I haven't tried to run them seperately yet - but actually I will look into it. Now that you asked I got curious myself.
I saw your posting about the P5E3 Prem - you ment the prem - right? That's the board i killed.....lol remeber me being fearless with the Maximus Extreme - paid off - in a bad way.... P5E3- Prem - awesome board!!! - 'specially with those sticks!
Originally Posted by saaya
I ment to post about how to get there - it actually did take quiet a bit of tweaking - 1950mhz Cas7 from the beginning - but after that I was playing with clock skews and really had to get the board in sync. - it certainly takes a bit of getting you mobo to run all minty in all respects - it was surpirsing to see how many voltages had to be just right for everything to be in harmony - something I will certainly add to the initial post. Now that I realized that 7/7/7/14 is actually a tini tiny bit faster ( about 1sec in 32m) I am just relearning the sticks all new again and have to adjust ( can adjust) secondary timings as well it seems.
Glad you staying clear of the blue kits so far - your Vantage scores are getting too high
just kidding - keep up the benching man .
AFAIK Saaya didn't think this kit was anything special - and they were not handpicked according to him - he was suprised himself how good they actually turned out to be. BTW I saw Saaya once binning ram during Cebit in Hannover - almost as impatient as I am
As Saaya actually asked me in an earlier post - on the 790i chipset compared to the X48- you have less memory tweaking options than on the X38 and especially on the X48 boards, and I was intitally not able to get above 1940-1950mhz Cas7 out of my set - it really takes a board that is in "full harmony" to get the full potential out of your ram. I not saying it couldn't be done - I just didn't have the patience on my crap 790i board to try any further, since my OS got corrupted every 20mins and gave up. I know you were/are running 790i - so maybe it's the board that was/is holding you back after all.
1.75V seems like a good VMch - Kunnak has actually achieved crazy good results with such "low" Vmch - I had previously wrongly thought that it was necessary to crank Vmch to the max in order to get better performance levels and more stability, but actually it seems I was wrong.
Giorgioprimo had mentioned that performance levels actually only need vdimm a while back - but it took me killing some boards to realize and see the truth in his wisdom. I had been running my vmch WAAAAY too hot - which also restricted my max mem clocks. 1.79/ 1.83V was perfect for my max runs with a high CPU multi - when I went to 1.87V and more I would crash in 32m runs usually after 5-6mins without realizing what the culprit was
Last edited by mike; 04-20-2008 at 12:46 AM.
In a terminator way "(MCH) come with me...if you want to live"
Sascha told me on MSN about your... extremely extreme NB voltages m8
Well...you've gone a bit too far, sorry for your boards.
On the P5K3 Deluxe's I've been using 1.7V ( the BIOS's max ) for some time.
After modding my best P5K3 Deluxe my max stable freq with the CellShock kit went from 2056 to 2088.
And I could bench ( 1M, Hexus, etc ) higher than before.
Performance level 6 was running fine, where it would instantly crash when set before.
Just make sure you keep the chipset under 55C loaded, and don't go that high.
1.8V was good enough for the P35s, and 1.85V for X38/X48.
Terminator - I love it - don't tell anyone how many I terminated - lol
Gautam had told me a while back that higher Vmch is needed - so it his fault!
(well he also told me that I needed to watercool the NB ....details details...)
It's also very true - if the NB stays under 58C( or 55C) - I now can go up to 1.93 without seeing as much instabilty!
Perfect quote here Kunaak - takes some of us ( me) longer to figure out than others - lloll
Last edited by mike; 04-19-2008 at 11:23 PM.
this isnt a cherry picked sample...
i sent out cherry picked samples to some benchers, but those were the 1800 blue kits and all of them clearly said its not a retail kit.
im the last one to send out handpicked review samples, didnt you see my rant about nvidia sending out cherry picked 790 boards with cherry picked crucial memory to the press?
