Question about OC'ing using Prime95

[Dragon41673]

When Prime95'ing and you fail at a 12x CPU multi...if you lower it to say 11 or 10, and raise the HTT to equal the mhz that the CPU was at when it failed...would that help to pass Prime?

For instance you have the following:

225 mhz @ 12x multi = 2700 and that failed in Prime95

would it help to do this to pass:

245 mhz @ 11x multi = 2695
or
257 mhz @ 10.5x multi = 2698.5
or
270 mhz @ 10x multi = 2700

Any ideas?

 

[stevty2889]

If you lower the multi and increase the FSB and keep the CPU running at the same speed..if the cpu was causing to cause the instablity then it would still be unstable since it's at the same speed...

You should also try to eliminate the ram as the culprit by using a divider to keep it below it's stock speed, and loosen the timings. Make sure you keep the HTT below 1000mhz. Use the 4x multi above 200 and below 250, use the 3x above 250.

 

[Dragon41673]

I am...that's why I was asking the question, just trying to make sure I'm covering all the bases. Thanks!

 

[furballi]

It's always best to use the highest multiplier when overclocking because you don't need to run as high of a RAM speed to achieve the same CPU core speed.

 

[Dragon41673]

Ok...thank you. That's what I've been trying to do but I keep failing prime at 220mhz CPU speed at 12x Multi. Yeah sure it's not bad...but so many other pople are getting so much higher then my 2640. I'm a bit disappointed

 

[freethrowtommy]

to get higher then that, you may need to up the vCore of the processor... on my 4400+, I need to up the vCore to 1.35 V to get to 2400 mhz. It is pretty nice OC at stock voltage for 2640 if that is the case...

 

[Dragon41673]

Not sure what vcore I need...I had it at 1.5 at first...right now it's 1.55...but I'm going to lower it back down and try a few things as I sleep tonight

 

[Capt Caveman]

You should make vcore adjustments upwards in small increments. If you're not monitoring your temps, you could fry your $1000 cpu.

 

[Dragon41673]

I am monitoring them now...lastnight while playing around with it on SuperPI running each test they have. As I went higher in tests, so did the temps obviously. The highest I hit was a 58c running 2640 with a 1.55vcore.

Newer development...I did try running the memory at 333mhz rather then 400 as you suggested. I passed a dual Prime95 test lastnight running 225mhz @ 12x, LDT 4x, and I dropped my vcore prior to running the test to 1.5v as a starting point. I heard, don't know if it's true, but as prime recycles into another pass it's harder to pass...is that correct?

 

[furballi]

10% overclock is about right for those CPUs. Not much headroom. Don't bump Vcore more than 10% from spec. It's best to run at DDR400 if you cannot raise the FSB by more than 6MHz at DDR333.

 

[TheLiberalTruth]

Originally posted by: furballi
It's always best to use the highest multiplier when overclocking because you don't need to run as high of a RAM speed to achieve the same CPU core speed.

Then again, lower multi + higher fsb = better performance (if you can achieve it) than higher multi + lower fsb.

 

[Dragon41673]

I thought that rasing the HTT (fsb) doesn't really make much of an advantage with the A64's...or am I wrong?

 

[Leper Messiah]

What kind of cooling do you have on that? If you're on stock 10% vcore is going to be the max i would recomend, esp. with such an expensive CPU, but, if you have a XP-120 and a decent fan, 15-20% vcore is okay. I run my winnie (default 1.4v) at 1.6 on stock, and it doesn't get past 54C. If you have it under 60C, your doing okay on volts.

EDIT: saw your HSFU, a 7700 would have been a better idea...or an XP-120/90C. 58C is about as high as I would go...

 

[Dragon41673]

2 80mm fans up front, 2 80 mm fans out back, 1 80mm fan in the side of the case blowing between the vid card and cpu. Zalman 7000B-CU on the CPU with AS5

 

[Bona Fide]

You should be getting more...just drop your memory divider to a 2:1 (DDR200) to absolutely move it out of the way. Also make sure you have the PCIe frequency locked, and that your HTT multi is 3x.

 

[Dragon41673]

Tried that already...for some reason I was still getting errors. Checked the mem out on memtest too.

 

[furballi]

Raising CPU core speed by lowering the core multiplier is not very efficient because you will force the RAM to run at much higher speed.

 

[coomar]

actually with xp-m's I found that the cpu would be stable at 12.5x208 but not 13x200 at the same voltage, don't know if this was true for a64's

doesn't matter for a64's you can use a higher multi or lower one, the ldt is not really important (as long as its under 2000 (effective fsbxldt multiplierx2)

some cpus don't like voltage, its better to go in small increments, kind of like weaning your cpu on the voltage

you can use a divider if you want for the ram, you will see a slight difference in performance using it but still the highest cpu speed is the granddaddy, put the ram to 1:2 and find your overall top cpu speed and then figure out the ram speed that gives you the best performance

 

[monster64]

You are partially wrong. My 3500+ venice can run at 250x11 (2750), with 1:1 on the ram at 2.5-3-3-7. If I raise the fsb to 305 and lower multi to x9, its still running at 2750, and I use the ddr 333 divider, the ram luckily it runs at exactly 250 MHz also. You get about a 5 fps boost in games is you do 305x9 than if you do 250x11. You ram must be running at the same speed though.

 

[coomar]

not sure whats going on there, the ldt isn't supposed to effect benchmarks signifcantly