Getting RAM to 3200 on AMD 1700 & Asus Prime X-370 Pro

[GunnerMan]

I have spent a pretty good chunk of time trying to get my RAM speed up from stock for some time now. I have read various OC guides and have used the AMD Timing tools.

The highest 100% stable I seem to be able to achieve is 3066.

My RAM is in my sig, its Samsung B-die, single rank, rated for 3200 at 15-15-15-35 1.35v.



I have SOC Voltage at 1.025v, DRAM Voltage at 1.35v, VTTDDR at 0.670.

It passes 4 hours of multithreaded memtest86 using all tests but #13 hammer test as well as 4 hours of the memtest (HCL?) provided in the Ryzen DRAM calculator tool. However, it will fail Prime95 large FFT test after 30 - 90 minutes or so or BSOD every so often in games. It was much less stable until I increased tRFC to 592 from 408.

I have tried just about every combo of ram voltage from 1.33 to 1.4 and SOC from 1.01 to 1.1 as well as the various procODT and RTT settings as recommended by the timing calculator. So far it seems to be happiest with the posted settings.

I was kind of wondering if it may be heat related since Prime95 seems to heat up things the most, about 65C CPU and it does not seem to fail when I reduce the number of test threads from 16 to 15 or 14 but then again, I am not overvolting or overclocking cpu and despite stock cooler my airflow is pretty decent I think, motherboard seems to hover around 35C under load.

I am kind of at a loss as to what to fiddle with next. It took me a fair bit to get this far. Most of my tweaking before resulted in no-boot or memtest errors right away. Do you think I may just be at a wall with my ram and cpu and should just focus on tightening timings at 3066? I seem so close to 3133 though. This great write-up,Demystifying Memory Overclocking on Ryzen on Reddit helped immensely.

 

[Johan45]

First question is have you updated to the latest BIOS for that board?

 

[GunnerMan]

First question is have you updated to the latest BIOS for that board?

Yes. I am running 5220 released on 2019/09/24.

 

[Johan45]

It's hard to say at this point but looking at your timings I do find some of them a bit off. Try saving your profile in BIOS then hit F5 to reset boot back to BIOS don't use DOCP/XMP but input manually. I have been using these timings at 3333 but try 3200 with the same settings. If you start with CL14 then GD can stay disabled and that has been known to cause some people issues.

 

[GunnerMan]

Thanks.

I tested those settings with DRAMv at 1.35v and SOCv at 1.025v. It forced me to Windows and it BSOD almost immediately after login. Rebooted into Memtest. Failed Memtest86 starting at test #6 Block move 64 byte. Passes earlier tests multiple passes fine. Increase DRAMv in 0.005v increments up to 1.365v and saw significant stability improvement (but still far from stable). It passed all Memtest86 tests so I booted and ran HCL and it errored within a few minutes.

I reduced RAM down to 3133Mhz and ran HCL. It errored after after 5 mins or so as well. Increasing DRAMv up to 1.4v had no effect. Increasing SOC up to default 1.1v also increased stability quite a bit here but it still errored in HCL rather quickly still. Seems to be stable at 3066 Mhz.

I increased tRCDRD to 15, set ram freq back to 3133, as well as set ram and soc voltages back to 1.35v and 1.025v respectively. Better stability than before but still failing after about 15 mins or so. I increased tRCDRD simply because it was a primary timing that, as I understand, does not directly effect other timings, can have a significant impact on stability, and the Ryzen Calculator recommended to be higher than 14 so it seemed like a good one to test first.

I noticed something off with the SOC voltage reporting and I don't know what to believe here. The soc voltage reported in Ryzen Timing Checker and HWiNFO both report the the vsoc is about 0.013v less than what it is set to. I increase the soc to 1.05 in bios (1.037 reported) and it passed a couple hours of HCL and ran normal usage for a few hours at 3133Mhz and tRCDRD at 15. Battlefield V crashed after about 30 mins of play though.

