OC Report - Cellshock DDR3 1866 BLUE D9JNL

[mike]

yeah, i hope it becomes common, clock skew adjustments really seem to improve things a lot, they are like the gtl for cpu tweaking
if tweaked right, you need way less voltages to get the same clocks stable, and you can reach higher clocks with the same voltages

camouflage, i chatted with him about that and he tried different tras but 14 seemed to be the fastest and lowest tras that was still stable iirc. thats the same i noticed too, hence the spec of 777-14

Re: Clocks skew ... we definetly need to look more into that. I tried comparing with a second Maximus today if the skew would be identical - but that board did not like me anymore and wasn't playing with me . I just ran across something in the latest Everest earlier - it might be able to see the skews?? Not sure - was in the middle of something else and only just remebered now in retrospect.
Sometimes I am a bit confused what is going on when I switch from auto to user selected clock skew and vice versa - it seems sometimes I get varying results ( or the move back to auto keeps the previous userselected value). Would be great to have an indication what auto does.
I had always thought that clock skews were only helpful to match 4 sticks, but I am seeing improvements. New reserach project...


RE: Tras - correct - I had added in the notes in first page that tras 12-21 was running strong in terms of max frequencies, but tras 12 took a big negative hit in SuperPi 32m times.
tras 14 was the fastest yet tras 21 was almost as fast as tras 14 for me in 32m ( within a second of each other), while Tras 20 again was a bit slower than tras 21.

Tras 21 gave me between 4-16mhz more frequencies. At 2.0V I went down from 1940mhz to 1936mhz with Tras 14; and 2.08V I went down to 1988mhz ( yet I also had changed other settings + CPU - so I am not 100% sure of the actual hit yet).
A benefit from being able to lower tras is possibly the abilty to tighten some of the other secondary timings.

[camouflage]

Hm, never played with clock skew so far, no idea what to set - just left it at AUTO.

Seems to me it'd be a great idea if each manufactor supports his sticks with a personal BIOS tutorial for different boards..........

[mike]

Hm, never played with clock skew so far, no idea what to set - just left it at AUTO.

Seems to me it'd be a great idea if each manufactor supports his sticks with a personal BIOS tutorial for different boards..........

I am quoting FCG from ANandtech:

Ai Clock Skew for Channel A/B: This setting allows you to either advance or delay the signal timings for Channel A or Channel B DIMMs. Because it's not possible to locate all memory modules an equal distance from the MCH, it's important to be able to adjust signal timing to ensure all data (or requests for data) arrive at the same point at precisely the same time. When experimenting with this setting either slightly delay Channel A (since it's closer to the MCH) or advance Channel B - do not change both simultaneously unless you find you need more than 350ps (picoseconds) of total skew (highly unlikely).

I have seen changes both good and negative with total skew as well - not to disagree with the above logic, but trial and error can sometimes give interesting results.


BTW you mentioned you prefer Maxiums bios 904 - in what way is it better then 905? havn't tried it yet, but am about to...

[camouflage]

Mike - 0905 is second best for me, but 0904 runs better. All other releases are crappy somehow, one (don't know which anymore) even killed my RAID so that all data were lost........

[saaya]

mike, if its really the same as read/write levelling then there are 3 options according to the intel white papers:

1. chipset training
2. module specified
3. chipset specified

1. the chipset does some quick test and adjusts the skew settings automatically
2. the module can specify a certain clock skew (dont think that would make any sense)
3. the board can specify a certain clock skew (dont think this makes that much sense either)

according to intel they use and recommend option 1, but from what ive seen that doesnt seem to be the case, at least not with the asus p5k3 dlx boards.
or maybe its a mix of 3 and 1. from my experience id say there is a bios preset, and the chipset does a quick test if this preset works, if it does then it keeps that clock skew adjustment.

