Ryzen 3700x/x570 tuf , ram overclocking issue

[CoSmA]

Hello,

I have the following machine: 3700x(stock everything), Asus x570 tuf plus(latest bios),Corsair vengeance lpx 3600cl18(CMK16GX4M2D3600C18) which I want to overclock to cl16.

I have been using Ryzen calcualtor and the machine seems stable(seems being the keyword).


My issues is that even thou the machine passes all the benchmarks I can throw at it(prime95 blend 6+h (avx disabled)), passmark's memtest , windows memory testing tool, dram calcualtor's own memory test(memtest. max ram , and 500 task scope), aida64(3h+).
All the tests have been taken and the time was only that long because I stopped it.

The strange behavior is that after some time, some games won't start, I have a list of them (AC Creed Odissey(wont start at all, no error is thrown), RDR2 wont start(error FFFFFF), Far Cry 5(it start , i get the main menu but I cant select anything)).
At the same time, other games like Apex Legends, Forza Horizon 4, Battlefield, GTA 5 whatever else.

Now , my questions is , how can ram be tested ? I have started overclocking in the AMD Bulldozer days, and only done cpu and gpu overclocking in the past, now i started with the ram and nothing else, all these tests that I have already passed indicate that the ram is stable , i also never have had a game crashing on me (also those that wont start, if they start I can play them for hours and hours, even leave them on, no issue), never had a bluescreen or microstutters, so in my oppinion everything ndicates that the machine is stable, is it really ? Is there something else ?

Also, ran cine-bench (11.5 and 20) and aida64 membench , no error (tried them before and after the games stop working).

I cant understand, please help me, i need a way to test my ram oc .

 

[CoSmA]

 

[DaveB]

If games and programs don't load, then the RAM isn't stable. A stable RAM OC from 18-22-22-42 to 16-19-16-36 is unlikely. I would try 16-19-16-39 which is a known standard timing for one variant of G.Skill DDR4-3600 CL16 Hynix DDR4-3600. If it doesn't work, maybe return your Corsair for a 16GB set of G.Skill DDR4-3600 16-19-19-39 which is just $83.

 

[Johan45]

AIDA64 stability is pretty good at catching RAM issues quickly. Set the test to RAM and CACHE only and let it run
As Dave says that particular RAM may just not like the timings. I have always found Corsair LPX to be iffy with RYZEN. It's gotten much better but still has fits now and then.

 

[CoSmA]

@DaveB ,@Johan45 , thank you for the reply and advices.

The timings shown are by using read-xmp option from Ryzen-calc, if I export the timings with taiphoon and use those to calculate I get a much more relaxed timing (16-21-21-42-64), however my questions remains, how can i truly test the ram, it's pretty annoying and time consuming to get thru a day of succeeded tests from different test suites and just having a game not working, I'm aware that this is the part of the overclocking process, but there must be a tool that can test the ram and fail when it is not stable.

@DaveB I'm from Romania, here every price is inflated to hell and back, also availability sucks ,returning the stick is not possible anymore . I will test the timings provided by you , maybe they will work .
@Johan45 The sticks run fine at xmp, I just wanted to lower the cl from cl18 to cl16, also I want to thank you for your extensive FX oc guide, it got my Fx8120 to 4.2 without much hassle

 

[trents]

It the RAM runs stable at XMP but not with overclocked timings, then the problem is with overclocking. How much do you really think you gain in real world computing by trying to overclock the RAM anyway? Cant' be much. Not worth the hassle IMO.

 

[Woomack]

DRAM calc doesn't work well with all memory kits. From time to time there are users who are complaining it doesn't work or believe it gives them perfect mem timings while something is constantly crashing. There are many posts around the forums about that. It's not any guaranteed software. Like any tool, it should be used as a possible help. Trusting a 3rd party software for memory OC is your first mistake.
Also, 16-19-16 is far from standard for this memory IC. The software sees it as a Micron E what is the most popular in Corsair 3600 CL18-22-22 kits and also in some other brands.
Micron E doesn't like much tighter timings than what you see at auto settings in SPD/XMP. Usually, you can set main timings at ~16-19-19 ... as long as you have the luck to good IC. Since it was rated at 18-22-22 then it's not guaranteed. Some of these chips run at 3600 16-20-20. I even had G.Skill Royal 3600 18-22-22 which couldn't run stable at less than 18-21-21.

Personally I would check XMP stability. Then set XMP and CL16. Then XMP and 16-20-20. At the end check 16-19-19 or 16-18-18. Don't mess with sub timings on this IC as it won't help much. You can play with the tRFC as it's one of the only sub timings which gives performance gain.

Here are 2 of my Corsair / Micron E tests. Maybe you will find something there that may help you:
https://www.overclockers.com/forums...o-RGB-2x8GB-DDR4-3600-CL18-CMW16GX4M2D3600C18
https://www.overclockers.com/forums...r-RGB-2x8GB-DDR4-3600-CL18-CMT16GX4M2C3600C18
Both tested on Ryzen 3000.

As Johan said, check AIDA64 stability test memory+cache only so it loads memory and memory controller to ~98%. It usually crashes quickly if anything is wrong.

 

[CoSmA]

Thank you so much for this reply.
I the timings form my post are when reading the xmp, which i saw that it's pretty much general consensus that it wont spit out realistic values(as my case ). When using the profile created by taiphoon ,i get more a stable oc, the values are ,like you also pointed out 16-21-21 , and they work wonders.
My issue was/is that no software make the machine crash if the oc is not 100% stable, don't get me wrong, if the voltage is below 1.38 for ram, i do get errors in memtest, however when having 1.4 and the crazy tight timings from yesterday (16-19-16) I cant get an error, or a test to crash, or the machine to bluescreen , they only thing i get is like i said, some games not starting or strange glitches like ,windows search wont show the text that i search for.
However , something that would crash if the mem oc is not 100% stable would be much better for me .