Most say that a SOC of about 1.025v should be plenty for 3200 but I am kind of concerned I was too low with it all along if the reporting tools are to be believed.

 

[Johan45]

I just checked mine in HWInfo64 and it reports 1.137V IIRC I had it set to 1.15V in BIOS. SOC is fine up to 1.2V

 

[PolRoger]

Thanks.

I tested those settings with DRAMv at 1.35v and SOCv at 1.025v. It forced me to Windows and it BSOD almost immediately after login. Rebooted into Memtest. Failed Memtest86 starting at test #6 Block move 64 byte. Passes earlier tests multiple passes fine. Increase DRAMv in 0.005v increments up to 1.365v and saw significant stability improvement (but still far from stable). It passed all Memtest86 tests so I booted and ran HCL and it errored within a few minutes.

I reduced RAM down to 3133Mhz and ran HCL. It errored after after 5 mins or so as well. Increasing DRAMv up to 1.4v had no effect. Increasing SOC up to default 1.1v also increased stability quite a bit here but it still errored in HCL rather quickly still. Seems to be stable at 3066 Mhz.

Your kit should be B-die but did you confirm with Thaiphoon Burner?



I have a somewhat similar setup with a Ryzen 1700X and ASUS Prime X470-Pro and I was able to boot and run OK with the "Johan45" settings and my TridentZ (1R) 3600C16 kit.

"Johan45" settings:



Maybe you should try/test your combo @3200 MT/s with Primary settings @16-16-16-16-36 and tWCL 16 and all other timings left to BIOS auto defaults? I would try to dial in settings for memory at C16 first before trying to lower tCAS any further.

3200C16 tWCL 16 and all auto/defaults:



"PiLsY" 3200C16 settings:



"PolRoger" 3200C15 settings (CPU Auto) with DRAM ~1.35v and SOC ~1.025v:

 

[GunnerMan]

Your kit should be B-die but did you confirm with Thaiphoon Burner?

I have confirmed with Thaiphoon.


I have also tried quite a bit to get stable with 16-16-16-36 and auto (though not sure what I had tWCL at) and had similar struggles. I could get it to seemingly approach stability but never could quite get it there. I will try it out again following your tWCL and other recommendations as well as play with a bit higher vsoc.

I tried with Johan45's settings because I had tried from XMP timings up to very loose never made much progress. I always figured I was probably missing something fundamental.

Thanks for the help.

 

[PolRoger]

I always figured I was probably missing something fundamental.

You may not be missing anything fundamental... It could be just something ??? about your particular combo... CPU IMC; motherboard/BIOS or even the DRAM kit.

I have an early sample DDR4 (Z-170 era) Crucial Ballistix Sport kit with 2x8GB (2R) sticks... Micron A-die? That runs at default 2400 MT/s and that about it for this combo. Anytime I tried pushing higher... Major instability/incompatibility.

Try those straight 16 with tWCL 16 and also maybe try those C16 "PiLsY" settings originally configured for a 2x16GB (2R) B-die kit but I found them to work nicely with my single-sided B-die kits. If you can't get 3200 MT/s C16 dialed in? Then maybe go with something lower... ~3000 MT/s speed and tighten up both the primary and secondary timings.


I tweaked/tightened up some more on my previous 3200C15 settings:

 

[Zerileous]

It could simply be the CPU and /or board doesn't like to do it. As Johan suggested more vSOC may help. If you prefer not to run that voltage 24/7 that's fine, but at least it will demonstrate whether the CPU is holding you back or something else.

Also not all b-die is created equal. There must have been some reason that they rated it at 15 instead of 14, however anecdotally 1 series Ryzens do not do well with odd primary timings. So I think you're on the right track trying 14 and 16 based timings.

 

[GunnerMan]

It could simply be the CPU and /or board doesn't like to do it. As Johan suggested more vSOC may help. If you prefer not to run that voltage 24/7 that's fine, but at least it will demonstrate whether the CPU is holding you back or something else.

