• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

DRAM:FSB ratios

M

msjones

Weaksauce
Joined
May 21, 2008
Messages
71
Hi,

I have an overclocked E4600 CPU running on a P5W mobo (specs in sig). I have a few questions about the RAM.

I am using Corsair XMS2 2x1GBPC6400 C4 sticks. When over clocking the CPU i didnt let the motherboard set the speed or timing for the RAM. I clocked my CPU at 12x240 (2.88GHZ) and set the RAM to 800MHZ and the timings to 4-4-4-12.

After booting up, Everests CPU-Z app shows the following:

newpicturew.png


Now I have read that memory shold be 1:1 when over clocking but mine is 10:6. Is this safe? I ran stability test for 8+ hours over night and no errors.

thanks in advance
 
msjones said:
Hi,

I have an overclocked E4600 CPU running on a P5W mobo (specs in sig). I have a few questions about the RAM.

I am using Corsair XMS2 2x1GBPC6400 C4 sticks. When over clocking the CPU i didnt let the motherboard set the speed or timing for the RAM. I clocked my CPU at 12x240 (2.88GHZ) and set the RAM to 800MHZ and the timings to 4-4-4-12.

After booting up, Everests CPU-Z app shows the following:

newpicturew.png


Now I have read that memory shold be 1:1 when over clocking but mine is 10:6. Is this safe? I ran stability test for 8+ hours over night and no errors.

thanks in advance
Click to expand...


1:1 only matters if your going up into the 400FSB range.. at 1:1 400fsb = 400mhz(800mhz DDR2) so with your current overclock 10:6 @ 240mhz fsb = 800mhz.. so its fine..
 
sirmonkey1985 said:
1:1 only matters if your going up into the 400FSB range.. at 1:1 400fsb = 400mhz(800mhz DDR2) so with your current overclock 10:6 @ 240mhz fsb = 800mhz.. so its fine..
Click to expand...

Is performance any better the way he did it, vs setting ram to 240 mHz (DDR2-480)?
 
There isnt any difference in performance that I have noticed. I'll leave running at 10:6 @ 4-4-4-12 all seems to be good.

Want to push to 3GHZ but cant seem to get passed 299MHZ FSB even with multiplier change, just wont post.
 
yes theres a huge performance difference between ddr2 480 and ddr2 800..


try setting your memory timing to 5-5-5-18, msjones and then see if it will post at 299.. your probably pushing the memory to far.. if it doesnt work with the timing at 5-5-5-18 then try dropping the memory multiplier 1 more notch and see what happens.. also could be the cpu voltage needs to be bumped up as well..

when overclocking its a good idea just to set your memory to 1:1 timing until you get to your max cpu overclock that you want.. then slowly raise your memory multiplier up til you get back to 400mhz(800mhz DDR2)
 
sirmonkey1985 said:
yes theres a huge performance difference between ddr2 480 and ddr2 800..
Click to expand...

It makes sense if both are 1:1 ratio, but I hear there's no performance gain if you go above 1:1, so 10:6 is the same as 1:1, both perform like DDR2-480.

Dunno if that's true, I just heard that.
 
Thanks for all the replies. This is where I am at now.

Leaving the clock multiplier at 12x I increased the FSB to 250 giving me the 3GHZ, system would not post. So after power down I went back into the bios and upped the voltage and kept trying to boot the system. I successfully managed to boot into the OS at 1.3375v. This has me a little worried because looking on intels site it says the voltage range for this chip is 1.162V-1.312V.

I left my RAM timings set to 4-4-4-12 but changed the memory voltage settings from 2.1v to auto. On my mobo after setting the FSB my RAM frequency defaulted to DDR2-667 which I have left it set at, this gives me a 8:6 ratio.

newoc.png


This is my current state of OC as I type. My idle temps are hovering between 27C & 30C (Checked using both Core Temp & Everest). I have not yet tested the load temps as I am a little weary of my current CPU voltage settings.
 
the voltage intel gives are the rated VID's.. basicly it means any cpu for that series will not be lower or higher then that voltage.. but its for stock speeds.. the e4600 is a 65nm process if i remember correctly.. which means it has a max voltage of 1.6v.. 90nm is 1.7v 130nm is 1.9v.. 45nm is 1.4v(intel) 1.55v(amd), i7 is 1.5v
 
Meh, I am going to leave the OC at 2.88GHZ with the RAM at 800. The stability test has just fell over and rebooted my PC.

I suppose 2.88GHZ isnt bad really, I am getting a little extra power and it is noticable. So I assume i am Ok to set my RAM back to 800MHZ (10:6)
 
if your temps are all within the norm try bumping the voltage up some more.. you probably werent that far off from what was needed since you got into windows.. usually getting into windows is a good sign that your close to the needed voltage to make it stable..
 
Well my idle temps are at about 27-30c average about 29c and my load temp is at 45-52c. What do you think, are these good temps to push a little more voltage through?

