I'm trying to overclock my e8400 to 3.6Ghz. Just a nice sedate amount for me.
I have a strange problem that manifests itself whenever I hit this speed.
First I upped the FSB to 380 (380*9 = 3420MHz) and set the memory multiplier to 1:1 (760MHz DDR)
I gave the FSB a +0.1v tweak, and set the CPU voltage at 1.28v
The PCI-e frequency is set to 100MHz
I ran prime 95 (not for a huge amount of time, but enough to convince me the machine was not about to reset) and all seemed good.
I could then pull about 11500 in 3dmark06.
Now the problem.
When I increase the FSB to 400MHz for my 3.6Ghz clock I can run prime95 fine.
I had upped the CPU voltage a little more (in fact I've tried voltages from standard (fails) to 1.3v)
However when I run 3dmark06 the graphics seem to lag. The framerate is fine, and I will score around 12500. But the graphics stutter. It's as if half or quarter second chunks are missing.
For example, in the canyon ride test where the air-boat thing is coming along the canyon it's very jerky.
I've been watching the temps (using the motherboard utility) and they don't rise above 42 C.
Core temp tells me the cores are up at 62 C but I know this is bogus.
Wanna know what drives me nuts? I dont know if you are runnin' the tests after ya boot or what, but stupid pre fetch can do that to me sometimes. Though it doesnt effect whatever its doing, it looks like it does. As in, my mouse jumps around and freezes like that, though its not "really" doint it! And if I release a huge amount of RAM, like from prime 95s blend test and swap to a test that uses way less ram, pre fetch is at it again!
Oh, anyone know how to kill pre fetch? Damn thing trashes my Crysis!
Damn, Didnt even mention the other thing. Core Temps! Is your Bios up to date? I just opened a brand new p5k e with whatever bios, and my temps are funny too. Core temps are like 47 max under load, lol, but Tcase gets to 47ish. and I have noticed that core temps like to be about 14 c hotter. So its obviously detecting them wrong. But Idle, now thats funny... cores at like 32 and t case at like 36. yeah right! Check the bios, and get HWmonitor!
Then again, you could be on some cool liquid, and not get that hot!
not sure about the pre-fetch thing, but it seems to be consistent when I go over 385-390 MHz on the FSB. Can you explain what you mean by pre-fetch for me?
BIOS is up to date.
Maybe it's temp related. My mobo utility tells me I'm 25C idle and 40C load, running on air. Maybe these temps are wrong.
Core temp is correct with the OC you're doing at that Voltage. I'm running 3.6 on my Wolfdale with 1.296v(per CPU-Z) and I'm running a 25C idle(Tcase, this is what your Mobo utility reports) and 35C idle under CoreTemp (which reports Tjunc1/Tjunc2) Under load in Prime95 it will peak at 70/71(Tjunc), but settle down after several hours and go into the 60s...and thats with me on an Intel HSF!!! Tcase for me peaks @ 60C, which is fine considering that throttling doesnt take place until 72.5C Tcase. 100% stable after just short of 12hours (overnight run) and i got 16,414pts on 3dMark06. im running the X38-DQ6 board, so i have better power regulation, so there may be a voltage drop on your PCIe bus that's causing the stutter. I wouldnt take too much stock in 3dMark "looks" because i average 230fps in HL2:Lost Coast and looks soo smooth that i can tell whoever made the flythrough didnt make a very smooth movement pattern at all
I ran the standard prime95 stress test, and checked the 'rounding errors' options as per the suggestion in the C2D guide.
I think the ram *should* be fine as I'm not overclocking it any. I'm running it at DDR-800 at 4-4-4-12.
My PSU is a cooler master Real Power Pro 750W, which should be more than enough for my single card set up.
Hmmm, you're right, at at first I thought it may be because the volts weren't high enough to support that speed and timing consistently, but if you were running them before at 4-4-4-12 and 800 then its most likely not the problem.
I'd first try the 1:1 memory ratio (on this board it is described as 2.00B (you can also try 2.00D although the 2.00B seems more stable for me)
Then up the FSB bit by bit. I had to hit the 1.2875 Vcore setting to remain stable at 3.6Ghz. (400Mhz FSB)
Remember to run Prime as you go to ensure everything is stable.
then benchmark when stable at 3.6Ghz
if you have DDR1200, try one of the 3.00 memory multipliers to hit your 1200DDR. However, you may get away with tighter memory timings with the 2x multiplier, so all you can do is bechmark and see which gives you better performance.
One thing I have discovered is the 1T command rate at 800DDR 4-4-4-12 doesn't seem to yield much performance increase in a benchmark, but games seem to be a little more responsive....
2. Up your CPU Host frequency. A nice starting point might be to go from 333Mhz to 350Mhz REMEMBER YOU NEED TO MONITOR YOUR CPU TEMP to avoid damaging anything. If you get problems, check temps, check vcore (more vcore required for higher speed HOWEVER more heat needs to be dissipated)
3. Turn Performance Enhance to STANDARD
4. Set your PCI Express to 100 Mhz (you don't want this to rise as you up your FSB)
5. Set your System Memory Multiplier as I described EARLIER (ie 2.00B) You can tinker with this To Your Hearts Content once you get the basic setup stable.
6. As your ram will be essentially underclocked initially, take the tightest timings the manufacturer recommends.
If I use the 1:1 ratio (2.00B), would I have to set my DDR to Auto or leave it at its 1200Mhz speed before the reboot?
I realize that I didn't SPECIFICALLY answer your question.
So, Now I shall, and then We Will All Be Happy
You don't actually set the DDR speed AS SUCH.
It's a product of the FSB and the memory multiplier.
So using the 1:1 ratio your memory will be running at DOUBLE (DDR, remember) your FSB speed.
1:1 (2.00B) at 333 Mhz = DDR-666
5:6 (2.40B) at 333 MHz = DDR-800
1:1 (2.00B) at 400 Mhz = DDR-800
2:3 (3.00B) at 400 Mhz = DDR-1200