that doesnt mean the ideal clock skew adjustment is configured though, the current bios preset might work, but a different adjustment might work way better, and i think thats what we are seeing here. it seems that if you set things back to auto it keeps the last used skew adjustment, does the usual quick test and, it works, so it keeps using that value and doesnt readjust things.

thats why once you changed clock skew and set it back to auto, you might still run the same value you last used.

well, pure speculation here... but thats the gut feeling i have from my experience with clock skew on the asus blitz board and from what you just told me.

so on the premium you can adjust address and data skew independantly?
thats really interesting! thats basically like compensating the trace length anomallies of the board your using, plus compensating possible irregularities in the memory controller or of the memory chips interface... really neat options!

i wouldnt have thought this can make that much of a difference...
im starting to think pcb layout has a pretty big impact on how high you can clock your memory if adjusting this gives such big gains.
either that, or the memory controller of the chipset has high irregularities that prevent high memory clocks... the latter actually matches my current experience with intel ddr3 chipsets...

oh and the only option everest reports that comes close to this is phase adjust, but i dont think its trd phase adjust, i think tamas means the clock skew phase with that option. ill check with tamas...

btw, i really love the testing your doing man, really good stuff your finding out!!!

[BadNizze]

This is a bit out of my leauge that´s for sure. But I will try anyway, Should I leave CMD Skew @ Auto on my P5E3 Premium and focus on CLK Skew? Or do I need to "compensate" by changing CMD Skew when I play with CLK skew? I will try to find a Borderline stable voltage for 8-7-6-21 @ 900mhz and then play with CLK skew to see if I can get i stable, does that sound like a good way to test?

Edit: Will try with CMD\CLK skew @ Normal to find a baseline voltage for my STT @ 8-7-6-21 900mhz PL 7.

[saaya]

wait for mikes reply, hes really the clock skew tweaker here
unfortunately i didnt have much time to play with clock skews so far, the boards im using unfortunately dont have that option in bios, and while i DO know its possible to adjust it in windows using a register tool, i dont know which registers to change, and what to

hmmmm hey mike, could you make some dumps with bareedit from felix88?
dont change anything except clock skew, then we can see what registers get changed! 1. felix88 might add an option to memset, 2. people can tweak clock skew live in windows which is faster than rebooting all the time, 3. people that dont have a clock skew bios option can still tweak clock skew

i went through my notes about D9JNL testing and figured i could post some here

1.9v single channel P5K3 333 strap
memtest test no5
increased each timing in steps of 1 clock until i saw an improvement
then increased clockspeed in 25mhz steps to the max stable speed
timing combinations that arent mentioned didnt bring any boosts in mhz:

554-15 1275
555-15 1400 trp = 125mhz boost
565-18 1425 trcd = 25mhz boost
665-18 1575 cas = 175mhz boost
666-18 1675 trp = 100mhz boost
676-18 1700 trcd = 25mhz boost
776-18 1850 cas = 125mhz boost
776-21 1875 tras = 25mhz boost
777-21 1925 trp = 50mhz boost
888-24 1975 cas+trcd+trp+tras = 50mhz boost
888-27 2025 tras = 50mhz boost

note the order or limiting timings:
trp
trcd
cas
trp
trcd
cas
etc

until you reach high speeds, then tras plays a more important role all the sudden.
and note that going from cas7 to 8 doesnt bring the expected boost, at least with my board and configuration...

[saaya]

just chatted with SIOUX, i sent him a 1866 kit as well and he also managed to reach 2000 777
and also on an x48 board
im starting to think handpicked intel chipsets will clock the same if not better than handpicked 790 chipsets.

so it seems its not a freak kit you got there mike
i still dont think all kits can do this, but from 4kits i saw results of 2 managed to get 2000 777 with tweaking and vdimm above 2v

[camouflage]

I need 1084 8-8-8-x to run this here with 1:2 divider:



Hope to receive a fine kit on Friday.........

[metro.cl]

i had to bring my PC to the office to start testing the RAM, nice to know about skews will have look into it.

