Next Question! Running RAM at 1:1? How Important?

Associate
Joined
17 Dec 2008
Posts
103
Location
Birmingham, Uk
Hey guys,

I know your sick of me, but while my .NET projects compile i have loads of time to read stuff and think up new questions for you all!

I'm planning to overclock my e8400 + P5Q-E to 4.0ghz.

This means that i will need a FSB of 445 (x 9)

I run DDR2 1066 Memory and my BIOS will let me run it at the following speeds:

667/800/835/887/1002/1066/1111

Assuming i actually achieve the 445 FSB (4000/9) would it be better to underclock my memory to run at 887mhz?

i.e.
FSB: 4000 / 9 = 444.4 FSB

DRAM:
1066 / 2 = 533mhz
OR
887 / 2 = 443.5 (Near 1:1)


Surely having the RAM set to as fast as it will run would be more benificial that running it in a 1:1 relationship with the FSB frequency?
 
its not important, just run the ram as fast as it will go, any much over the JEDEC rated 800mhz there really isnt much increase in performance
 
i think 1:1 ratio makes it more stable, but if its stable at a faster speed then go for it :D

although lowering your multiplier and upping the fsb to achive the same speed will be better, higher fsb better performance :D
 
It used to be important with the Athlons due to their integrated memory controller, but these days, raw speed gives you more of a benefit than low latencies, as long as you're not going higher than 5-5-5-12.
 
Lovely ... so there is more of a benefit of keeping my frequency at 1066Mhz and running at a dodgy ratio rather than running it underclocked to keep it at 1:1 with the FSB.

Jesus those were some fast replies - thanks all :)
 
although lowering your multiplier and upping the fsb to achive the same speed will be better, higher fsb better performance :D

Lol - just read this again. If i up the FSB to match the Ram Freq then i'll be running at 4.8 ghz on air!

I think i'll just keep my target at 4ghz because i'm not planning on running Nasa anytime soon :p
 
Lol - just read this again. If i up the FSB to match the Ram Freq then i'll be running at 4.8 ghz on air!

I think i'll just keep my target at 4ghz because i'm not planning on running Nasa anytime soon :p


Not necessarily - dropping your multi to 8 would give you 4.3ish ghz at a FSB of 540 with your RAM maxed out at 1:1 (not saying the mobo would do that though)
 
Ah .. sorry, i miss read it, missed the bit about dropping my multiplier.

Not sure i can get it up to 4ghz yet but it might be something to think about ...
 
Assuming i actually achieve the 445 FSB (4000/9) would it be better to underclock my memory to run at 887mhz??
No it wouldn't be a good idea, you want your effective memory speed to match your System Bus as closely as possible *not* your FSB.

If your running an overclocked E8400 at 4GHz (9x445) your system Bus is running at 1780MHz which is nice and fast however it means jack **** if the memory is crawling along at DDR2-890Mhz . . .

FSB= 445MHz
System Bus= 1780MHz
Mem (1:1)= 445MHz/DDR2-890MHz

Once you know how fast your CPU can run and how fast your FSB can run you will get a faster system if you then take the memory from 1:1 sync and use a higher multi so that the memory can keep up with the *effective* system Bus speed.
 
Thanks for that Wayne ... IIRC the system bus is the FSB x 4. Due to the System bus being 'Quad pumped' right?

So what your saying is, that my RAM of 1066 still won't be fast enough for an overclock of 4.0ghz? i.e.

1066mhz is still slower than 1780mhz i'm trying to achieve in my overclock?

Will this matter and/or make a difference?

I've also heard that LOWERING my multiplier and making my FSB faster is better than what your suggesting? Which is better for my 1066 RAM?
 
1066mhz is still slower than 1780mhz i'm trying to achieve in my overclock?

