How far can I push this DDR4 memory?
How far can I push this DDR4 memory?
I’ve been managing my own hardware setup, focusing on self-reliance. The current configuration gave me decent performance, but not the best. I considered mixing components between machines to create a compatible group. I found a collection of four Kingston ACR24D4U7S8MB-8 (DDR4-2400) chips, Micron 6TB75 D9TBH models from late 2016 to early 2017. How much potential lies in them? What specifications should I aim for? I’m beginning with my existing DDR4-3200 (133x24) 16-19-19-52-1 at 1.35v and gear 1. If they’re matched, what would be the next targets? The motherboard is a Z790 with a 13700F chipset.
This IC is a Micron 8Gb revision B, distinct from the 16Gb version. The documentation for Rev. B is similar to Rev. E but lacks the high-frequency headroom you typically see. Most users report speeds near 3200MT/s, so you likely won’t achieve much higher speeds (3600 might be the upper limit, though that could introduce IMC challenges). You may still experiment, but the main risk is needing to reset CMOS. Adjustments to tRCD, tRP, and tRAS can be made slightly lower, giving your primary timings around 16-18-18-36 or 16-17-17-34 depending on your setup (the looser kits may allow more flexibility). Voltage scaling is uncertain due to limited data, so sticking with the 1.35V supply is advisable. If you decide to fine-tune sub-timings, search online for Rev. E settings and test incrementally.
Input the printed code (D9TBH) into the FBGA decoder tool. It will display the datasheet for that specific IC. The final digit in the full part number indicates the die revision. I think it might not be worth the time, though. I’d consider testing higher frequencies just in case, but I wouldn’t invest effort unless it stabilizes quickly. If you need further adjustments, set tREFI to 65000 for better memory performance and stability.
I tested two D9TBH ICs in dual channel—one Kingston and one Micron OEM. The results matched DDR4-3600 CL18 kits (typically 18-22-22-42), which seems acceptable. At 3733V the system was unstable, so no further adjustments could be made. AIDA64 benchmarks passed, indicating a solid start. For future builds, consider using four matched sticks tomorrow. The DDR4 overclocking guide you referenced suggests older Micron chips can be scaled down beyond 1.35V, but only up to around 1.40V. It also recommends a tCCDL of 8 for Intel, which helped improve performance by about 19.5% bandwidth. A 60ns latency appears reasonable for this configuration.
It's worth noting that most references mention 4Gb ICs, but I've also encountered Samsung B die, which can handle voltages above 1.7V. These guidelines aren't always reliable since each kit differs. If your setup needs 3600MT/s CL18-21-21, you might aim for 3533 with four sticks based on your IMC quality—still a solid result. For memory stress testing on the 13th generation, my successful runs were using Y-Cruncher VT3 (though it sometimes fails at frequency) and TestMem5 (Anta777 Extreme1 preset works well for DDR4, though some prefer Absolut or 1usmus_v3). I’d allocate an hour with VT3 and four hours with TestMem5 (roughly eight loops of Extreme1, even if it's been a while since I tuned 13th gen DDR4). If you're interested, there’s a 8Gb C die from Samsung that consistently delivers 3600CL18-22-22-42 with minimal variation. Micron 8Gb Rev. B seems comparable, though it might support CL16 with better sticks.
Today I own four Kingston ACR24D4U7S8MB-8 DDR4-2400 sticks. Two are dated week 45 of 2016 and two from week 15 of 2017. Three units support DDR4-3600 (133x27) with serial numbers 18-21-21-38-1 at 1.35V, while gear 1 was set up yesterday. One unit won’t work with the system. The best performance it achieves is around 3500 at week 15 of 2017, and even that drops when combined with others.
Looking at voltage levels, a baseline of 3200 at 1.40V works, 3400 also at 1.40V, but 3500 fails at 1.37V. The 3600 unit only boots partially at 1.35V and fails at 1.37V and 1.33V. This suggests a negative relationship between voltage and stability. I’m leaning toward the idea that the Samsung B-die chips behave negatively, but since I had only one of those, I replaced it with a Micron OEM unit that ran fine at 3600 the previous night. I’ll continue testing and consider swapping the problematic stick for the new one tomorrow.
I have four DDR4-3600 modules (133x27) with voltages set at 18-21-21-38-1 at 1.37v. I increased the voltage from 1.35v after a crash yesterday. Based on negative stability trends, this might help with the second stick's higher capacitance per channel. I'm not sure if it will work perfectly. So far, AIDA64 benchmarks passed without any crashes during sleep, though I'm not entirely confident. My AX210 WiFi stopped working completely, requiring a driver reinstall to get it running again. I've noticed Chrome tabs crashing occasionally, which isn't common for Google products. What adjustments could improve stability?