do 2400 passing most benchmarks...

I hope these 2400 can do 2666... but I have also heard about the IMC limits..
I guess there's only one way to find out...

 

[hokiealumnus]

Yea, you'll probably hit the limit of your IMC before 2600. Heck, you're lucky already; mine tops out below 2400 on the 3960X.

Perhaps leaving tRCD up would allow for more wiggle room on other timings? I'd try 10-12-11(or 10 if you're lucky)-28 and see if that helps. Cycle time is probably as low as it's going to get at 220 as Woomack mentioned.

Remember, you're working with a 32GB kit rather than a 16GB kit, which has the potential to make it a little more difficult. Not a LOT more difficult, as these clock surprisingly close to their 16GB brethren, but you can expect a little more resistance with 8GB sticks.

 

[Woomack]

My TridentX 2x4GB/Samsung are making 2600 stable and have really big problems with stability @2666 no matter what timings or voltage.

 

[bluezero5]

I can't get 2666 to post.. no matter what.. guess my limit is before that..
I will try to get to 2500 first, and see how it goes.
on x79, the bclk OC is limited, so I will need to use the CPU strap to try.. but that just makes it harder. heh.

DRAM cycle seems ok at 208.. but ANY below won't even Post...
if the system pass large FFT custom 1024 to 4096 FFT, I will call it stable.. let u guys know.

meanwhile..
on this set, I have tried experimenting with tRCD and tRP... in fact, findings as such..

CAS: OS boots on CAS 9, but fails SuperPi32M, exact not round.
tRCD: possible at 11. but I can't see performance improvement.. so keeping it at 12.
tRP: at 12.. and can't lower at all. dunno why.
tRAS: can boot from 27 - 31, will do super-pi just fine, but fails Prime95 large FFT.


tRRD: 4 ok.
DRAM cycle time: testin 208 now... but under 200 won't boot.
tWR: can do 8. but 8-10 sees no performance boost, so keeping at 10.
tRTP at 4. don't think need to go lower.
tFAW: tried 16, (tRRD x4), but crashes Prime95, 20 is min.
wWTR : 4 is fine.
Latency: using 6. haven't tried lower.... should I?

overall, the RAM set is very nice.
but between 2133-9clock and this set, improvements are 'very small'. Unlike when I overclock the 2133 kit to 2333.. it was a monsterous set there.. (but fails large FFT too.. so won't call it stable)

I need Gurus like woomack and Hokiealumnus to guide me a little here.

 

[hokiealumnus]

Stability tip - Run HyperPi with 12 threads, priority High. It will likely be faster than your Prime95 testing and will root out RAM instability pretty quickly.

 

[Woomack]

If you can't move with tRP lower then it looks like Samsung IC. For most samsung based kits 2400 10-10-12-26 1T 1.65V should be possible ( at least on Z77 boards ).
You already set 10-11-12-31 1T so it's about max and can only play some with dropping tRAS. DRAM cycle won't improve your results so much but it helps with stability so setting it to 250-300 can let you oc memory some more and overall give better effect ( that's maybe as I dont have X79 to test anything ).

I would leave secondary timings on auto as performance difference is minimal but it affects stability so also chance for higher clocks.
In 3rd table you already made really nice timings but set it back to auto to check max clock. When you make higher clock then back to timings.

RTL is thing that was helping on X58 but it's not working at all on P67/Z68/Z77 ( I mean that +/-1 is ok but if set higher or lower then all is crashing or can't boot ). Don't know how it's working on X79.

Play some with DRAM CLK Period. Something near 6 should be optimal.

No more ideas for now and I can't find any other useful options in this bios.

 

[bluezero5]

thanks both!

Woomack, I already tried DRAM CLK period, 6 is best indeed. 7 works, but not as good.
3 is surprisely ok, but also not as good as 6.

Let me try to set the rest on auto.. and try you 10-10-12-26-1T

I tried 10-11-12-27 before, it works, but wasn't stable enough, maybe cause I didn't keep things on auto.. I will go down that path and see how it goes... x79 wasn't as easy for me comparing to z77... so still learning....

and thanks Hokie on HyperPi, you might have saved me a few hours of test per stability!

 

[bluezero5]

alright, think I found a good sweet spot for the 2400 Trident.



however, needs 1.72V to be fully stable. (which I am fine with)

on my 3930k x79 @ 5.0Ghz.

superPi32m: 6m 50.121s
superPi1m: 7.59s
Hyper-Pi 32m x 12 = Pass.
Battlefield 3 for 3 hours = Pass
Prime95 large FFT for 4 hours = Pass

as such, I call it stable for daily use.

special thanks to Woomack and Hokie for their input. has sure made some decision making a ton easier.