1:1 ratio can someone pls explain it to me

zeroyb

Distinguished
Nov 10, 2001
104
0
18,680
I understand why 1:1 would be an ideal ratio, but I don't quite understand how it works. Can someone please answer a few qustions? By the way, I'm assuming timings are equal for argument's sake. Thanks a lot!

1. If I overclock FSB to 400mhz, why would I want 800mhz RAM? Would there be any difference if I got 1066mhz RAM and ran it at 800mhz instead of getting 800mhz RAM?

2. Maybe I can figure this out if I had question #1 answered, but... So, when overclocking, are only FSB speeds such as 667mhz, 800mhz, and 1066mhz ideal for best memory performance? Or can you get optimal memory performance with any FSB speed you want, such as 450mhz, as long as you run RAM at 1:1 (900mhz in this case)? If so, would OC'ing a 800mhz RAM be better, or would 1066mhz RAM be better?
 
Okay, there are several other posts that cover most of this material, so my first suggestion is that you peruse older posts, such as this post.

1. the "800MHz" DDR2, actually only runs at 400MHz. DDR stands for Double Data Rate, so it sends data on the rise and fall of the clock signal, or twice per clock. The 800 MHz is then the effective speed of data being sent twice per clock at 400 clocks per second.

2. DDR2-1066 is really just overclocked DDR2-800. DDR2-800 is the highest JDEC standard, so anything more is just verified to function in a given overclocked state. The reason that 1:1 memory timings are preferred is that in general, having memory that can send data at a faster rate than the MCH can handle it doesn't really provide a performance benefit. Also, it puts extra stress on the MCH, which has to deal with incoming memory which is not synchronized with the FSB clock.
 
Thanks. I don't think you've really answered my questions. I've actually read that post and many similar ones. But, I couldn't understand it 100%, that's why I posted.
 
I'm still a bit confused about the whole QDR thing....

QDR = FSB x 4 (eg 400MHz FSB x 4 = 1600MHz QDR)

SO, what is the function of QDR? what is its job in the system? is it the speed at which the CPU communicates with the northbridge?? if so, that would be double the RAM, and not 1:1....

Someone please explain?
 
I think u answered your question,in some boards like NVIDIA 680i/780i boards u have something called QDR which as u said is the FSBx4.But in some other boards u have the normal FSB (for example if i have a
356MHZ FSB then in 680i/780i boards its 1424 FSB)

So the QDR is the normal FSBx4
 
My answer to point 2 should answer point 1. To make it more clear: since DDR2-1066 is really just overclocked DDR2-800, when you run the DDR2-1066 modules at DDR2-880 (400MHz FSB 1:1, or something like that), it will behave the same as memory that claims to be DDR2-800. They are the same exact modules, just tested, rated and packaged differently.

As for 2: Optimal memory performance is really dependent upon memory throughput and timings. Tighter timings mean that the memory and MCH sit around waiting to send data, or waiting for data less time, and spend more time transferring data.

Optimal memory performance is actually quite a tricky subject. Running your memory at a 1:1 FSB ratio is simple, and it tends to give good performance and put less stress on the system. Higher dividers won't hurt memory performance, and may enhance memory performance, but it's usually just a synthetic difference that makes very little difference in real-world applications. There are some applications which are quite sensitive to memory speed, though: WinRAR and UT3 are both sensitive to memory bandwidth, as can be seen in many of Tom's articles.

So, you can get good memory performance at FSB 450 if you can run your memory at 1:1 divider. It'll be better than trying to down-divide the memory speed, and it will be easier than trying to overclock the memory more to fit some other speed.

Basically, get your FSB as high as you can when your CPU multiplier is locked. Keeping your ratio at 1:1 will probably be the best. An FSB frequency of 450MHz is great. Run your memory at 900MHz effective if it'll take it, and keep timings as tight as possible. Keep your tRD as low as possible too... that usually means backing off the FSB speed. Check out this fascinating Anandtech article.
 
Hmm... According to wikipedia (a far more reliable source than I), QDR is the way Intel processors communicate with the memory controller, transferring data on the rise and fall of dual clock signals, which are 90deg. out of phase. So I think the very utmost optimal situation would be QDR memory which would operate then at the same effective speed. I'm not sure if DDR in dual channels offsets this, though... perhaps.

At any rate, very few applications seem to gain a benefit from running your memory asynchronously with the FSB. I'll have to dig around to find a better explanation as to why this is, but I think the Anandtech article above touches on the subject.
 



well, if you had 800mhz ram, you would have to OC it to 900mhz by adding voltages and changing memory timings and such which is a lot more hassle than if you just have 1066mhz ram and ran it at 900mhz because your 1066mhz would be rated at that speed and should be able to run at any lower speeds just fine. it depends on the amount of OC-ing you have and if you are comfortable to work with such added variables of stability.

to my knowledge there would be no difference in running 1066mhz at 800mhz or 800mhz ram normally. timings and such are on the specifications of ram that you buy.

the increments of 667,800, and 1066 are there because of the increments of FSB that intel had increased over the years with its processors. mainly by 66-67mhz in generations...kentsfield = 266... penryn = 333....

take the Q6600 for example, it has a FSB of 266... 266 (fsb) x 4(quad pumped) = 1066 (fsb frequency) ....so in reality, to run 1:1 ratio with the Q6600, u only need 533mhz ram. but most people love to OC the Q6600 to a minimum to 333 FSB.

so 333x4 = 1333mhz fsb frequency....so therefore, you would need 1333/2 = 667mhz ram to run it in 1:1

on the new generations of penryns and such, the stock fsb is 333x4 = 1333 fsb freq.... but people usually go to 400fsb or beyond...
or 400x4 = 1600 fsb frequency... so at 1600mhz freq, you would need 800mhz ram to run 1:1 with it.

anything beyond 1600+ would require you to OC the ram, or get ram that are rated higher such as 1066mhz ram.
 
I think u answered your question,in some boards like NVIDIA 680i/780i boards u have something called QDR which as u said is the FSBx4.But in some other boards u have the normal FSB (for example if i have a
356MHZ FSB then in 680i/780i boards its 1424 FSB)

So the QDR is the normal FSBx4
dude read my question before u make some half arsed answer.... roflmao in 680i/780i only huh?? yeah well u need to learn a lot more then... i know HOW to calculate the QDR i want to know what it's FUNCTION is in a system.... ffs seriously sometimes i wonder.
There are two simple ratios that you need to remember:

(1) a front side bus ("FSB" ) for Intel CPUs is "quad-pumped":
it sends and receives data 4 times per raw tick of that bus clock;

(2) modern SDRAM is double-data rate ("DDR" ):
it sends and receives data 2 times per raw tick of that bus clock.

So, with recent Intel machines, a 1:1 FSB😀RAM ratio
can be easily achieved as follows:

(1) 1333MHz FSB / 4 = 333

(3) DDR2-675 / 2 = ~333


For this reason, it's easy to purchase RAM that is "too fast":
if one is not over-clocking, a 1:1 FSB😀RAM ratio can be
achieved with DDR2-675, assuming the FSB is 1333MHz.

Just scale those parameters up and down, for other
combinations of FSB and DRAM clocks.

ONCE AGAIN, i know HOW to calculate it, i KNOW the ratios, i do not know why the QDR exists!! what is its job in a system!!

Hmm... According to wikipedia (a far more reliable source than I), QDR is the way Intel processors communicate with the memory controller

THANK YOU!! finally someone can give me a definition of the function of QDR.... now that wasnt too hard now was it? :bounce: