Hi guys,
I'm here because I have some questions about my ram that I would like to resolve.
My actual PC:
i7 14700KF
Corsair 2×16 4000 cl18-22-22-42 CMH32GX4M2K4000C18
RX 7900 XTX Gaming OC (GByte)
Corsair RM1000x (1000watt psu)
Asus TUF VG279QM
I used to have these ram.
cl16-19-19-36 3600 1.35v
I didn't know they failed, but they did. It wasn't until I upgraded from my old gtx970 to a 7900xtx, that I didn't realize they were crashing (xDefiant crashes every 5 minutes of gameplay).
And I didn't really realize it, it was by chance… After formatting the PC a thousand times, trying different drivers and ways to install things, different board bios, different OC and UV to the GPU and CPU… .
After two weeks I woke up so bored that I decided to do a memtest86.
I felt a great chill when I saw all those errors, and compared them with the hours and days I had wasted blaming other components.
It was as simple as first activating JDEC and seeing that there were no errors, and then activating xmp and lowering the freq to 3500.
I was able to play for hours without crashing (the loss of performance when using JDEC was brutal)
I decided to do RMA and bought these memories.
To my surprise, they didn't crash but I felt something strange. The memories had a lot of latency.
I went to the bios and I don't remember if I saw gear2 or commandrate2.
But I put both in 1 and let's play.
BOOOOM crash.
How could it be? Yes they were new and super stable.
From here on I don't want to expand further as if it were a novel. But the fact is that I was playing around and testing the latencies first, lowering them. But memtest86 kept failing.
I got tired and decided to make the XMP stable with gear 1 and Commandrate 1.
After three attempts I succeeded (the second was a false correct, it was stable until it crashed at the hour mark).
What I did was raise the ram voltage to 1.38v. Failed.
Then 1.40v (this time I also set the bios memory training to 1.40v). Failed.
And finally: 1.43v ram and 1.43v training, in addition to raising the VDDQ to 1.25v (thanks to a ram guide (credits) that suggested raising the VCCIO).
After this, it has been a rock, playing and in memtest86, even having Gear1 and CommandRate1 (taking into account that from 3750 Mhz in ram, command rate 2 is usually set,or so I understood). I can see BRUTAL immediacy in everything.
I play on a 280hz monitor and I have all Windows (ioT 2021) debloated with NTLITE, both features and components.
As you can see, I prioritize latency a lot.
And my doubts are:
1 – Obviously it is clear that I have only touched the VDDQ as a last resort and I don't know whether to attribute the stability to those additional 0.3 in the ram, or to the 1.25 of the VDDQ, but that was two hours ago (the successful tests) and I didn't want to test but for today. Should I lower the ram voltage?
2 – Is it wrong to have forced the xmp [VS] Try to lower the latencies more by hand?
Taking into account that I am the child of the PC without latency (it is a hobby)
What can you tell me? I really have little experience with RAM OC, and i having 3600 before, idc if to have less freq to be able to reduce the CL is worth?
Thank u for Reading.
1 Comment
The dram voltage may not have mattered, but for DDR4 on 12-14th gen it is also supplying VDD2 afaik.
The IVR Tx VDDQ (aka CPU VDDQ) is very important, one has to find the right value for their combo. The window of passable values seems to narrow as ram frequency goes up.
Otherwise SA voltage seems to matter for IMC speed, but default bios values are often pretty generous here.
You will want to probably test with some other software as well (like TM5, y-cruncher, Karhu Ram Test, etc). But at least as important as that is testing across multiple retrains from cold boot to make sure it trains consistently stable.