With testing out of the way, it is time to see if this memory kit has any additional headroom. For Intel, we start off by using the XMP profile and increasing the frequency until the loss of system stability. After finding what can be accomplished without changing any of the timings or voltage, the second step can begin. This is where we go for the maximum frequency and lowest possible timings. Voltage modification above the XMP profile is allowed. After all, this is overclocking!
Intel's 11th Gen Intel Core processor paved the way for things to come. The introduction of the memory controller Gear Ratio allowed the system memory to run in synchronous 1:1 mode (Gear 1) with the CPU memory controller, or in a 2:1 ratio (Gear 2). With the release of Intel's 12th Gen Alder Lake based processors came DDR5 support and the additional 4:1 ratio (Gear 4).
It is generally considered that between 3600 and 4000 MT/s is the upper limit for Gear 1 support when using a Alder Lake (12th Gen Intel) CPU. This of course is partially dependent on the CPU memory controller and supporting voltages related to memory. In rare instances, higher-end motherboards can increase this slightly and offer better overall compatibility due to shorter trace length, higher PCB layer count and a better memory training algorithm. For instance the Intel Core i9-12900K used for these memory reviews maxes out at DDR4 4133 MT/s for single-rank memory. Achieving 4133 MT/s is quite a unlikely occurrence judging based by the sheer number of forum posts of many users struggling to get 3800 MT/s stabilized. It is safe to say that anything greater than 3600 MT/s using Gear 1 will often require a bit of hands-on tuning.
Since DDR5 has a higher operating frequency and a dual 32-bit data bus, synchronously operating it in 1:1 does not function at all. So far there have been no confirmed reports of this working for DDR5. That only leaves 2:1 ratio and above as a viable option for any DDR5-based setups. The motherboard should automatically switch to the 2:1 ratio for both DDR4 and DDR5 above 3600 MT/s. If all else fails, you can manually enforce Gear ratios in the BIOS as well. With this information on hand, we can deduce that Intel 12th generation processors using DDR5 in theory will benefit the most from the highest-possible frequency, until the 2:1 ratio is not longer possible. At that point the cycle starts over again with a new 4:1 ratio and even higher frequency system memory.
Those looking to overclock on a Intel platform will generally find a hard barrier around 6600 MT/s using the Intel Z690 platform and an Intel 12th Gen CPU. Switching to Intel 13th Gen, this value increases to 6800-7200 MT/s when using Intel Z690 motherboards and 7800-8000 MT/s for Intel Z790 4-slot motherboards. Special overclocking specific motherboards like ASUS Z690 Apex, Gigabyte Z690 Tachyon, MSI Z690 Unify-X, ASRock Z690 AQUA OC, and EVGA Z690 Dark can reach up to DDR5-8000+. While the new wave of Z790 versions start at DDR5-8000 and can reach up to DDR5-9000+ with exotic cooling such as LN2.
Caution is advised when raising DRAM voltage over the rated XMP profile. Direct airflow or a waterblock may be necessary for long-term stability. This extends to the CPU as well. Raising the integrated memory controller voltage (VDD2), System Agent (SA), and VDDQ_TX above Intel specifications may cause irreparable damage. Please proceed with care and do research before attempting this. Do not copy and paste values without understanding the impact first, especially if simply taken from screenshots posted on Discord or Reddit.
Overclocking can often be a challenging and rewarding as you traverse the unknown. Identifying what memory ICs are being used is a key component to a successful overclock. XPG is using SK Hynix A-Die here, which is still considered the best as we enter the second half of 2023. It will eventually be dethroned, as DDR5 is still relatively new in respects to the life-span of DDR4. In any case, whether it is SK Hynix A-Die, M-Die or an entirely new revision, the same steps are taken to find the perfect balance of frequency and primary timings. Once that is set, you can focus on the sub-timings to improve performance metrics further.
However, that all goes out the window when the memory doesn't place nice. Here, a few hours were wasted troubleshooting due to being overly confident and skipping the basics. The problem ended up caused by raising the voltage past 1.45 V, resulting in memory errors regardless if it was the 7200 MT/s XMP profile or 8000 MT/s. It is unknown if the memory is at fault or the motherboard has a bug with this particular APENC PMIC. In either case, once the culprit was found, a quick overclocking session ensued. Nothing too spectacular. While SK Hynix A-Die can certainly run 8200 MT/s at the same timings using only 1.45 V, not all ICs are created equal and this kit certainly needed 1.5 V to pass 8000 MT/s. Scaling back to 1.45 V, 7800 MT/s was the maximum value achieved. More adjustments to the sub-timings could be made, but were not attempted for this particular overclock.
While having a limit of 1.45 V is an inconvenience, the overclock was still successful. However, when the memory frequency was increased, so were the primary timings as well, negating some of the benefits that faster memory can provide. As always, it's important to point out that no overclock is ever guaranteed, which makes memory overclocking a new adventure every single time. This also means that the same settings may require more voltage for a different kit, or may not be possible at all depending on the motherboard, CPU and how well the memory performs beyond the rated XMP profile.