(Notice: I changed the topic title half way of the post due to better relations to the topic)


I have a KHX6400 Kingston HyperX 2x1G (800mhz), 5-5-5-15

Here's the thing regarding of harnessing the highest possible performance out of this Ram:
800mhz with tighter timing of 4-4-4-12 (with increase vMod) @ 1:1 FSB ratio VS. increasing the mhz to 1066mhz+ but with default timing (5-5-5-15) @ 1:1.2 FSB ratio?

OK, people are buzzing about that you need to set the FSB ratio to 1:1, but I test out in Everest's memory benchmark and I get higher latency score (72ns) compared to when I just plain increase the ram to 1066mhz (63ns), so what I am trying to figured out is that:
1. Is bumping your ram clocks can squeeze out more performance than setting the ratio of fsb to ram 1:1 with 4-4-4-12)? (higher clocks > tighter latency w/ stock clocks)
2. 1:1 DRAM:FSB ratio is an old-school approach, meaning is no longer necessary is the future of ram OC'ing?
3. Or is just that this synthetic benchmark won't do justice compared to the real world applications, meaning 1:1 still reign supreme when for e.g. intense gaming?

FYI, I set both settings of higher clocks and tighter latency at default/same value of PL (Asus Ai Transaction Booster)
After hours of Googling, still I can't reach to any conclusion regarding my inquiries, hope anyone can help me out...tq

This post has been edited by JoBigShow: Dec 13 2009, 11:22 AM

the ram and timing affect little on overall performance, while the PL have higher affect.. to get lower PL at higher fsb, u need to choose the right divider, usually 4:5 or 3:4 or 5:6 can achieve lower PL, so even at looser timing, it still perform better..

easier to say, those performance binned chip rams are all meant for benchmark.
in terms of normal usage, u wont notice the difference .

d result in running 800mhz 4-4-4-12 and 1066 5-5-5-15 is almost similar .

but 1066 is having higher bandwidth .

Interesting, so in terms of achieving performance, I am to orientate my efforts all on getting the lowest PL value?
And the timing and clocks are just the means/'tools' to get me to tweak ever closer to the lowest value of PL possible right?

usually 4:5 or 3:4 or 5:6 can achieve lower PL, so even at looser timing, it still perform better..
This is really interesting... I have my FSB set at 400, Strapped at 333, I can achieve 1:1.2 by increasing my DD2 from 800mhz to 961mhz.
Can someone clarify me this, is 1:1.2 technically 5:6? Mathematically 5:6 is 1:1.2( 5/5 = 1 : 6/5 = 1.2) but probably a totally different thing in computing i guess...

What other means I can attain stability in getting the lowest value of PL?
FYI, ATM I set my PL @ 10. I'm using Asus P5Q Turbo btw, Any suggestion that I can get lower PL?
Ai Clock Twister?,etc.

-------------------------------------------------------------------------------------------------------------

Alright, I heed your advice about the the favorable ratios for stable low PL operation and I set my Strap to 400 and the ram clock to 1066mhz, therefore achieving 3:4.
My PL is set to 8 and I 'enabled' all the 'Pull-in of CHA/B PH1/2/3/4', which in my case shows channel A 1 to 3 and Channel B 1 to 3.



As for the result above, I hope I can get lower latency in my next 'trial and error'...that is, with your kind assistance

At this point, I really like to hear how you guy (especially the Pros) handle this similar situation as mine.

nb voltage and vdimm can help achieve lower PL stability after u get a right ratio. usually for 400 fsb, pl5 or pl6 is achievable..
this is the sample, it was a quadcore @400fsb PL6, dualcore is easier to do it.




