***Asus Rampage III Extreme Owners Thread***

[skuldarin]

well i'm sure my system just doesn't like going over 1866 on mem. i think it's specifically mem related as my dominators are of those non hyper chips (can't remember off the top of my head, they are the one's that don't like going with even numbers on the cl's, it's gotta be 8-9-8 or 7-9-7 and so on.) no big deal really, not much of a different between 1866 and 2000. i've increased rTL's as far at 74, 75 and 76 with +5 down the board and still can't get through full windows boot without crashing so... ohwell.

i'm thinking of trying 1800 and going for 6-8-6-24-72 2T though or 8-9-8-24-72 1T... either way i'ma go for the faster latency / response.

[CryptiK]

Hey don,

My VTT is 1.41 at the moment, i'll try increasing it and dropping my Vcore later on ... will let u know

Do try playing with PLL too. For an easier point of comparison and so we can get an idea of scaling, what does you CPU need for 4GHz LinX/Prime small FFT etc and 4.2 or 4.3GHz LinX/Prime small FFT?

Hi guys!

I just got a 990x last week and trying hard to find a 24/7 OC for it.
What do you think is the max vcore for 24/7, if i only use it for gaming? Im watercooled with the Swiftech h20-320 kit, and NB waterblock also from swiftech. Currently im stuck at 4.5ghz, 1.39 vcore in bios, in cpu-z it reads 1.376 in idle and from 1.343 to 1.349 in load under prime95, cpu pll at 1.31v, all that with LLC at 50%. I have the rampage III extreme board, of course. I would like to use it in the primary rig for at least 3 years, so I would like to find a sweetspot between safe vcore and performance, something that can give it a run for the money but not devastate it (i know will probably sound stupid to you guys, sorry). Batch nuber of cpu is 3045A709 . Im a little dissapointed with this CPU, i thought according to info on the web prior to the release, I could have a higher 24/7.

Thanks guys, please give me some advice!

This is like asking how long is a piece of string. No one's going to be able to tell you with any semblance of accuracy how much these things can take 24/7 for the arbitrary time period of 3 years without degradation.

Around 1.36v would be my personal limit, but I'm conservative with 24/7 PC's. When I bench I do it on bench only gear and on DICE and LN2 so my work/gaming PC has to be stable and I go for a balance between performance, heat production and reliability. I'm running my 990X daily at 4GHz @ 1.184v / 3600MHz uncore @ 1.244v / DDR3-1600 6-6-6-18 @ 1.55v because with that configuration I'm pulling ~78-79GFLOPS in LinX and ~17GB/s bandwidth and all at low voltages. FPS are still great with my card in the games I play and encoding/rendering/batch photo conversion is faster than my bloomfield was at 4.33GHz.

Since I can't seem to increase my core clock or bclk at all past 200, I decided to tune Uncore freq.
4.2GHz @ 82.9GFlops:

x18 up from x15! I had to increase vQPI from 1.3375 to 1.3500.
It creates a nice look with QPI and Uncore synchronized.

I think I might not have enough cooling to go faster, since I top out at 84C with Linx at 4.2GHz. Although, I feel like I have a lot more performance at 4.2 than many have at 4.5! I see many 4.5GHz club members with less GFlops.
Right now I have 3x 120mm rads. Two GTX 120's and one GTS 120. One GTX has push/pull. One GTX can only have push because of its location. The GTS only has pull on it now, but I was thinking I should add another fan for push/pull there. It will take some very careful alignment and I have to swap my barbs back to straight from the nice rotary 45's:
http://s144.photobucket.com/albums/r...t=2e6b80b7.jpg
http://i144.photobucket.com/albums/r...h_2e6b80b7.jpg
http://s144.photobucket.com/albums/r...t=85256708.jpg
http://i144.photobucket.com/albums/r...h_85256708.jpg

Nice, I run ~3600MHz too. I see a lot of GT users leave this at <3000MHz and I wondered why since it has a large impact on performance. It's the speed of the L3 cache as well as IMC and so is critical in tuning overall performance.

Thanks! I was thinking that since 4.5GHz is kinda of a big feat I should loosen ram timings to 8-8-8-24. I saw some other users running up to 1720 at CL6 on the same kit I have so I just went for it. It was a combination of strong memory, cpu and imc, and mobo.

I love memory overclocking so I'm very tempted to tighten TRFC and make some other tweaks. But for now I want this feeling to last lol

Yeah 4.5GHz was awesome to hit for the first time. I also rather enjoy ram overclocking and getting it tuned for good bandwidth and low latency can have a great impact on performance and how the PC feels to use. These kits will do 1700-1720 CL6 on 1.65v with tRFC ~60 (maybe have to use 72 depending on your kit) and tight subs, so have a go when you get the chance. I will play with mine a bit now I have a different CPU and board to learn.

haha i know i mistyped there yes, 4v, i use compressed super LN2 and turned my 980 into a super conductor and power it with a nuclear reactor.



i already loaded 0003 actually. at the moment i've only been trying to get 4.5 stable again with it, i haven't done much mem tweaking yet. but so far i lowered my rTL's to 60, 61 and 62 down from 58, 58 and 60 (stock) but that's about it.

