cancel
Showing results for 
Search instead for 
Did you mean: 

STRIX x470-I Gaming can't make 3200 MHz ram stable

dalaeck
Level 7
Hi, some help from an expert please!

I'm an X470-I owner and I've been trying for nearly a week now to get my 32 GB of ram at the rated CL to not give errors in memtestx86.
My RAM is F4-3200C14D-32GTZ. It's not QVL but it's "high quality" Samsung B-die rated at 14-14-14-14-34 (1.35 V).

Here is the problem: The RAM is very close to stable, just barely not and I can't get it to fully complete memtestx86 without errors.
It's always a few errors, test 7 and 9 fail, but only on repeated passes, it can also just plain pass the test quite regularly.
The system is so close to stable it will pass realbench, aida64 for hours, and stresssaptest likewise.
But the system is NOT stable, it will crash, on the timescale of a day. So I trust only memtest86.

RAM works flawlessly with 16-16-16-40 and passes all memtest 86 no errors, 3200 MHz and base frequency. So it's not bad RAM.
In fact the problem is that there's no way to find any sort of sweetspot for anything. It'll post and boot and run 3200 MHz CL 14 at low voltages, like 1.29 V and high voltages likewise.
It always small, single digits number of errors in memtest.

I've tried many, many things now and I just can't get it to run CL-14 3200.

What I've tried:
1-D.O.C.P and adjusting up and down the soc and ram voltages
2-Ryzen DRAM Calculator safe, fast and extreme presets
3-Vipeax's settings from http://www.overclock.net/forum/27487784-post1636.html
4- harrysun's settings from http://www.overclock.net/forum/11-amd-motherboards/1635742-ryzen-memory-support-duel-rank-2-x-f4-320...
5- From this forum https://rog.asus.com/forum/showthread.php?101707-16GB-RAM-modules-for-Crosshair-VII-2700X/page2#post...

And I've likewise tweeked voltages and sub-timing settings in all the above. I set the timings loose or tight, I adjust sub-timings,
it's always the same result, a close-to-stable but still errors . It's very frustrating, crazy frustrating.
Is it time to give up? There's something just odd, because it's so close to stable over such a wide range, I can't pin anything down.

Any help much obliged,
9,207 Views
10 REPLIES 10

dalaeck
Level 7
Well 50 views but no replies, so here's an update that might add something.

74254

I can get 3000 CL 14 with fairly aggressive sub-timings completely stable, passes all tests.
These are essentially the same sub-timing I've input to my attempts to make 3200 CL 14 stable.

Is it possible that the memory training is different at 3200 vs 3000?
Because the only time I've seen more errors in memtest is when I clear the BIOS via shorting the pins, input aggressive timings first and then dial back towards looser timings.
And likewise if I start with loose timings and then try for more aggressive.
Neither loose nor aggressive once again are initially truly stable.

I know there are memory training options, but I've spent the better part of a week learning the black-box of sub-timings, the memory training looks like a black hole...

Anyhow, any help much obliged.

aaraujo
Level 7
What CPU are you using?

Sorry, it was mentioned in linked threads but I forgot to do so here.

It's a 2700, not X, plain 2700. Because of low TDP, my (tiny) rig cannot take the thermals of a 2700X.
Right now the CPU is overclocked to 3475 and undervolted to 1.1 V.
I have also run it stock, the memtest errors are not a result of the overclocking or undervolting.

aaraujo
Level 7
Did you try raising the SOC voltage?
I have the same motherboard and I have the SOC voltage at 1.05V. The CPU is 2700X. I am running a 8GBx2 kit at 3600Mhz with the fast timings from ryzen DRAM calculator. The memory is also Samsung B die. The RAM voltage is set to 1.405V as suggested by the calculator. It went over 4 memtest passes with any error!
Besides the SoC and RAM voltages I am using ProcODT at 60 ohm. I have tried 53 Ohm with success. Also disabled "Gear Down" and "Power Down" as suggested here
https://rog.asus.com/forum/showthread.php?93722-Flare-X-3200MHZ-16GB-2x8-F4-3200C14D-16GFX-Performan...
Everything else is left at AUTO.

Thanks, yeah I have raised the SOC voltage, when I let it go auto it sets to 1.1375. I've brought it manually to 1.4, which is the max suggested.
It doesn't seem to do anything unfortunately.
Likewise I tried boosting many times procODT , I've stopped short of 80 because of potential thermal problems.
Geared down mode and power saving I have disabled, in my stable CL16 or stable 3000 CL14 though neither of these settings seem to impact errors, there just are none.
CmDT likewise, 1T or 2T, no net effect. 😞
Looking at the guide you linked, the only thing I haven't done is set vboot, I'm not even sure I have that option in the newer software. It has no problem booting, like ever. I don't want to damage anything so I haven't tried crazy/extreme options but it has never failed to boot 3200 CL 14. It just never passes prolonged memtest....

aaraujo
Level 7
There is no vboot option. But I don't think it would make any difference. It is needed to boot only. I think 1.4V SoC is too much. Mine goes to around 1.137 when set to auto. But I don't need that much. It works very fine at 1.05V! Your RAM is a 2x16GB kit. I remember that there is a difference with dual rank memory not overclocking very well. Is it single or dual rank?

PS: What software is that from the picture above?

It is definitely dual rank and it does run at much lower vsoc, as the snippet shows 1.025 seems fine.
The program is Ryzen timing checker, it's freeware and quite useful!

aaraujo
Level 7
Tested with 1.025V soc voltage and it passes memtest 4 passes without a single error. But fails CPU and Memory stress test under AIDA64 in less than 10 minutes. Raised SOC Voltage to 1.1V, and it ran AIDA64 System Stability Test for 30 minutes until I stooped it. So definitely 1.025V SOC is not enough! Nor 1.05V. At least I tried 1.025V, 1.031V, and 1.05V, and all failed under AIDA64! 1.1V SOC is fine so far. But I have to run the test for several hours at least, to be confident.

Yep, I've seen similar. I can get my kit to run stable 3200 but only CL16 at SOC 1.05 V. I'm still struggling with the CL14...
Again it's test 7 & 9 in memtest that fail.
I've noticed that the largest gain in latency from sub-timings comes from adjusting tRFC (and associated 2 and 4) . Pretty much all of the other sub-timings are very small, incremental improvement compared to it. I'm doing this laborious comparison using the memtest latency it shows before you run the tests and double checking with Intelmemchecker.

Unless I get some additional advice I think increasing the tRFC is kind of my last shot,