---------------------------------------------
about cherry picked samples:
the chips change cause micron keeps tweaking their mfg tech and the design, and they dont tell us about it...
so if your pushing memory to the limit naturally there will be differences between the chps/sticks/kits.
just like with the D9GTR i adjust the speed we bin the memory at every now and then, sometimes up, sometimes down, to make sure the high end kits really use the best chips. For the 1866 kits we always use the top 20% of the chips.
The first batch of the 1866 kits were slightly better, for the second batch the chips were slightly worse, about 25mhz.
Now a problem with some skits clocking better than others is... tricky...
I dont have time to test several kits and find out where they max out, then find out what the rough average oc the kits is, and then pick some kits that represent the average kits ideally, reaching the typical average oc.
Ive actually done this cause it pssed me off that i sent out samples and some of them DO oc better than the average kit which hurts my credibility.
But its really too much work man... and then the next batch of memory will oc different again...
so lets say the average oc is 200mhz, then i send out a review kit that ocs 200mhz, and then the next batch ocs 150mhz on average...
see my dilemma?
and then there are kits that clock MUCH better on one board then on another...
i had a kit here that clocked really really well, tried it on another of the very same board, and it clocked 50mhz worse...
i guess it comes down to trust. i dont send out cherry picked review samples UNLESS i say so, and the few times i did i told those guys to clearly say its a special sample, like for victor wang, and they did.
Sending out handpicked review samples only makes sense if you plan to get attention of the average user, who doesnt know sht and trusts the press 100%.
And even then i dont think its worth it... usually new products are great, so if you send out cherry picked ones, and people then find out the average stuff is worse, you make your product look worse than it actually is!
people will be dissapointed even though their average ,whatever product we are talking about, is actually great, its just worse than what they expected.
So in that regard it would actually make sense to send out cherry picked samples that clock BELOW average
sorry for the long text, but i wanted to make clear where i stand on cherry picked samples
---------------------------------------------
bill, how much did you gain in stability mhz wise? with extra vmch?
mike, yeah hehehe, giorgioprimo showed up at cebit with a box of memory and said "saaya, my friend... can you please bin them for me and find the best sticks for me?"
yeah well, with the older p35/x38 chipsets you really DID need that high vmch, but it seems with the new revisions 1.7v is already enough to reach the max clocks.
about needing more vdimm to reach tight PL... i dont know... thats not the experience i made... when i used low vdimm slow mem speed and low vmch i cant get tight PL, if i increase vmch i CAN get tight PL...
hmmm so with watercooled nb higher vmch (1.9v) DID help after all?
btw, vahid from techpowerup got 1900 777 without any tweaking:
http://www.techpowerup.com/reviews/C...CL8_2GB/3.html
how much were your max clocks before tweaking anything?
Last edited by saaya; 04-20-2008 at 02:13 AM.
I can vouch for saaya anytime.
He's an honest person.
Whenever he sends someone a cherrypicked sample, he does that for a reason ( Benching ? ) and he DEMANDS the receiver to say that it's a cherrypicked sample and not a retail kit.
And of course, I've never "seen" him sending a cherrypicked sample for reviews.
I have more than a reason to be sure about that
Like I said before, from DDR3-2056 8-7-6-21 1T I went to DDR3-2088 with the exact same timings.Originally Posted by saaya
I've also gained about 30MHz over my previous SuperPi 32M stable frequency/timings-set at 6-5-5-16.
The extra voltage also allowed me to run 1 performance level lower ( better ) than before at the same FSB/DRAM Divider.
On a friend's P5K3 Deluxe that couldn't do DDR3-2000 stable with my samples no matter what, with 0.05V over the stock ( unmodded board ) voltage it managed to run DDR3-2000 stable at the same settings as my good P5K3 Deluxe.
saaya i still want my hp hahahahaha j/k
i hope to get soon those 1866 kit..weird i don't see them on cell's site lol
i should have it in 10 days tho' hopefully they clocks well like the 1800 i have
but are they still black or blue kit?