Will this matter and/or make a difference?
Well 1066MHz is still faster than 890MHz and as such any data passed from the CPU to Northbridge to Memory (and back again) will be quicker, any tasks you peform from gaming to encoding will be faster. The process of memory tweaking is best left to the final stages of your overclocking as you first need to work out how fast your CPU can go (and what volts it needs) and how fast your FSB can go. During CPU and FSB testing it is recommended to leave the memory at 1:1 Ratio.

Sadly DDR2 speeds can not keep up with the effective System Bus speed especially when overclocking so just do your best to keep the Processor fed with information by running the RAM as fast as it can go.

Don't just take my word as gospel though but instead do your own testing based on what I have said, you should find a nice overall speed increase with your RAM running blazing fast!


I've also heard that LOWERING my multiplier and making my FSB faster is better than what your suggesting? Which is better for my 1066 RAM?
Again you are wandering into advanced overclocking territory and it will probably be easier if to begin you stick with the *default* chip multi (in your case x9). Once you have accomplished the basics you will be in a much better position to start tinkering with non standard multipliers (i.e walk before u can run).

It is a seldom known fact that the chipset has to work a lot harder when running the CPU with a lower divider than it's default and you will run into a whole new set of problems which could fry your brain at this early stage but yes in Theory the overall system speed can be increased by lowering the stock multi and increasing the FSB . . . just leave that bit out for the moment! :p
 
Ahh i'm on the bus, Thanks for that, a good explanation is always welcome :)

Just so i'm clear tho, if i do your earlier calculation on my existing stock speeds (9x333):

FSB= 333MHz
System Bus= 1332MHz

My top spec 1066mhz memory still isn't fast enough for a 1:1? How is that possible? :eek:
 
Your top spec 1066mhz memory is more than fast enough for a 1:1 ratio, your forgetting that the memory speed is doubled so when your talking about 333MHz-FSB/1332MHz-System Bus the memory when running a 1:1 ratio would actually be running at 333MHz/DDR2-666 :)
 
Whats the point in trying to match the ram speed to the CPU's internal clock. FSB is the system bus (more or less). Sure the CPU is quad pumped on the FSB, and DDR is "only" double pumped,

FSB= 445MHz
Effective speed Quad pumped = 1780MHz
Mem (1:1)= 445MHz/DDR2-890MHz

Unfortunatly this only tells half the story. The FSB is a 64bit interface, but the memory is on a 128bit dual channel interface. PU

So in reality...
CPU = 1720mhz x 64bit
Mem (1:1) = (445mhz/DDR2-890) x 128bit

End result 1:1 memory can fully saturate the cpu's available bandwidth, so the potential lower latencies which can be achieved with 1:1 memory is still a viable way to improve system performance.

Dont believe that wide memory bus's work... Intel i7 has a 192bit DDR3 memory controller, or look at graphics cards up to 512bit DDR controllers. Unlike multicore CPU's memory controller / bus width scales almost perfectly. It doesnt matter if you get 64kb as a very fast 1 bit wide bitstream, or a slower 512bit wide interface, as the overall bandwidth will be identical. If all things were equal and low latency was possible at high clock speeds then the narrow bus would give better performance on very small transfers, but on larger transfers there really is no difference.
 
Nicely written Corasik! :)

You sound like you got a degree in Computer Science or something! :p

Sadly I am just an experienced layman and I can't understand a lot of what you wrote! :o

I ran an Intel® Core™2 system for 18 months using a 1:1 memory ratio, it worked fine, fast and stable etc etc but since doing some intense research into the MCH/Northbridge/NBCC/Straps/tRD etc I found that everything works faster once you run the memory asynchronously on an upward divider.

At the moment I am firmly believing that a Intel® Core™2 system running using stock 333MHz-FSB/1332-System Bus will run better if matched to overclocked RAM running asynchronously at 666MHz/DDR2-1332MHz and not synchronously at 333MHz/DDR2-666

If you care to break your findings down into bite-sized chunks I would be happy to read! :cool:
 
Back
Top Bottom