I first tested CAS8 with bios 151 that came with the board, then flashed couse there was a newwer bios 204, and i didnt like it so flashed back to 151, but seems like CAS7 is what was limiting me. Will start testing after lunch

[metro.cl]

at 2.0V 2000mhz was for sure a no go as well for me.

Actually on the P5E3 Prem. I had needed DRAM CMD SKEW +350PS and DRAM CLK Skew of +200ps to achieve max results ( was worth about +50mhz).

I noticed that Skews differ from board to board - something that might be worth looking into and could give you a little boost.

Hmm interesting, did you apply that value to both channels?

[mike]

Hmm interesting, did you apply that value to both channels?

Argh - you caught me on the wrong foot here - my own fault - yes I had it on both channels - which does not make much sense - yes even to myself.

I only played for a few hours on the X48 with clock skews - but from my experience now with the Asus Maximus X38 - it appears that clock skew only have a "big" impact when running max voltage + clocks.
Today at 1.98V @ 1900mhz clocks skews hardly had any impact, as opposed to when I was running 2.08 1975-2010mhz.

I just received another X48 today - I'll be able to give better answers in the next couple of days

[|SiLA|]

i'll be having my 2 kits next week

hopefully they'll clock well :d

[Oliver]

Good to see some more results...

A kit worth playing with...

Looking forward to further results...

[camouflage]

Received mine one hour ago.............

[metro.cl]

Argh - you caught me on the wrong foot here - my own fault - yes I had it on both channels - which does not make much sense - yes even to myself.

I only played for a few hours on the X48 with clock skews - but from my experience now with the Asus Maximus X38 - it appears that clock skew only have a "big" impact when running max voltage + clocks.
Today at 1.98V @ 1900mhz clocks skews hardly had any impact, as opposed to when I was running 2.08 1975-2010mhz.

I just received another X48 today - I'll be able to give better answers in the next couple of days

Samething over here, skews help but in weird ways, i left channel A at normal and then checked delays for channel B, and 250ps got it stable. But then i found another setting that helped me to get it stable without the need to yse delays

Will update SS soon but 2000MHz 7-7-7-21 2.1v is OK for SP32M. There is a new BIOS for me in Asus FTP and my E8500 should arrive today

[saaya]

so you got 2ghz cas7 too?
stable?

what real vdimm are you using?
did you meassure it?

[metro.cl]

I dont know where to measure the vdimm but set in BIOS and boot from BIOS y got this.



Didnt actually tweak much, only changed 3 subtimmings and left everything else at auto, the thing that brought stability was All precharge to activate set to 10, it was at 8 before.

[begrip]

omg nice rams

[Gautam]

Heh, all this talk about clock skew is giving me an idea for that Foxconn article contest.

[BadNizze]

just chatted with SIOUX, i sent him a 1866 kit as well and he also managed to reach 2000 777
and also on an x48 board
im starting to think handpicked intel chipsets will clock the same if not better than handpicked 790 chipsets.

so it seems its not a freak kit you got there mike
i still dont think all kits can do this, but from 4kits i saw results of 2 managed to get 2000 777 with tweaking and vdimm above 2v

I spoke to friend over at a swedish HW forum, Ho also got this ram and his kit does 777 @ 2000 1,9v semi stable. I´l ask for screens of stability.

[saaya]

I spoke to friend over at a swedish HW forum, Ho also got this ram and his kit does 777 @ 2000 1,9v semi stable. I´l ask for screens of stability.