Good post saaya
Thanks for clearing things up for me, I guess there were some misunderstanding on my part, sorry for that.
Part of the fun from purchasing and benching hardware is that there is a chance for you to receive a "special" part that OC like crazy, or required very low voltage to run at certain speeds. Kinda like playing lottery.
I would hate to think that these "special" kits are already filtered out, and all you'll receive is kits that ran at rated speeds. Little chance >> no chance!
I guess I'll be getting a pair of these blue sticks to try out after all.
the 1866 are blue, all of them
e_f, well there is still a difference from one 1866 kit to another... its not like if you bin the mem you end up with a certain amount of kits that all clock the same...
we are talking about the highest bin here, there is no from x to y range for the memory like with lower bins.
depending on the yields of the chips you usually have a strong typical yield curve, and if you take the centre, what most of the mem can do, then yes, all kits will roughly clock the same. but if you cut off the top, then you will have a bigger difference from one kit to the other, than when comparing different kits in the center.
know what i mean?
i mean i even posted here and said "better wait for more results" because i dont want people to think all kits clock like this, cause im not sure they will.
why would i send out a handpicked kit and then tell people dont take theese results serious
if it helps, from my experience and the feedback i got regarding the 1866 kits, they do 1850+ 777 and 2000+ 888
especially depending on what board you run it on. if you get 3 boards each will oc it differently, even if they are all the same. and it doesnt even mean one of them is better than the other, cause with a different kit suddenly another board clocks higher... its the same with ddr2, but with ddr3 the difference from one board to the other seems to be bigger.
ive seen differences as high as 100mhz max stable clocks from one board to the other... but im starting to think it can be leveled out with clock skew adjust.
its too bad most boards dont have that bios option
Last edited by saaya; 04-20-2008 at 05:44 AM.
oh finally for the first time i'll have blue kits
I just ordered a pair from memory Configurator.com I just hope they will play nice whith my Evga 790i
QX9650orE8500 Cellshock blue 1866mhz 2*1gb zeus 850W Asus 9800GX2 Evga 790i vappochill LS Lianli pc75B
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.
Still we probably need to be careful...it was reckless of me to mention how well 1.9v vmch was working out for me (to you at least )
For some reason the though of killing the northbridge never even crossed my mind. Under water the IHS of the NB is literally cool to the touch...the NB temperature in BIOS is under 30C. If it can die so quickly under air though, then maybe water does nothing but postpone the death. I'm going to try to stay around 1.8v and keep my yack shut about high vmch values from now on. They just make things irresistibly easier though...
At 2.0V I was around 1940 with fully tweaked settings and 7/7/7/21. I assume that the dram clock skew added between 20-40mhz at that point . Also me running Tras 21 vs 14 is worth about 10mhz ( yet with hardly any negative SuperPi time impact), and Vahid was using Tras 14 - so I think we are very close in Cas7 2.0V results without tweaking. I will run some more tonight and report back.Originally Posted by saaya
Last edited by mike; 04-20-2008 at 02:04 PM.
Just giving ya the blame...you take no responsibilty at all. Benchzowner made an important point - keeping the NB below the 50C is the keyin terms of how much voltage the NB can take. You ( Gautam) had mentioned to me how much cooler the NB was running for you compared to X38 boards - this is probably the real advantage of the X48 bin over the X38bin.
hey mike, you said you also noticed clock skew seemed to be tweaked different at different cas latencies? did you find out anything in that regard? im really curious about clock skew adjustments, your the second guy to tell me they can make a really big difference. maybe the third, stilt told me a looooong time ago that ODT tweaking can make a huge difference... and i never really figured out what he meant by that maybe he was referring to clock skew after all.
