Originally Posted by Mahawka
I have been doing some overclocking with my i52500k. I am quite new to OC so I just have a few questions.
Right now i am running at 1.355 Vcore at 4.7Ghz. I feel like i have a decent chip compared to the research i have done.
I have run Prime for 5 hours and it has not crashed yet.
Is 1.355 a safe Vcore for 24/7, if yes what's the max vcore for both 24/7 and benching.
Will i get a better OC if i OC both Multiplier and Base clock Vs just Multiplier
Do i have to overclock the Base Clock in order to OC my Ram? (corsair vengeance 1600mhz)
are there any rule of thumbs i should follow when i OC my 2500k
Thanks a bunch.
1.355V is well-within design spec for the 2500K and I'd feel very safe there. My 4.7 GHz 100% CPU load voltage is 1.336V. Although no two CPUs are alike, you sound like you're in the ballpark.
Intel whitepapers list the maximum VID as 1.520V. (Keep in mind that VID is not the same as Vcore, but you can see what your current VID is if you use programs like RealTemp.) Source: page 80 of this http://www.intel.com/content/www/us/en/processors/core/2nd-gen-core-desktop-vol-1-datasheet.html
However, each processor has its own pre-programmed maximum VID which may be above or below that value and VID is sort of like the Vcore demand, it's what the processor is "asking" for, but not what it gets. Droop comes into play and the VID may be 1.500V but the Vcore setting might be at 1.450V and the actual Vcore reading at that moment may be 1.350V. Then, during an unload or load transient, the Vcore will spike up and down. If it spikes too high, it can damage the core or cause a BSOD and if it spikes too low, it can cause a BSOD as well. Changing the LLC (load line calibration) can reduce the chance of an undervolt BSOD but can also cause your high spikes to exceed maximums more easily.
I personally feel that it's best if you keep the Vcore (not VID) at an absolute maximum of around 1.420V or lower, as that's safe enough for 24/7 use with adequate cooling to not cause too many issues. My own preference is to never exceed 1.400V in any situation.
Changing BCLK may increase your memory frequencies some which may give you a "better OC" in terms of memory bandwidth, but most people have found that a BCLK of +/- 5 is about as far as you can go. (95 to 105). Going to 106 and beyond usually introduces many instability problems to both memory and CPU and can even damage SATA components in some cases. Thus it's advised to keep your BCLK between 100 and 105.
That being said, I prefer to keep BCLK at 100 and just do all of my overclocking via the multiplier. I think that 100 MHz increments are good enough and that there's no reason, if you're stable at 4800 MHz to push it up to 4850 MHz if it is crashing at 4900 MHz. Again, that's just my opinion.
You may be able to enable the XMP profile in many BIOS to overclock your RAM to 1833 values or just set the memory frequency manually to 1833 MHz and then edit your timings accordingly for stability. However, your motherboard may not boot or this may introduce instability to your system. Since overclocking RAM and CPU for the first time simultaneously would be too many variables to handle, it would be best to leave BCLK at 100, get your CPU stable at whatever frequency you can, record all of your settings that it took to get stable there, and only then overclock your RAM manually or start messing with the BCLK to overclock your RAM (and adjust your CPU multiplier down accordingly so that the CPU net frequency is the same, or lower than your last stable OC without the RAM overclock.) I've found that RAM errors are easy to cause with these chipsets, even with great DIMMs.
Huh, well, the obvious stuff, like keep your memory voltage close to 1.50V and not the 1.65V that the older boards had. Install your memory in true dual-channel mode in the preferred slots (read your manual to find out if it's 2 and 4 or 1 and 3) and also don't fully populate your DIMM slots. 2x4 = 8 GB of RAM will out-perform 4x4 = 16 GB of RAM in all gaming situations and the majority of computational situations, unless you're doing heavy video encoding, heavy 3d animation/video production, or editing huge image files, like satellite images or manipulating very large autoCAD or solidworks projects.
I consider my rules of thumb to be max core voltage 1.400V, max temperature in Intel Burn Test (the hottest thing out there) of 80C after five consecutive passes on maximum, and ability to pass memtest386+. I also like to use Prime95 to test for CPU stability. The 1344 and 1792 FFTs seem to be the most troublesome, and 30 minutes on each is a fairly good indication of stability, if you customize your blend to use 90% of total system memory.
Just my thoughts:
I personally like to run a Prime95 blend test after I've passed all of the above and change the blend settings to custom to use 90% of total system memory and run it for 18 hours, so that it tests every FFT. Then, once I've done that, I can be sure that both my CPU and memory are stable.
Once I'm 18 hour blend 90% RAM stable and I know that memtest386+ is passing and IBT at 5 runs maximum has temperatures below 80C I will go in and fine-tune my OC to reduce voltages as much as possible with the lowest LLC setting possible. My first OC at 4.7 GHz was 1.360 to 1.368V and I had the LLC at 75% or so. With these settings I was hitting thermal and/or current/power throttles silently. Even though my CPU frequencies stayed at 4.7 GHz throughout the IBT tests and my temperatures were 85C, which was well below TJMax (which is 98C!!!), I was noticing that my GFlop/s were inconsistent, as low as 122 in some tests and as high as 125 in others.
So I reduced voltage significantly by changing my offset voltage (I like speedstep enabled and C1E, C3, and C6 reporting enabled) and my Vcore came down accordingly. I did this until I got BSOD 101 errors from undervoltage and saw that my GFlop/s had increased some. So I struck out to get stable with the highest GFlop/s possible in IBT, which is the most power draw and temperature abuse I can find in a test. I was able to reduce my LLC to only 25% instead of 75% which increased my GFlop/s and then I was able to reduce my voltage from the 1.368 down to my current 1.336. The two changes combined reduced my temperatures in IBT between 5 and 10C and increased my GFlop/s from a variable 122 to 125 to a very consistent 125.5 to 126.0. My benchmarks in PCMark and the PhysX scores in 3DMark11 increased accordingly.
Check out the Sandy Super Stable club and P67 / Z68 owners club links in my signature if you want to read even more detail about settings and information on your motherboard (which you didn't list in your post).