My adventures in Overclocking (Part 2)
I would just like to share this. Everything is based off from reading the excellent guides found at the First Page of this thread. I am new to overclocking and so far, it has been fun!
Achievements by following the guides:
- Manual Method of overclock (fixed voltage)
- Offset Method of overclock without Load-Line Calibration due to BIOS version restrictions (variable voltage, i.e. VCore voltage behaves "like stock" voltages where it scales down when idle)
Guides for Initial Overclocking: Please see First Page, First Post. :)
In this post, I would like to share my adventure into tweaking the CPU PLL Voltages as indicated in the First Page. :)
Important Note: prior to this, I have already "found" my overclock settings via the methods mentioned above. I would suggest, just as other experts here always say, to try the first two methods first. This gets you "acquainted" with how your processor responds to various VCore Voltages, and how the processor behaves in each method of overclock. Not technically a pre-requisite, but I would say "yes it is a pre-requisite" because:
- It lessens the guesswork (since you already know the optimal VCore voltage for your designed overclock).
- It reduces the number of variables you will need to tweak at the same time (typically it's VCCIO a.k.a. QPI/Vtt, VCore, and CPU PLL). Having already "found" your optimal voltage prior reduces the variables to just two of them.
- The approach is relatively methodical hence, less confusion.
Tip: like what the experts here say....always record, always take notes.
My Settings for a 4.5GHz Overclock:
- VCore as reported in BIOS: 1.320v
- DVID/Offset: +0.010v (resulting to 1.320v VCore as reported in BIOS)
- DRAM Voltage: +1.644v
- VCCIO (QPI/Vtt Voltage): +1.100v
NOTE: for #3 and #4, I initially selected XMP Profile, noted down the DRAM Voltage and QPI/Vtt Voltage and applied it manually instead in BIOS so that I can run my RAM at its stated settings of 1600MHz at 1.65v with a Command Rate of 2
- CPU PLL: to be discussed below
- Since I manually set my RAM as mentioned above, I may have eliminated one of the two variables (VCCIO, a.k.a. QPI/Vtt and CPU PLL are the two variables), so that leaves me with finding out the optimal CPU PLL Voltage.
- Per the guide in the First Page, I put my CPU PLL Voltage to 1.5v as the starting point (on my board, Gigabyte GA-P67A-UD7-B3, it was 1.520v).
- I ran Prime95 "Custom Blend". I ran two types of "Custom Blend": one with 1344K (min and max) and another with 1792K (min and max). Cycle Time is 1 minute. RAM: between 80% to 90%. Duration of test: at least 30 minutes, ideally 1 hour. See this post for an example.
- I initially got a BSOD 124.
- Reboot and go into BIOS.
- Adjust CPU PLL Voltage one "notch" (increment) higher than the previous one (in my case, previous was 1.520v, so I changed it to 1.540v).
- Save BIOS and reboot.
- Repeat Step 3.
- If you end up with BSOD 124 again like Step 4, repeat Steps 5 to 8. In my case, I observed that after each increase in CPU PLL Voltage, it took a longer time before I encountered a BSOD 124.
- What we are hoping to see here is that the BSOD will change from 124 to 101, because BSOD 101 is relatively a clearer indicator that you need to increase VCore...in other words, you may have already "found the sweet spot" for CPU PLL Voltage.
Explanation: as we can see, I started with 1.520v of CPU PLL Voltage. Initially I encountered BSOD 124. At each BSOD 124, I went into BIOS, increased CPU PLL Voltage by one "notch" and tested again. By the time I set it at 1.600v, I encountered BSOD again, the error message was no longer 124, but 101. :) So, on the next reboot to BIOS, I did not touch CPU PLL Voltage anymore but touched the VCore by moving it from +0.010v to +0.015v.
After I post this message, I will run Prime95 "Custom Blend" again to continue. I am anticipating that:
- It would take longer than my previous attempt before I see a BSOD.
- The BSOD may show 124 or 101. Either way, I would increase VCore. In my experience, I noticed that typically, when BlueScreenView application shows alternating 124 and 101 (as you increase VCore), it is getting closer to stability.
- I have "pre-defined" a limit to the VCore voltage addition that I would use. From my previous attempts, 1.344v was the highest to target, ideally should be 1.332v or 1.320v only (these were the voltages when I posted my official submissions).
- Without using LLC, these voltages give you (inclusive of drooping) as low as 1.296v, 1.284v, or 1.272v of VCore under full load (specifically, while stress-testing with Prime95).
Suggestion: in my experience, the effort involved in trying to push for getting 1.272v of VCore was not worth it as far as "net gains", i.e. normally we lower VCore with the objective of achieving lower temperatures. I have recorded my stress-test sessions where I got 1.296v, 1.284v, or 1.272v as the lowest-recorded VCore during the stress-tests, but the temperatures they all gave me were the same. At mild to moderate (4.5Ghz = moderate) overclocks, a 1-3 notch variance on VCore did not show me significant temperature advantages. Perhaps on higher overclocks this will be a very important factor (as well as usage of LLC). With that, I would say as a matter of suggestion, to just stay within that 1-3 notch variance, leaning towards which one gives you more stability.
If staying at the lower VCore (of the 1-3 notch variance) gives you good gains in temperature, then definitely go for it. It just so happened that in my experience i never saw it happening (unfortunately). The best "improvement" that I saw was 1 degree Celsius when I compared them, at the cost of time/effort/electricity bill. :D
Hope this helps! By the way I am not posting this to "reinvent the wheel" since there are already well-established documentation and guides in the First Page. My post is to merely confirm all those guides. :)
Just finished running Prime95 Custom Blend for 1344K and 1792K, each at one hour, and passed both without BSOD error messages.
Following my post above:
- My VCore Offset Voltage (at the BIOS) prior to this run is at +0.015, which is a BIOS-reported VCore of 1.332v (EDIT: but wavering still at 1.320v in BIOS). In the Operating System at full-load (during stress test), it was wavering between 1.292v and 1.284v.
- VCCIO (QPI/Vtt) is at 1.10v and DRAM Voltage at 1.644 as indicated above as well, to account for manual configuration in place of using XMP profile for the Corsair RAM.
- As far as Power-Saving Options, C1E and EIST are Enabled, and then C3/C6 and Thermal Throttling are on Auto.
For me, this is a good indication that I am right on track, following the guides from the First Page of the thread. :) Next step: 12-hour Prime95 Regular Blend...but I may need to schedule it for a later time. Will keep everyone posted!
P.S.: how do you run a 12-hour Prime95 Custom (not Regular) Blend? If you have screenshots on what to check/uncheck or numbers to put in, I would appreciate it. I still plan on doing Regular Blend though...just curious so I asked.
So I tried running a Prime95 "Custom Blend" Test and failed after 40 minutes. Following on from Update #1...
Edited by topet2k12001 - 5/28/13 at 8:59pm
- BSOD Failure Code is 124.
- From here I have 2 options: either increase CPU PLL Voltage from 1.600v to 1.6200v (next "notch" up), or...increase Voltage Offset from +0.015 to +0.020 because at +0.015 Offset, the BIOS-reported VCore was still switching/wavering between 1.320v and 1.332v.
- I will try CPU PLL Voltage first, test with Prime95 "Custom Blend" 1344K and 1792K for an hour each, and then move with Prime95 12-hour "Custom Blend" test.