1866 cellshock kit or some other brand? heh
1.9v real? that would be sweet! does it scale with more vdimm tho?
i can tell you guys, some d9jnl chips dont scale with vdimm above 1.7v!
but there are some that reach the same speeds with 1.7v that other chips need 1.9v for... so go figure

its better to have the same speed at low vdimm than at high
then again, the best chips are the ones that do scale with vdimm, cause no matter how good a chip is, if it cant take 2v then it wont be able to run really REALLY high speeds... i figured what you guys want is max speed, and not good speed at low vdimm, so im looking for the chips that scale best with vdimm and reach the best speeds, and the top 20% or so of the chips then end up in 1866 kits

i think i figured out why i dont get as high with the kits as you guys.
1. im using relatively old boards that surely dont clock as well as the newer boards out, all the really nice scores are on 790 and x48 boards.
2. im not tweaking vdimm to the max spot where it still scales, increasing vdimm to that sweet spot usually gives a boost of 25-75mhz
3. you guys are tweaking clock skew and subtimings, and even if you dont, i have feeling the bioses are better tweaked on the new boards regarding to clock skew adjustments.

oh and btw!
about clock skew... well im not sure if it was clock skew, but i think so.
listen to this: i was curious how high i could push the 1866 kits im working on right now (received a nice batch of chips! this batch is going to be even better than the first batch of 1866! )

anyways, i put in one stick that didnt do THAT well, and checked how high i could push it, and then whats limiting it, etc...

then threw in a better stick and did the same.
now guess what, i accidently tried the previous stick and...
it clocked a good 25mhz better...
i compared different voltages, and it was def clocking better than before...

i kept pushing it, and you know how when you push things too high the board wont boot properly and the monitor stays black, all fans spin on max, and if you reboot again the board loads defaults to let you enter bios, right?
well, after that happened, the mem performed exactly like it originally did...

i checked all the time, i was 100% running the exact same settings and timings, im using the bios profiles you can safe and load... so this sounds to me like the chipset somehow adjusts some timings or some setting by itself...
could it be that it does adjust clock skew? and with the better stick it adjusted clockskew to a different level, so then the worse stick performed better with that adjusted clockskew value?

and when the board reset to defaults, the clockskew value was reset even though i loaded the bios profile?

ill try to test this again during the weekend and will keep you posted

[BadNizze]

Cellchock JNL, I´l ask if it scales with voltage later need to get to work now...

[mike]

I dont know where to measure the vdimm but set in BIOS and boot from BIOS y got this.

.

You can measure vdimm on the fet circled here - sorry for the clear pic


Best CLock Skew on this board for me for Cas 8 2T- Advance 100ps on both channel A+B

Saaya said:
load... so this sounds to me like the chipset somehow adjusts some timings or some setting by itself...
could it be that it does adjust clock skew? and with the better stick it adjusted clockskew to a different level, so then the worse stick performed better with that adjusted clockskew value?

and when the board reset to defaults, the clockskew value was reset even though i loaded the bios profile?

This weird behaviour with motherboards ( Intel X38s) has been driving me nuts for months now - killz me that you constanly have to find old settings again, over and over.
Might be clockskew - not sure myself - maybe other chipset related timings. On the other hand - it actually might be clockskew - good chance it is since they clocked better and then went back to slow.

[mike]

Update on new board - Asus P5E3 Premium:

Compared to the Asus Maximus Extreme I lost about 20mhz at equal voltages ( 2.07V actual )with this board. Max 32m I can pass now is at 1980mhz. With my previous P5E3 Prem. I was clocking the Cellshocks also at 1998mhz - just like the Maximus. BUT I am running a different bios on the New X48 board - might be the reason for the loss - but doubt it - I will verify this later. Dual 32ms on the other hand appear to be better on this particular board.

Great news is that with particular board I can push Vdimm a bit more - now I can push 2.13V actual in these Cellshocks and still scale for Cas7 - unlike with the Maximus and the P5E3 I had before Cas7 2020mhz 32m is closing in!

Other good news is that I can run Cas 8/8/8/16 2T ( 2.06V ) without being limited by the motherboard - well now I think I am limited by the CPU




Unfortunately again I HAVE to use high NB voltages - I am @ 1.85V - I wonder how long this board will last...and I have a feeling it needs a bit more.
Overall the board allows for tighter secondary timings - a nice bonus

Settings For the P5E3 Premium for the above 1m ( timings not optimized):