I wonder if clock skw adjust has anything to do with read/write leveling.
compared to ddr2, ddr3 gets the command signals in one line. for ddr2 each chip has its own command line, for ddr3 the command line goes to one chip after the other. well it actually has a T split between the left and right 4 chips, and i think on 2gb sticks probably 2 T splits. But as a result of the trace going to one chip, then the next, then the next etc, there is added latency depening on what chip you talk to.
if you tell chip1 to do something itll reply faster than chip2, which will be faster than chip3 etc... in the intel docs this is sometimes called ODT, on die termination, sometimes read/write levelling...
and it seems asus now calls this clock skew adjust
http://en.wikipedia.org/wiki/Clock_skew
this sounds spot on, doesnt it...In circuit design, clock skew (sometimes timing skew) is a phenomenon in synchronous circuits in which the clock signal (sent from the clock circuit) arrives at different components at different times. This can be caused by many different things, such as wire-interconnect length, temperature variations, variation in intermediate devices, capacitive coupling, material imperfections, and differences in input capacitance on the clock inputs of devices using the clock. As the clock rate of a circuit increases, timing becomes more critical and less variation can be tolerated if the circuit is to function properly.
There are two types of clock skew: negative skew and positive skew. Positive skew occurs when the clock reaches the receiving register later than it reaches the register sending data to the receiving register. Negative skew is the opposite: the receiving register gets the clock earlier than the sending register.
Two types of violation can be caused by clock skew. One problem is caused when the clock travels more slowly than the path from one register to another - allowing data to penetrate two registers in the same clock tick, or maybe destroying the integrity of the latched data. This is called a hold violation because the previous data is not held long enough at the destination flip-flop to be properly clocked through. Another problem is caused if the destination flip-flop receives the clock tick earlier than the source flip-flop - the data signal has that much less time to reach the destination flip-flop before the next clock tick. If it fails to do so, a setup violation occurs, so-called because the new data was not set up and stable before the next clock tick arrived. A hold violation is more serious than a setup violation because it cannot be fixed by increasing the clock period. Positive skew and negative skew cannot negatively impact setup and hold timing constraints respectively (see inequalities below).
and its really easy to calculate the best clock skew adjust it seems:
i dont know about you guys but... i prefer trial and error
not that i wouldnt know how to calculate the ideal clock skew, pfff, so easy i just... like trial and error... its... more fun
Last edited by saaya; 04-20-2008 at 11:28 PM.
Mike - when You tested them on Maximus Extreme, did You use 333 strap or 266?
BTW - X48 might be better on the subtimings but on my Maximus Extreme I can run 8.5 x 542 @ 1.65 Vmch with the NB on water:
Sadly the Dominators don't run 1:2 divider @ 500FSB, so I'll give theese blue Cellshocks a chance....., will receive a kit this week.
Then I'll do some testing @ +500FSB and 1:2 divider, hope to push theese baybies up to 1084.....
What about Your 7/7/7/14 testing?
hmmmm quite interesting, did you guys compare the results from vahid at techpowerup, metro and mike here? mike reached very high cas7, cas8 almost got him no gains. vahid got high cas7 or 1900+ and cas8 got him almost no gain, metro got very high cas8, but cas7 didnt clock that well.
well, this could all be different tweaking and adjusting... but while testing and playing with d9jnl thats what i noticed too. some seem to be limited by clockspeed, others by cas. interesting huh? if the limit is by clockspeed then its most likely the memory chip interface thats holding the chips back, not the actual logic. intresting huh?
I was running both 266 and 333 straps - I think the 9 multis were on 266 strap.
Tras 14 cost about 4-8mhz in my max clocks so far
Just been runnning some SuperPi 1ms - got sidetracked with a new E8500 on air today...
Another little crappy vid of the 1m run -- boot into XP @ 2020mhz
http://www.youtube.com/watch?v=gWE3k8EpaNo
Last edited by mike; 04-21-2008 at 04:33 AM.
Damn, crazy chip there.
Clock skews are indeed quite annoying in IC design. Quite interesting that they are giving the user some control over clock distribution these days in BIOS...
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
Bookmarks