Yeah, I think this combo may just be too finicky.

I had 3133 pretty stable with Johan's initial settings and tRCDRD at 15; passed memtest86 a few passes, ran HCL and Prime95 for a few hours and worked fine for a couple days. So I said ok, lets revisit 3200.

Ran some 3200 tests and could not find stability with 16-16-16-36 or 14-14-14-34 and various other tweaks I tried. Brought ram voltages up to 1.4 and vsoc up to 1.2 and it did nothing for stability. I logged 10 runs of Memtest86 3 passes of test 6 with xmp settings and 1.35 vram with varying SOC. Test 1 with soc at 1.03125 I got 70 errors over 3 passes. Runs in the range of 1.0375 to 1.05625 (4 runs) produced between 30-50 errors. soc in the range of 1.0875 to 1.15 were the best runs, ranging from 14 to 27 errors (1.0875 was best). Over that the error rates jumped up again.

Changes to dram voltage did not seem to affect things too much. Tweaking tWR, tWTRS/RL, tWCL, and tRC did not seem to do much either.

I was done for the day and decided to go back to my 3133 stable settings...no boot, had to clear cmos. Eventually got it booting and from then on my "stable" settings produced MemTest errors right away. Dialed back to 3066, still errors. Tried 3200 xmp timings and auto for the rest on 3133, errors. Reduced speed to 3066 and it bsod's on me after a couple of hours. I am kinda thinking the system is changing internal timings on me which is messing with stability.

This has happened before, stable at one setting, make a tweak, and revert and all of a sudden it goes from stable to massively unstable.

I did notice in Memtest that the errors happened in the address range 0x2E4CCC0A0 (11852 MB) to 0x3DEE88C80 (15845 mb) 100% of the time. Of course, I have no idea how the IMC maps addresses to physical memory but the first thought in my head was ah it may be a single dimm that is not happy.

So I guess I will play with 3066/3000 and tighten things down as I can and call it a day. I know these early Ryzens were finicky with the memory. My work machine with a 3700x on X570 and 32 gb of G.Skill F4-3600C19-16GVRB runs 3200 with ease.

I would like 32gb at home anyway, maybe I will find some different sticks on the QVL and see how that does. Granted, more memory may indeed exacerbate my troubles....

 

[Zerileous]

For a 1700 32GB will be even more strain on the IMC which you're already finding the limits of. Being on the QVL may help, it may not.

 

[trents]

Take out one stick of RAM and rerun Memtes86. Try both sticks individually. I'm beginning to wonder if you have a bad stick.

 

[GunnerMan]

Take out one stick of RAM and rerun Memtes86. Try both sticks individually. I'm beginning to wonder if you have a bad stick.

I think I will try that.

Also, I am wondering if temps may be an issue but I can't confirm it yet as I was not really keeping good logs on temps and fail times, though I have a few hunches.

After my last reply I went back and set everything to Auto except I left vsoc at 1.025 and reduced a few of the primary timings slightly from auto to 16-20-20 and set ram to 2800. Ran 2 passes of memtest and then a combo of Prime95 large fft test and CUDALucas benchmark for 2 - 4 hours. I repeated the process, only increasing ram speed. At 2933 I had to increase dram voltage to 1.225v for stability. After I hit 3000, I began with 3 passes of memtest and at least a combined total off 8 hrs stress test and 1 day of general use and gaming before making any changes. I got up to 3066 which is where I am at now. My plan was to stop at 3066 or 3133 and focus on timings there once I found a good base stability setting. At 3066, it passed 4 hrs of Prime so I decided to leave it run overnight. It BSOD'd overnight so I increased ram voltage to 1.25.

At this point, it will run stress tests for hours no problem but it will BSOD about once every 24hrs (usually critical_structure_corruption). I noticed the BSOD's happened mostly in the PM when my office is the warmest and there is sun coming through onto the computer area. During warm times, with Prime running my ram temps are around 51 and 47c but I have seen them as high as 56c with 1.35v. My CPU die average also climbs upwards of 75c.