What do you think about my XMS2, should I also up the voltage to 2.1v?
 
As long as you keep the load temps around 65C and the vCore under 1.5V, go for the highest OC possible within those parameters and sustainable by your mobo.

Set the RAM voltage to the highest officially supported voltage for that RAM.
 
Nice one, will give that a bash when I get home. Ill report back later.

Thanks for all your help and patience for the new guy!
 
Quick update.

I have upped the vcore to 1.3500v and the memory to 2.1v. I am nipping out for a few hours so I will leave the stability test running.

Hopefully this will sort it!
 
If it were me, id go with 400 with 1:1, 400FSB which would make the ram run at DDR2-800, and lower your multiplier to 7, so you get 7x400=2.8ghz, or if you want to push it further, you can do 8x400=3.2ghz, if you can get it stable up there.
 
nissanztt90 said:
If it were me, id go with 400 with 1:1, 400FSB which would make the ram run at DDR2-800, and lower your multiplier to 7, so you get 7x400=2.8ghz, or if you want to push it further, you can do 8x400=3.2ghz, if you can get it stable up there.
Click to expand...

I have already tried this but wouldnt post, even after increasing the vcore to past 1.4500v. I am happy with my 3GHZ OC I have dropped the RAM to 500MHZ to give me a 1:1 ratio.

Which app would you say is best for measure core temps? I have open at the moment Everest, Core Temp and Real Temp.

Core Temp & Everest = 27-30C idle - 49-57C load
Real Temp = 32-35C idle - 55-59C load

Both seem to be well within range I was just wondering which would be more accurate.
 
raise your memory timing.. 500mhz ddr2 is worthless and bottlenecking your system..

for intel systems real temp is your better option..
 
sirmonkey1985 said:
raise your memory timing.. 500mhz ddr2 is worthless and bottlenecking your system..
Click to expand...

Entirely incorrect. His FSB at 250 is the bottleneck.

msjones said:
So ignore my ratios and what up my RAM freqs.?
Click to expand...

The best way to do it is to keep them 1:1. Running your ram at 500(250x2) or 800(400x2) doesnt matter since your FSB is at 250. If your ram is rated for 800(400x2), then youd get the most by raising your FSB to 400 and lowering your multiplier accordingly. If you can get your CPU to post at 2.9ish with 250x12, but not 7x400, with the same CPU vCore than the issue is most likely with your north bridge voltage.

EDIT: I use RealTemp.
 
Don't go straight from 250 to 400 FSB, cause the CPU's limit is probably somewhere in the middle, like 350 mHz.
 
Dude, it dosen't matter if it's 1:1. Use another divider, 10:6 is fine. The extra bandwith will help make quite a difference. 480mhz thats way slow.

Just run the Cache Bench on Everest or Download Sisoft Sandra for free. You'll see a difference.
 
Super pi should also tell you whats better. You could run IBT and see what kind of Gflops your running. There are tons of benches you could do that would prove a 1:1 divider is not the way to go, unless you have to.
 
Lol. Yeah youll see a difference in SiSoft Sandra...which im sure will translate to real world performance. The only time it makes sense to not use 1:1 is when your RAM limited before CPU FSB limited. Which isnt too often these days.
 
He should still lower his multiplier though, to raise the CPU's FSB.
 
nissanztt90 said:
Lol. Yeah youll see a difference in SiSoft Sandra...which im sure will translate to real world performance. The only time it makes sense to not use 1:1 is when your RAM limited before CPU FSB limited. Which isnt too often these days.
Click to expand...

Honestly, that dosen't even make sense. If you can get more bandwith, which he will, then why not. Yes, going from 480mhz to 800mhz will translate to real world gains. What do you think all of your data fits on the L2 cache?
 
nissanztt90 said:
Because your limited by the bandwidth of the FSB.
Click to expand...

How so? The L2 cache is genealy 3 times faster than your memory bandwith. You will never see your memory bandwith faster than your L2 cache. Your ram also recieves and sends data to more than just your cpu. The extra speed will help him alot. Also FSB does not directly translate to the bandwith on your cpu, he also has a 12x multi.
 
Thanks for all the replies. So my final stable OC is 12x250 (3GHZ) and RAM @ 4-4-4-12 2.1v 750MHZ using a 12:8 ratio.

Temps are still low with Real Temp saying idle is between 30-35c and load 50-57c.

I have tried lowering the multiplier to get higher FSB to run 1:1 with RAM but it doesn't seem to want to post.

Final CPU-ID output.
newoc2.png


Is everything looking in order? I thought i'd start out learing to OC on a budget system so I can move on to higher end components. This way if I kill anything it cheaper to replace till I know what I am doing.
 
nissanztt90 said:
The CPU has its own FSB limit as well. Try 333x9, run your ram at 667.
Click to expand...

Should I try a lower voltage?

Its my impression that with a lower multiplier you can use a lower voltage, or is that onply applicable with the FSB.... forgive the dumb Qs
 
msjones said:
Should I try a lower voltage?