Attached thumbnail(s)

5:6=1.2x
4:5=1.25x
2:3=1.5X
1:2=2x
That the ram divider and the speed of the ram to FSB
For lower PL,you need a good ram,right amount of vmch
BTW read anandtech.com rampage formula review,they go deep into this subject

Wow, I see you manage to set it up near to the "golden configuration" of: 1000hmz + tRD 6 + 4:5, but from the article from Anandtech that OC4/3 recommended (a freaking eye opening article, thanks O/C4/3!), you can set the CAS# to 4, so I just want to know if you able to get that stable on your system, if so, from my understanding you should achieve that golden configuration of 1000mhz + CAS# 4 + tRD(PL) 5 right?
Or like you said its a bit difficult for the C2Q compared to the C2D?



Later tonight i'll have some freetime to "play" with the BIOS
But I hope there are still more I need to know before fiddling around the bios.

Briefly ATM, i can say tell you guys to get 1000mhz (or closest possible), i have to set my Strap @ 266 on my E7400 (FSB now @ 400, default FSB @ 266), set the FSB at 400, therefore achieve 4:5 ratio...now I am in the process of tightening the CAS# with 4 and PL of 6, and balancing the whole configuration by fiddling around the vDimm and vNB. Correct?

I hope right now my way of thinking is correct that the 1:1 synchronous ratio is an old habit dated back from DDR1, and for the current Intel technology such ways of thinking is no longer necessary because compared to back then, now we have more independence in terms of bios fiddling(straps, tRD, MCH, etc) therefore asynchronous ratio(4:5 or 3:4 or 5:6) is more favorable these days ?
Please confirm me on this


Correct me if i'm still wrong
But I think I'm nearly there of understanding about this whole take on system Performance, thanks to that article:

http://www.anandtech.com/mb/showdoc.aspx?i=3208&p=1

This post has been edited by JoBigShow: Dec 12 2009, 10:52 PM

i dont have lga775 system anymore so cant try more
btw, i found that DFI board will give better PL when same setting applied to it

Alright, I manage to lower the PL to 7, the comp will not POST at 6.
I have my divider set @ 4:5. My guess is, i need a tighter timing on the DDR2 (4-4-4-12) to get lower than PL 7, but since my ram is 5-5-5-15 setting it to 4-4-4-12 causes the system to be unstable, hence will not perform POST. So I'm sticking with 5-5-5-15 ATM.

But to make up for the inadequacy, I manage to set the Ai Clock Twister to 'Moderate' and running it stable, will try 'Strong' later.
Here's the benching result of before and after OC'ing the system performance:

Before: (PL 12, 1:1 ratio, Ai Clock Twister on Auto and 800mhz clockspeed of ram)


Current (13/12/09): (PL 7, 5:4, Ai Clock Twister on 'Moderate' and 1000mhz clockpeed of ram)


As you can see the latency is lowered noticeably. I'm still working out on how to get the latency below 56ns, so I'm going to play around with the ratios that PL and getting the ram to stabilizes under 4-4-4-12 setting by bumping up the vDimm/vDDR or by other methods.

Hope anyone kindly can show me how you get the PL at 6, or best at 5

Bigger CPU Cache have lower latency
And as for lower to pl5 or 6,it is matter of vMCH but ram play a part as well
Good ram like D9GKX or D9GMH can hold PL5 no issue

I don't know if my KHX6400 (800mhz) Kingston HyperX use elpida or them D9's, I think they used to use d9's, but will research my rams more, the problem is I don't know where to start looking..

As for the vMCH/vNB, the heatsink on the NB felt warm and borderline-hot when I juiced it up to 1.26v..
Since I'm pretty new to vMCH fiddling, can you recommended me what range of vMCH is good for 'safe' NB OC'ing? (on Asus P45 P5Q series MoBo)

And do you know more on 'Ai Clock Twister'? I know it sets the Dram performance-to-compatible perimeter, but what does the Overclocker's society take/views on that 'Ai Clock Twister'?

This post has been edited by JoBigShow: Dec 13 2009, 11:22 AM

Absolute max 24/7 vmch v are roughly 1.35-1.4v or so
Kingston 800mhz Hyper X should be epida base,D9 base one are 2006 or so stock
D9 for DDR2 pretty much impossible to buy new