Ah they're your RTL's, tRL is different In BIOS 0003 just below the RTL values are the tRL values.

well i'm sure my system just doesn't like going over 1866 on mem. i think it's specifically mem related as my dominators are of those non hyper chips (can't remember off the top of my head, they are the one's that don't like going with even numbers on the cl's, it's gotta be 8-9-8 or 7-9-7 and so on.) no big deal really, not much of a different between 1866 and 2000. i've increased rTL's as far at 74, 75 and 76 with +5 down the board and still can't get through full windows boot without crashing so... ohwell.

i'm thinking of trying 1800 and going for 6-8-6-24-72 2T though or 8-9-8-24-72 1T... either way i'ma go for the faster latency / response.

Elpida BBSE IC's. I have a GSkill kit with them, they are still very good. Some Hyper clocks better with staggered timings too.

RTL = round trip latency

tRL = read latency

Raju noted that this board can set tRL too tight sometimes, which is probably why we now have BIOS 0003 with that setting so we can offset it. tRL has a large bearing on dropping sticks when overclocking.

[skuldarin]

Ah they're your RTL's, tRL is different In BIOS 0003 just below the RTL values are the tRL values.

Elpida BBSE IC's. I have a GSkill kit with them, they are still very good. Some Hyper clocks better with staggered timings too.

RTL = round trip latency

tRL = read latency

Raju noted that this board can set tRL too tight sometimes, which is probably why we now have BIOS 0003 with that setting so we can offset it. tRL has a large bearing on dropping sticks when overclocking.

the slightly lower speed vs low latency is really showing it's worth for sure though. just at 512mb (light stability testing) on LinX, i went from about 72-74gFlops to 80-82gFlops going from 1866 @ 7-9-7-24-72 2T to 1734 @ 7-9-7-24-72 1T. i still want to get it to 1800 but it's helping alot.

yeah see i've been adjusting rtl's and i'm at 68, 69, 70 with +3 on all 3 channels. i'll lower some of the read times aswell, i think last i saw they were something like 5-15-10 or something like that.

[Toritechie]

the slightly lower speed vs low latency is really showing it's worth for sure though. just at 512mb (light stability testing) on LinX, i went from about 72-74gFlops to 80-82gFlops going from 1866 @ 7-9-7-24-72 2T to 1734 @ 7-9-7-24-72 1T. i still want to get it to 1800 but it's helping alot.

yeah see i've been adjusting rtl's and i'm at 68, 69, 70 with +3 on all 3 channels. i'll lower some of the read times aswell, i think last i saw they were something like 5-15-10 or something like that.

Sounds really good; mind if I ask what multi and uncore you're using? I've just hit 4.5Ghz stable; but an inordinate amount of voltage was required and my Gflops are less than some of my previous runs, since I am forced to use 2t or higher for my memory.

[skuldarin]

Sounds really good; mind if I ask what multi and uncore you're using? I've just hit 4.5Ghz stable; but an inordinate amount of voltage was required and my Gflops are less than some of my previous runs, since I am forced to use 2t or higher for my memory.

i've been keeping my uncore at around 3200-3300 range for 1800-1866. the main thing i've found, lower base clock and higher multi helps. the higher the base clock, the higher the amperage pull on the cpu from what i see through ROG (though ROG isn't the best to use, it does give a good idea of things).

also, i've notice a bit more usefulness of my PLL being around 1.25 than the 1.629 i had it at before.

i've tested 115x39, 133x34 (my current), 155x29 and 187x24. all of which kept at about 3200-3300 uncore and kept as close as possible to 6.4gbits for QPI. also know that every one of these wwill show 4.480-4.5 in bios but will be AT LEAST 4.5 to 4.55 after boot up due to small variation in your ration when the bootup takes control of things.

QPI of about 1.35 to 1.37v seems peachy, vcore around 1.4125-1.425 so far.

EDIT / UPDATE: testing 1866 @ 7-9-7-24-72 1T RTL 57-59-61 rTL +2 @ 1.749v. all 12 gigs registering, LinX running full board. Uncore 3200 (3204), QPI @ 6.4gbit w/ 1.3625v, 133x34 @ 4.539ghz w/ 1.41875 vcore. currently with 1.4v PLL.

