How far can I push this DDR4 memory?
How far can I push this DDR4 memory?
That was just now. We'll monitor the outcome. Do you recall which BIOS setting locks the base clock to precise 100MHz versus something like 99.8? It sounds like a PLL control. I saw SkatterBencher demonstrate it in one of the videos, but I can't remember the details. [EDIT] I didn't locate any BCLK options at all—it seems this board doesn't support them, or they're not visible for non-K processors. MAG Z790 Tomahawk Wifi DDR4...
It should turn off spread spectrum, but based on my observations unless the circuit includes an external clock generator, it often fails to raise it from 99.8MHz.
Through several elimination steps I determined this value isn’t consistent with my memory, board, or processor setup. Do you have a method or formula to determine a better tREFI setting? Currently I’m using the default values the board provides so I can complete memory-intensive tests. Thanks, it looks like a strong challenge. I’m struggling with the DDR4-3600 configuration—it only holds for a few minutes before errors appear. So far I’m only getting results at 1.37v with DDR4-3466 and specific timings. It also supports CL17 up to 3400 but not beyond. The system boots into CL15 at 3400, but not higher. 3500 is the least reliable, making it hard to boot into Windows. 3600 offers more stability—it can run for a full day (though I didn’t let it go longer than that), and it will download a 100GB torrent with only a few damaged files. You get the point. However, it still fails the overall test. Here are some observations: voltage above 3200 at 1.37v is the most stable for four sticks; 1.38v or lower won’t work. Stability drops below 1.35v and 1.36v. This memory seems to prioritize a command rate of 1T and CL18 (no higher). tRCD and tRP values over 22 tend to be less stable until around 30. There appears to be a reason behind the preference for longer timing delays—possibly related to PLL tuning or clock spread. Any suggestions on improving the 3600 setting? Yes, I understand that getting a perfect PLL value isn’t always possible. It seems my BIOS lacks advanced options like BCLK and spread spectrum. Forum users recommended updating the BIOS. This might be a last resort—I’ll consider it once I have a better undervolt setup and verified configs are saved.
The system really doesn't have a clear solution. TREFI measures how often memory refreshes occur, and since RAM chips need recharging after time, errors often appear during stress tests due to temperature effects. I don’t adjust settings finely; instead, I test around 262k (the memory limit) with cooling present, then about 131k without it, and keep reducing until it stabilizes. I haven’t found a kit that reliably works at 65k, but some do. I’d try lowering it to roughly 32k and see, or just let it auto—auto is thought to be around 12k. Adjusting the CPU VDDQ can help sometimes, though the lack of VCCSA control is a big hurdle. The VDDQ value you use depends heavily on the board and BIOS version; for example, MSI boards often prefer 1.1V or 1.45V, while 13th gen controllers have quirks where certain voltages perform better at specific timings. Intel’s 1.33 multiplier (or strap) tends to be more stable at high speeds than the 1.00 strap, and 4100MHz chips run smoothly at 4133MT/s, whereas 4100 is quite unstable.
The board is currently around 13.5k. Interesting, I'll check those figures later once I locate the stable data rate. System voltage settings? You have "CPU SA Voltage Mode" and "CPU SA Voltage." I just began with VDDQ at 1.35v and an offset of +0.1v on SA (0.934 + 0.1). It looked like that caused instability, preventing a boot into Windows. I switched to 1.37v for both, without any clear reason. During TestMem5 it lasted slightly less than before. I'll go with 1.45v for both now, as it seems safe. That matches what happened on the 11th gen system you assisted me with last year, I believe.
Sure, system agent. The non-K LGA 1700 CPUs keep their SA voltage at 1V or less, meaning even if a motherboard allows a higher setting, it will always round down to 1V.
You're on the right track. Checking HWMonitor confirms SA 0.934v, which suggests the system didn't recognize your settings. Trying 1.0v might help, but VDDQ 1.45v caused a BSOD at startup.
It appears some CPUs and motherboards have SA readings that behave oddly, especially when voltage adjustments aren't reflected unless you cleared the CMOS first. Recent BIOS updates generally favor lower VDDQ values, so interpret those changes accordingly. I don’t know your exact BIOS version, but 1.35V was typically ideal for older systems—only a few required 1.45V.
I performed a CMOS reset and quickly entered the BIOS. I loaded the profile, adjusted the SA offset to +0.05v (+50mV), but the hardware monitor still showed the default 0.934v instead of the expected 0.984v. The board’s default VDDQ is 1.2v. A previous issue had 1.4v causing a BSOD during boot, and earlier it was 1.1v.
It might not be a Clear CMOS that does it, I forget how I've gotten my boards reporting correctly after changing it. There is a way to do it, but I'm too tired to spend the hour troubleshooting on my Unify-X to double check. It should be running at 1V though, which is the highest it works at. Try 1.3V VDDQ and see if that helps much. That said, VDDQ is probably not the limiting voltage, and therefore won't help you get high memory speed to work. It's also worth noting that there is another you could do, tuning termination settings manually, but generally that takes hours, if not days to dial them in by pure trial and error, and there's no guarantee that it would even help. If you value your time at all, I wouldn't do this.