Someone said online that DDR4 is rated up to at least 81C but that temps >40c and especially into 50c could cause instability. Higher temps also means more capacitor? leakage on the ram so a faster tREFI may be needed.


UPDATE:

I tried both sticks individually in both dimms and both sticks in recommended single stick dimm. The same stability issues cropped up in about the same place and time as when together. I did get slightly more stability with single sticks so I left it with 1 for a few weeks and played around.

I found they did have the same particular settings. Won't boot at x speed with cl15 but will at 14 kinda things. So particular I wont even try and write them down. I thought random at first until I created a log to detail all settings and benchmark times, time to benchmark error, what test(s) failed, what address the error was on. I even logged the ratios given in the SPD and XMP and graphed a line connecting the points on each timing in ns as a function of MT/s and ensured any change I made stuck to it's respective line and the differences between them. I came so close to 3200 stable. I got it booting at 3200 and tried a different approach. Reduce down to 3133 and set timings that follow the corresponding linear scaling. Apply any rules I have decided to follow (sticking to scale and accounting for constants like tRAS >= tRCD + tRP meant I rarely needed to make a change that did not stick to my linear predictions. Then I tweaked voltages up/down until it was just stable. Took note of changes to voltages (I was looking for stability windows where too low or to high may cause stability problems.

I then booted into 3200, right now it was on the brink of no boot. Again, I adjusted timings up following linear progression. I did this in a change one timing, benchmark 20 mins Memtest86 test 9? Just about the only test that failed ever so its all I ran for these "test sessions". I liked the single test multipass because I could get a better view of how many errors there were, where they occurred, and what rate they occurred at. Funny thing I found, it's easy to pass the first pass in Memtest86, subsequent passes are much more difficult. Don't know why, I figure it has something to do with caching, but first pass is always much faster and much easier to pass.

Anyway, I did the timing change, benched, more errors or less errors? Inside margin of error? Based on that I would either leave the timing or revert it back. In some cases where the error difference was high I would then move the timing the opposite direction and see if I had the same effect. This helped me learn what timings made big stability differences, which ones didn't matter much, and which ones you needed to be ginger careful with. Once I went through the primary and most secondary. I would make a voltage tweak of either DDR, SOC, VDP?(the one that is 1/2 ddr...), CLDO. Again, same process. 1 by 1, by now my desk is just covered in post it notes (because this was going too smoothly to bring out the laptop). My wife comes in, picks one up and says, "so this is what you're doing in here," and walks off.

So, here I was, days and days in of painstaking benchmark sessions and I am 2hr+ stable on 3200 (the secret sauce (but not the meal!)) was CLDO 953 and up the RTT_Park to RZQ7? 1 step up from 48Ohm. So I mem test it over night, all tests 5 passes. 2 errors. Adjust, run all night. 0 errors. Boot to Windows. 10 mins in BSOD. Slapped in the face. I made a few adjustments but I couldn't go 3 hrs in Windows without a bsod. Prime also did a good job too. Then I did the same process, only in Windows. Can I get prime stable? Yes* but the intermittent (once a day avg) BSOD's plagued me.

Finally, I said well if I am this close at 3200 I can bump down to 3133 and call it a win and drink some beer. Nope. BSOD right out of the god damn gate. None of my 100% Stable saved profiles worked any more. All random BSOD's still, some even errored or would not boot. Down and down we go, landing at the default speed, optimized default bios and it BSOD'd on me this AM. ****s got a mind of its own.

(Yes, I have tried new OS installs, no I did not exceed posted voltage limits, yes I figured out the mid day sun was magnified directly onto the motherboard with no side panel)

I gave up. I have a 3900x, Asus Crosshair VIII, and 32gb of some 3600 Samsung GSkill Royal **** on the QVL on the way.