Man this is the highest I've ever gotten stable... :\ Just under 13 hours prime stable... http://valid.x86-secret.com/show_oc.php?id=227164 I simply bumped the vcore up to 1.5v (didn't try to see if it'd run 1.45v - doubt it). 4GHz is inconsistent. :\ Still trying to find out why. Getting 50 damn MHz is a PITA!
I finally managed to find out why my E6850 is sometimes stable, sometimes not. It's the temperature! I just switched from aircooling to watercooling, and started playing around with my E6850 again. Moving to watercooling improved the temperature for at least 5C compared to Scythe Infinity with dual 2600+RPM fans. Latest E6850 overclocking results: #1 3906MHz @ 1.42-1.43v #2 4005MHz @ 1.46-1.47v Both are Prime95 small FFT test My E6850 is a weird one, it likes low temperature There are certain minimum temperature requirements for different speeds, it doesn't really need a lot of voltage bump. 3906MHz: <63C to be stable 4005MHz: <53C to be stable (Abit uguru readings) 51-53 range is impossible to achieve without aircon
Well, I think as is, this is probably the best I can do. Something is holding me back but I still can't figure out why. Prime95 small FFTs is grueling indeed. http://valid.x86-secret.com/show_oc.php?id=227164 Alas this is the end result. Perhaps temperature is the issue right now. I think if I had the GPU off the loop temps could be better. Either that or another rad. Voltage scaling is pretty nuts though. 1.45v @ 3.9GHz is prime stable 1.5v @ 3.95GHz is prime stable (would fail @ 1.45v) I've gone as high as 1.6v to try and get 4GHz stable but won't clear prime. Orthos on the other hand seems to be less strenous. I can do 4GHz @ 1.5v for 12 hours on their small FFTs. :\ Fails prime after 30 mins. Close but no dice I guess... :\
I still doubt my E6850@4GHz can survive longer/multiple runs of Prime95 small FFT test at that voltage though, I stopped after a few hours and turned off the aircon LOL, but at least that's a good sign, if it can survive for a few hours at 1.46-1.47v. I can't always do stress test @ 4GHz, because I had to turn on the aircon @ full blast to get some 1337 ambient temperature
Changed the tubings to 3/8" ID, got a 1-2C decrease in temperature lulz. Oh yeah, new radiator too decreased 3C with it. 3942MHz @ 1.43v My new 24/7 overclock. no voltage bump needed yay /me tries to summon D-Tek Fuzion
They already posted their replies to your question here - http://forums.techarp.com/reviews-a...duo-e6850-overclocking-report.html#post316359
Thanks for the help. I read the article, I was able to OC to 3.7(9x411) and I upped the volts to 1.45. I understand that I can now back down the volts until the system is running stable. I will have to learn more on the memory OC. But I have a question. I installed the PC Probe II and there is an alarm for the FSBT. It is currently at 1.46v. Is this a bad thing? What should it be running at? So far my temps are: CPU:23c MB:31c NB:37C SB:38C DRAM:2.05 Again thanks a lot for the help!
Could you please gimme all your system specs? If you could, please download CPUz and validate it (and get a link). CPU: Mobo: Graphics: Ram: PSU: CPU Link:CPUID CPUz Validation:CPUID It will give you a link like this: CPU-Z Validator Database Also please let me know the voltages for the CPU/RAM/Chipset. For more accurate temp readings use CoreTemp. Core Temp Welcome to TechARP forums!
On Saturday I was reading and tinkering with the bios and all of a sudden the bios screen was flashing blue and red. I shut down the system and reset everything back to default. So I was pretty nervous the entire weekend thinking I fried something! Ill try OCing the system again and post the stats. Also, I ran the orthos program while the system was running at 3.6ghz and it failed 7 minutes into the stress test. I guess I need to understand a little more about OCing. Here is my hardware list: CPU: e6850 CPU fan: Zalmans 9700 Mobo: maximus formula - 0505 bios Graphics: XFX GTX8800 Ram:4 GB (2x2) kingston hyper x ddr2-800 (pc-6400) PSU:cool master 1000w
When you increase the FSB, make sure you set the ram divider properly so that you ram will not run too fast. I see that you are using DDR2 800 and ~3.6GHz, you kinda run out of ram dividers (which is why I got some 1066 ram). That could be one of the reasons. This is because the ram speed is derived from the physical clock of the PC. So increasing the FSB of your processor increases the ram speed (need to switch to a lower multiplier to run at lower speeds).