[spoolindsm127]

Yeah 4.5GHz was awesome to hit for the first time. I also rather enjoy ram overclocking and getting it tuned for good bandwidth and low latency can have a great impact on performance and how the PC feels to use. These kits will do 1700-1720 CL6 on 1.65v with tRFC ~60 (maybe have to use 72 depending on your kit) and tight subs, so have a go when you get the chance. I will play with mine a bit now I have a different CPU and board to learn.

I think I got pretty luck with my kit since it has no problem doing 2005MHz at 7-7-6-20-1T tRFC 60 with 1.65v. I'm thinking it shouldn't be a problem at a lower frequency. I'm going to work on memory timings next and possibly increasing the Uncore a bit.

And I just wanted to throw my $0.02 in about RTL values. I've been using Raja's rtl calc with 'Min Sep Delay ps' value set to 0.58 instead of the original 0.67. It seems my board runs tighter RTLs than what the original value of 0.67 calculates. At least for my setup it seems the best or most stable settings for RTLs are in a +2 and +1 pattern in relation to Ch.A. So for ex. my last run at 1716MHz RTLs were set at 48,50,51. Not the +2 +2 pattern I see people using. Only sometimes when the rtl calc calculates an RTL like 47.9, the board seems to want to still run RTL 47 on Ch.A and not 48. In this case a +3 +2 pattern for Ch. B+C works well, like 47,50,52

From my experience, RTLs have a BIG part in overclocking and making a successful LinX run. RTLs being set too tight can also cause dropped mem channels.

[skuldarin]

i actually had to do it +1 per channel from one another. when i do 2, IE 58-60-62, i BSOD if i test more than 2gb of mem in any burn test. but if i put to +1, IE 58-59-60, it's perfectly fine and actually stops on error instead of straight up crashing.

[spoolindsm127]

Awesome kit indeed, DDR3-2005 7-7-6-20 1T @ 1.65v

Thanks chief . Believe it or not I made 100 passes in LinX using 5120MB memory at those timings

i actually had to do it +1 per channel from one another. when i do 2, IE 58-60-62, i BSOD if i test more than 2gb of mem in any burn test. but if i put to +1, IE 58-59-60, it's perfectly fine and actually stops on error instead of straight up crashing.

Yea it can either crash from RTL being set too tight, or too loose also. Did you give +2/+1 ie. 58, 60, 61 a try?

[skuldarin]

i have not yet. if i start getting instability again when testing higher amounts of mem then i'll give it a go. right now i'm getting the best stability with +1/+1 at 66-67-68 with +3 rTL's

EDIT: on another note, i've come to find from multiple sources that the reason why gulfy chips have such out of control temps is due to intel using rather trashy sensors in their 32nm chips in general.

i was scared about some of the temps i was seeing, but they were always on the same cores. first i tried relating the cores to the physical position on the CPU but 1 out of the 2 cores that read REALLY hot is on the outside and the other near the dead center of the chip. because of this, i've learned to have a better trust in HWmonitor's and PC Probe's temps, among others aswell, over RealTemp GT as RealTemp GT is only giving what it sees from these trashy sensors. because of this i've been doing 1 of 2 things, 1 - the method earlier stated of looking at the highest temp and taking 80% of that as your average of a sort and 2 - adding the temp of all cores in RealTemp GT and getting an average between all 6 cores. either way i go about it, temp ranges from 82-88*c at full load at 4.5ghz. this temp range is also including the hours of stress testing which heat soaks my water system as it's a full single loop between cpu, full MOBO and a dual chiped VGA card. because of this i do need to periodically let the soak cool out over about 10 mins to get back to normal levels otherwise i start getting that higher 88*c average after a good few hours of none stop burn tests.

i really need to isolate my cpu with a second loop >.<

[Cookiesowns]

Hmm, COME on asus! evga even released the microcode update already, how about you??

http://www.evga.com/forums/tm.aspx?&m=932958

[tomovk]

Hi guys!

I was working on my 990x OC, and i got to the point where i am stable on 4.5ghz, 1.375v in bios, 1.363v in idle (cpu-z), and 1.330v in load (cpu-z), but im only stable in prime95 small fft. when i do blend, it is reporting errors, never BSODs. I flashed to bios 0003, and i can modify tRL values now, but i dont know where to start. whatever i try, i gate errors. i have g skill trident 2000mhz 9-9-9-24 kit, and i run it at 1804 with uncore of 3608. PLEASE HELP. I tried raising qpi/dram voltage and ram voltage, no use. temps dont get above 72c.

[tomovk]

What do you think about using loadline calibration on this board? If i need 1.4v for 4.6ghz, is it better for the cpu if i set 50% LLC, and the idle voltage is higher, dropping on load, or set 100% LLC, and have lower voltage in idle, so it climbs to 1.39v on load?