Its my impression that with a lower multiplier you can use a lower voltage, or is that onply applicable with the FSB.... forgive the dumb Qs
Click to expand...

No worries, i was in your shoes a few years ago asking the same questions.

The multiplier multiplies your FSB speed to give you your CPU speed. If you keep the same FSB speed, and lower your multiplier, you lower your CPU speed, and yes you can lower your CPU vCore. In your case, if you went to 333x9, your lowering the multiplier, but your increasing the FSB speed, so youd need to keep the same voltage.

The voltage needed generally depends on the frequency its running at. So the CPU vCore shouldnt need to be any different. For example, you can run 250x12 = 3.0ghz. Or you can run 500x6=3.0ghz, or you can run 333x9 like i suggested. The vCore should be roughly the same, however, the CPU itseflt has an FSB wall, just like the motherboard also has an FSB wall. A "wall" is basically a dead end...wont go any higher no matter what. Sometimes the CPU reaches it first, sometimes the motherboard reaches it first.

Then you take the speed of the ram into consideration. It is best to maintain a 1:1 ratio. The only time you wouldnt want a 1:1 ratio is if your ram cant handle a higher speed than your FSB. Your ram is rated for 400 (400x2 = 800) but your FSB is only 250 currently, so your ram isnt a limiting OC factor.

You shouldnt need to alter the CPU voltage too much to go from 250x12 to 333x9. You will may have to up the northbridge voltage to get the higher FSB speed stable or bootable.

For example on my P5Q i can do 400x9 = 3.6ghz, my FSB needs 1.2, for 445x9=4.0ghz, my FSB needs 1.32-1.34.
 
Thanks alot for that. That was much more understandable than the other 10 guides I have read about OC'ing.

Just another question, on the ASUS boards I cannot see anything for adjusting the NB voltage?
 
Your welcome. Different boards may call it different names. On my Asus its simply called NB Voltage. Take a look in your manual to see if you can spot it.
 
Honestly you would be crazy to run your ram any lower than you can. Intel boards love high mhz ram while nvidia boards love a 1:1 ratio with tight timings. A higher mhz clock will net you lower latency and more bandwith and at the speeds that you are running you WILL see real world gains by running 800mhz+.

I have never seen a benchmark that would give me a reason to run my any multi lower than it can go. A high FSB clock will need more VCC and VTT than the same clock with a lower FSB and higher multi.

Also I have found Intel Burn Test to be very good at detecting northbridge and memory errors.
 
BababooeyHTJ said:
Honestly you would be crazy to run your ram any lower than you can. Intel boards love high mhz ram while nvidia boards love a 1:1 ratio with tight timings. A higher mhz clock will net you lower latency and more bandwith and at the speeds that you are running you WILL see real world gains by running 800mhz+.

I have never seen a benchmark that would give me a reason to run my any multi lower than it can go. A high FSB clock will need more VCC and VTT than the same clock with a lower FSB and higher multi.

Also I have found Intel Burn Test to be very good at detecting northbridge and memory errors.
Click to expand...

Just to clarify what you mean - so for example on an e7400, on Intel board:

2.8 GHz @ 266 mHz x 10.5 (memory divider = 3:2 or 400 mHz)

2.8 GHz @ 400 mHz x 7 (memory divider = 1:1 or 400 mHz)

You're saying the top one would perform same or better, even though the CPU FSB is a lot slower?
 
Just ran SuperPI...2 million digits.

This is on a Q9650 with an Asus P5Q

FSB at 266, 1:1 with RAM, CPU at 2.4ghz, RAM at DDR2-533, timings 4-4-4-12
Time for 2M digits: 51 seconds

FSB at 266, 1:2 with RAM, CPU at 2.4ghz, RAM at DDR2-1066, timings 5-5-5-15
Time for 2M digits: 48 seconds.

Doubling the ram speed knocked a whole three seconds off.
 
nissanztt90 said:
Just ran SuperPI...2 million digits.

This is on a Q9650 with an Asus P5Q

FSB at 266, 1:1 with RAM, CPU at 2.4ghz, RAM at DDR2-533, timings 4-4-4-12
Time for 2M digits: 51 seconds

FSB at 266, 1:2 with RAM, CPU at 2.4ghz, RAM at DDR2-1066, timings 5-5-5-15
Time for 2M digits: 48 seconds.

Doubling the ram speed knocked a whole three seconds off.
Click to expand...

Thats just one bench that Im sure uses mostly L2 cache, but 3 seconds is alot in spi. Now what is the difference in bandwith/latency?
 
Since you know so much, you can tell me. Im going to actually use my PC instead of bench it. If you wanna go nuts over how much better it is to have a higher ram speed than the FSB, go right ahead and play those benchmarks.
 
Try 3dmark 06 or wprime. He will se an improvement in day to day apps from a quick tweek.
 
You must log in or register to reply here.
Back
Top