If I understand you correctly...I should use 8x450 to get 3.6? Will this help correct the ram issue? Before I was using 9x400.
When I meant by FSB I meant the actual physical speed of the processor. ie: The stock FSB of a E6850 is 333MHz. With the 9x multiplier it runs at an effective rate of 3.00GHz. At stock with your stock ram (DDR2 800), the speed is derived from the 6:5 multiplier. (333x5)/6 = 400. Since its DDR so 400x2 = 800MHz. So each time you increase the FSB, your ram speed goes up. So lets say at 3.6GHz using the 9x multiplier. 400FSB. (400*6)/5 = 480MHz, DDR effective = 960MHz. That is 160MHz over its rated speed. Unless you have some really binned stuff, 'normal' ram will not have that over head (as proved by your PC crashing). The mobo I have does not have a divider lower than 6:5 so that is why I bought 1066 ram. 4GHz = 444FSB. For the 6:5 divider, its (444*6)/5 = 533Mhz, DDR Effective = 1066. The reapers have been binned pretty heavily (D9s that are meant to run at 800 I think) and has quite a bit more head room. Having a higher FSB will only hurt you more in this case (lower effective speed but higher FSB) as your ram is holding you back unless if you have a lower ram divider. Everything on the mobo is derived off of the FSB of the CPU. Ram, Bus speeds, etc. "Enthusiast' level boards usually have locks for most of them by default. THat is why a PCI lock (and PCIe lock) is important as the moment you start OCing the processor, the bus speeds go up as well and it can only go so far before it goes out of spec and things go hay wire.
Ok, I am trying to understand the mathematics of both the clocking of the cpu and memory. Correct me if I am wrong. So if I set the fsb to 333 then (333*6)/5 =399.6 which doubled would give me 799.2. Ok, so if I now have a mulitpler of 9 and my fsb is 333 then my cpu clock speed would remain 3.0. So how do you increase clock speed with out going over the memory limits. I am sure I am missing a piece to this puzzle.
Generally when one OC's the CPU, the ram speed will go up. Every ram module has a specific tolerance (again if its been binned for higher speeds - you have ram that has a higher tolerance or capability of overclocking). So as one increases the CPU's FSB to ultimately increase the CPU speed, we will choose a lower ram divider that would drop the speed of the ram. Think of it as shifting gears in a car. You up shift before the engine hits the redline (just in the context of ram though - in this case we drop the multiplier down so that the ram will run at speeds that it is capable of). There are a lot of ram dividers on a motherboard. I do not know what is the lowest one you have on your motherboard. You're going to have to play around and see. Drop the multiplier as you crank up your FSB (a calculator comes in real handy here!).
Is there a seperate multiplier on for memory and the cpu? Can you give an example of how I should set this up: If I wanted to clock to 3.6ghz: I would increase the fsb to 400 with a 9 multiplier. You stated the I should reduce the multipier. So should I use a 8 multiplier with an fsb of 450. This is where I get confused. If the fsb is now 450. using your equation(450*6)/5=540 DDR effective 1080. So if I did this that would mean that my memory would be over by 280mhz...Correct?
Ram divider and CPU multiplier are two different things. The ram divider is applied to the FSB of the processor. As its a derivative, any changes in the CPU's FSB will result in a change of the ram speed as well. So lets say you're on 6:5 ram divider. @ 400 FSB and 9x multiplier you are running at 3600MHz. For your ram it will be (400*6)/5 = 480MHz or 960 MHz effective DDR. 160MHz higher then the stock speed of 800MHz. So to answer your question, that would be correct, you would be over 280MHz in that example. Different ram has different overclocking headroom. Generally 'normal' ram does not have a big OCing headroom.
