Overclock.net › Forums › Intel › Intel CPUs › Low Voltage Overclocking: Is it all wrong?
New Posts  All Forums:Forum Nav:

Low Voltage Overclocking: Is it all wrong?

post #1 of 4
Thread Starter 
Before you turn on the flame throwers hear me out. I'm reading through lots of threads where everyone is very proud of their processors OC'ng at lower than usual average Vcores. Ultimately, I understand that the higher the voltages, the closer you are to reaching the thermal limit and reducing the life of the processor.

Here's my experience from playing around with what felt like two very different 3570k
  • #1 3570k, idle 26c, IBT load @ 4.4 GHz 76c 1.23v
  • #2 3570k, idle 34c, IBT load @ 4.4 GHz 78c 1.19v

I could never get that low voltage #2 3570k to be stable or post at anything higher than 4.4 GHz. But #1 would be able to post even though it sucked down more voltage. I was able to compare the performance of #1 and #2 on the same motherboard, rig and cooling setup.

The scenario that is getting me to question the real value of the voltage required to get an OC stable was when I was running IBT on #2 @ 4.2 GHz with Vcore on Auto and the draw as only 1.19v. I ran that up to 4.4 GHz, still on auto and voltage indicated was still 1.19v but the Gigaflops wasn't going up. I went to manual to force it to 1.25v and Gigaflops went up for that 4.4 GHz level just like it should. So, as great as it is to run low Vcore for a given OC, I don't know if most anyone is looking at the actual performance besides just the MHz indicated.

Your thoughts?
post #2 of 4
It is good to be STABLE at the least amount of voltage because it means less heat. Less heat means greater overclocking potential, usually. Real temp tells you the CPU's VID, and the lower the VID from the factory, USUALLY means a better clocking CPU. Just make sure it is 100% stable!
EVOLVE
(13 items)
 
  
CPUMotherboardGraphicsRAM
|i5 750-voltage may vary| |Foxconn Katana[Non GTI]| 380 4Gb |4Gb/Tridents@Clocked| 
Hard DriveOSMonitorPower
|[V3/60Gb][500Gb]| |Windows 7 64 bit| |LG-EW224| |Antec HCG750| 
Case
|Tech Station| 
  hide details  
Reply
EVOLVE
(13 items)
 
  
CPUMotherboardGraphicsRAM
|i5 750-voltage may vary| |Foxconn Katana[Non GTI]| 380 4Gb |4Gb/Tridents@Clocked| 
Hard DriveOSMonitorPower
|[V3/60Gb][500Gb]| |Windows 7 64 bit| |LG-EW224| |Antec HCG750| 
Case
|Tech Station| 
  hide details  
Reply
post #3 of 4
I've also seen GFlops increase a little with more vcore in experiments. The little IBT testing I did ran without errors with the lower vcore. I assume the CPU can catch some kind of errors internally. Perhaps that's where performance gets lost, the CPU repeating instructions that failed or something. I looked for WHEA warnings but nothing was logged. The only sign that something was off were the lower GFlops numbers.

All those experiments were definitely not enough vcore for stability in any case. Even after increasing vcore enough to get max GFlops numbers, there were still BSOD problems after two hours of prime95 etc. It ultimately needed some more vcore increase for that to clear up.
post #4 of 4
Thread Starter 
Deepor- Ultimately, we're only seeing what is actually reported to the OS, everything inside the die is just a guess. What good is a high clock if it isn't producing the calculations or performance? I don't know the theoretical mechanism here but one could guess that even though you have a high frequency, without the voltage, you aren't feeding that frequency to get the calculation throughput. It is like spinning up a water pump to max RPM but only running half the fluid volume through it. Yes, the pump is at max RPM, but if your goal is to pump fluid, then you don't have the throughput. Voltage=electrons, and a CPU is just an electron pump via switching operations right?

Why is this relevant? The mistake that most people make with DC is to assume that DC voltage is dead stable with no standard deviation. But DC still has a waveform. In fact, one of my friends that works with RF processors was telling me that one of their current programs is to grab power lost as heat in the minutia of peak to peak deviation in amplitude in DC current. How this relates to OC'ng might be simplified in these terms, you're indicating say 1.2V vcore, but the reality is that under microsecond sampling the mean is actually 1.18 vcore due to the waveform of the DC. And this is why you're unstable or getting low Gflops @ 1.2V. When you turn it up to 1.22V, it all gets stable because really, you're giving out a predictable and stable 1.2V, with the hope that the occasional 1.22v spike isn't generating too much heat.
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Intel CPUs
Overclock.net › Forums › Intel › Intel CPUs › Low Voltage Overclocking: Is it all wrong?