Troubleshooting Overclocked InstabilityIf ROG Realbench causes the system to crash (BSOD), reset, or gives an error when overclocked, it could be due to a number of factors. If the system is perfectly stable at stock, then overclocked stability is usually down to one of these:
- CPU instability – not enough voltage or CPU frequency too high.
- Memory Instability – memory overclocked too far, or CPU memory controller not capable of sustaining the memory clock being applied.
CPU InstabilityIf the CPU is unstable when overclocked using auto settings for Vcore, then manual changes may be necessary. A hint towards CPU related instability comes from looking at BSOD codes. If the BSOD gives a message related to clock interrupts this is very often Vcore or processor frequency related. Depending upon how much voltage is being used and the frequency the processor is being run at, there may be some room to increase Vcore to see if it helps improve stability. Of course one needs to be mindful of temps and how much voltage is being applied.
Open CPU-Z and run the Realbench stress test. Note the voltage being applied under full load; CPU-Z displays Vcore. Enter UEFI and increase the voltage by adding 0.05V to it. As an example, if we see 1.30V in CPU-Z when the processor is under full load, we would add 0.05V to that value in UEFI to give us 1.35V. Be careful not to mix up values here, do not confuse 0.05V with 0.5V. Increasing Vcore by 0.5V would be catastrophic for the CPU!
We are basing the adjustments in this guide on the usage of “Manual” Vcore not “Offset” or “Adaptive” modes. The latter two are more complicated to use for newcomers and can be dealt with after one gains some experience of the platform.
Manual mode is set by doing the following in UEFI:
We can save and exit UEFI and now run the stress test again. Again, keep an eye on temps under load. If everything is within desired operating range we can leave the system to complete the stress test. If it passes as a result of an increase of Vcore, great, we can move on and see how the system holds up to regular use for a few days.
If it does not pass, then we need to make a choice at this point. We can accept a lower overclock. Rule of thumb is not to increase Vcore more than 0.05V to go from one multiplier ratio to the next. Sticking to this rule ensures that we do not increase current through the processor substantially for little performance return. Let’s say our CPU is perfectly stable at 4.6GHz with 1.35V, and we attempt 4.7GHz, which is not stable even with 1.40V applied. In such a scenario, it makes sense to settle at 4.6GHz at 1.35V, instead of increasing voltage past 1.40V for 4.7GHz.
Some boards have a fully manual mode option - which allows all other voltages to be set directly without using offset/adaptive:
Setting this to enabled makes setting the voltages easier - offset and cache voltage is harder to predict without and one has to rely on monitoring and guesswork. Recommend enabling this option to start off with if your board has it. Do be sure to enter all the associated voltages manually (guidelines are given below for start points).
Cache voltage can also help, though I’d advise more caution with increasing that as much as Vcore. General advice is to stay below 1.30V.
My musings on running conservative voltages are geared to help you obtain an overclock that does not result in degradation of the processor in a short time. Given the cost of these Pentium CPUs however, it is attractive to do away with such rules and be gung-ho about voltages and degradation. If pushing the CPU as hard as possible appeals to you, then by all means do so. I’ve been doing the same here.
🙂It’s also possible that
both the memory and processor are unstable, which can be more difficult to debug (more on memory instability below). Usually, it’s a good idea to lower either the CPU speed or memory speed and see how that affects the instability. Lower CPU speed first and check if the system still crashes. If it does, lower the memory frequency too and check if that has any impact. By doing so, one can work out where the instability is coming from and work on getting the system stable.
Memory Instability
BSODs with codes of “Page fault in non-paged area” or “PFN list corrupt” are an indication of memory instability. A system may also lock-up or reset if the memory instability is severe. If the system is locking up, then running a memory only stress test such as Memtest or HCI Memtest (the latter is a pay for version which I personally use) can help pinpoint the cause and debug it.
There are several voltages associated with memory on the Haswell/Pentium architecture:
VCCIO-D
VCCIO-A
VCCSA
DRAM Voltage
And to a lesser extent cache voltage.
- Most of the time, the auto parameters for these voltages will scale the voltages as DRAM frequency is increased. Unfortunately, some CPU require more or less voltage than the auto rules apply so manual adjustments may be necessary. I’d advise to use smaller steps in making changes upwards or downwards from the auto value - steps of 0.02V.
- Conservative advice is to stay below 1.20V for VCCIO-D, VCCIO-A and VCCSA. Cache voltage below 1.30V.
- Recommendations for a manual start point for VCCIO-D, VCCIO-A and VCCSA is 1.10V if running memory speeds higher than DDR3-1866. If running memory speeds of DDR-2133 or higher, start with 1.15V. Work up or down from there.
- There is some additional info on why voltages such as VCCIO-D, VCCIO-A, VCCSA and DRAM voltage need to be reduced instead of increased in certain situations. Granted, “beginners”, should not encounter such things, however, it does not hurt to improve our understanding. CLICK HERE
- DRAM voltage should not be adjusted if using XMP, as the profile will automatically apply the specified voltage. If using kits with no XMP (an unlikely scenario), then manual changes to ensure the specified level of DRAM voltage for the memory kit is applied.
This should be enough to get started. More to come!