I will check the information you provided me with, in the meantime this is what i currently run on .

 

[Woomack]

The difference in games is around 0-1% so I'm not sure if it's worth to push it to the limits and waste a lot of time on stability tests.
Loading errors are often caused by too tight timings or wrong settings. Sometimes one timing +/- 1 and all is acting weird. That's assuming that the main problem is with the memory and not something else.

 

[CoSmA]

You are right, and i will probably let it sit at this configuration, at most I will try to lower the voltages, i saw in your corsair vengeance pro tests that for you 1.35 was enough juice, maybe I'm lucky as well .
The reason why i was trying to get the lower timings was simply because I saw it as a free performance upgrade, small yes, negligible yes, but why not .

One last question, can you remember what trfc you where running when you tested the vengeance pro at 3600 CL16-19-19-39 1.35V ? Thx.

 

[Woomack]

I guess you can go down to at least 480 without losing stability. Can try lower till maybe 380. At auto tRFC is usually something around 650 at 3600 memory and something around 550 at 3200.
On the best Micron E you can set something like 3600 16-17-17 so it's not so far from let's say average kit. The worst will need something like 18-22-22 and that's why many kits have these settings.

 

[CoSmA]

Thank you for the replies and help.
Currently i tried the recommended values from ryzen master(the ones that are suggested when importing the profile create by typhoon burner), same configuration that i tried yesterday with 1.4v , but today with 1.35V .
The results speak for themselves

Just to be safe, I will continue with 2 hours of prime, blend no avx enabled and if that also goes well and the games start I might try the lower timings, or just for the sake of being able to
I gotta tell you, the whole experience is far more enjoiable since I started understanding the way timings should be configured.
What I hope to achieve is your CL16-19-19-39 1.35V if not so be it , i dont regret buying this ram kit anymore

 

[Woomack]

Nice

I'm fighting with a 2x32GB kit since yesterday.

 

[CoSmA]

I will be on the lookout for the results , if a post is to be made .
I'm curious is it running in 1t or 2t ?

 

[Woomack]

On AMD everything can run at 1T. It actually works in benchmarks and games at 4200 18-24-24 1.40V but isn't fully stable, maybe memory, maybe IMC, hard to say. 4266 can't even boot so maybe 4200 is just on the edge of stability and acts weird regardless of settings. It looks like that as voltages and timings are not helping at all. Max stable seems 4133 18-23-23 1.40V. It's something that passes 1h+ AIDA64. I have to check some more timings as benchable was 18-23-18-21-42 tRFC 620 but it was randomly crashing in stability tests. Either way, it's great for 32GB modules.
Also, one thing I mentioned in another thread. On lower capacity memory kits, performance is going significantly down once IF moves to 1:2 ratio. On 32GB modules, it also goes down but not so much and already 4133-4200 looks pretty good in tests. On 8GB modules, I had to go up to 4800+ to see a similar performance at let's say reasonable voltages and not pushed to the limits.

 

[CoSmA]

That's great, the IF ratio from what I read(and correct me if I'm wrong) should be 1:1 until 3733, I was wondering what real performance gains would have a ,,golden sample chip'' that can run a higher IF clock without dropping the ratio.
Things are so complicated now with ryzen, back in the days of the Bulldozer , a cl9 1600 kit was more than enough for about everything )

 

[Woomack]

The maximum 1:1 is generally 1900MHz but it depends on the CPU and used memory. For example, I can set 1900MHz on my 3900X but only with 8GB modules. 3733 works with 16GB modules and I can't even boot at anything above 3600 when 32GB modules are in use. On my previous processors (3600/3600X/3700X) some were stable at 1900MHz, some not. 1800MHz is pretty much guaranteed to work on every CPU.
The performance is mostly affected by memory controller speed and it's highly limited by IF clock up to at least 1800MHz. Above that, it depends on some other factors. For example on top OC motherboards you can find competitive benchmarking profiles at 4133 and tight timings so even though IF is 1:2 then for some reason these profiles were made.

For me, it looks more like a 3600 for daily work and gaming and up to 5000+ for overclocking and playing around with various memory kits as I can at least check the maximum memory frequency what is harder to do on Intel.

 

[CoSmA]

I have reached the end of my memory clocking journey, they might go further, but I don't want to continue, at least not in the near future.
Thank you very very very much Woomack for your help and for the Vengeance pro Review:


Thanks to everybody else that helped me with a suggestion/opinion, you guys rock !

 

[CoSmA]

Hey again..
I just could not resist the urge to try going higher with the infinity fabric myself , so I did...the results are not great .
All the speeds below have the correct infinity fabric also set, nothing else changed:
3666 = Worse CPU score (benched cpu-z and cinebecnh ), ram score went up(aida64 )
3733 = Fluctuating CPU scores, sometimes they are better sometimes they are worse, not as worse as they get on 3666, also in order for the ram t be stable I had to up the voltage to 1.4
3800 = Ram at 1.4 as before, CPU speed about the same with 3733, only this time it hard resets sometimes, no bluescreen , just a reset.

Looking in the hwinfo it looks like the infinity fabric is fluctuating ......however, the same behavior is occurring at normal 3600 ram speed .

All the ram speed where configured with the correct infinity fabric setting, also checked ryzen master to confirm that they are running 1:1 coupled mode ON.

Is this what unstable infinity fabric looks like, or am I missing something ?

 

[Johan45]

Is this what unstable infinity fabric looks like, or am I missing something ?

Yes that's an unstable IF