Corsair Vengeance LPX 3600-CL18 OC for Ryzen 5 3600

[Freeco]

I've recently done a new build based on a Ryzen 5 3600. I haven't done any overclocking since the early 2000's, but as Zen CPU's pretty effectively OC themselves and can benefit from quick memory, I'm curious what I can get out of my RAM. Just for fun, btw, nothing too extravagant. I want to keep the system rock solid.
I paired the Ryzen 5 3600 with an Asus Prime X470-Pro, and spent slightly more on the RAM compared to an ordinary 3200MHz CL16 set: I went for a Corsair Vengeance LPX kit at 3600MHz CL18-22-22-42.
After reading up on RAM OC'ing a bit, I found the DRAM Calculator for Ryzen on TPU. Helpful tool, definitely! To figure out which chips are on the DIMMs, I used Thaiphoon Burned. It tells me it has Micron E dies.

This is what the calc suggests as safe settings:

I started extremely conservative, and after a few iterations I currently have CL16, tRCDrd 19, tRCDwr 16, tRP 16, tRAS 36. Still at 1.35V btw.
I tried getting tRCDrd under 19, but I can't get the system completely stable, even after cranking up the voltage to 1.4V which the calc suggests as max.
The more secondary timings I pretty much left as is, for now. I had a little experiment with Trdwr, tFAW and ProcODT, but that miserably fails. I reverted them back to auto.

I'm looking for a little guidance here, how to progress...
What's the deal with the tRCDrd? How could I get it from 19 to the suggested 17. A higher voltage? Or do the other mentioned voltages in the calc play a role?
Secondly, how do I best test stability? I now use the MEMbench that comes with the calc as a quick 5min check. If there aren't any popups running the test, I do a +30min Prime95 run. No errors there I consider the settings stable.
How do I test whether a change actually improves performance, and with how much? Till now, I looked at the time MEMbench needs to run. I'm now at 300.14 sec. A little experiment moving from 3600 to 3733MHz actually resulted in a 1.5% perf loss: 304.66s. So is this a good tool to make an evaluation, or is there better out there?

 

[phanbuey]

the TRFC is the timing you want to get as low as possible (400? less?).

I personally like stability testing with Forza Horizon 4 - drive around for about 30 mins followed by a few hours of aida stability test stressing memory.

I also like to test performance using Far Cry 5 benchmark (VERY latency sensitive) as well as AIDA 64 memory test.

 

[Taraquin]

Some good News for you: tRCDRD and tRCDWR has very little impact to performance. tCL and tRFC on the other hand matters a lot. You may not notice any difference in games or apps what-so-ever With 18\19 on tRCDRD and tRCDWR, especially tRCDRD is really Close tied to stability and lowering it affects stability a lot.

My suggestion: Try 16-19-18-36 and try lowering tRFC (along With CL it is usually the value that affects gaming the most). Micron has the most problems With low tRFC, but maybe you can try 550, then lower to 500 etc. tRFC has a tendency to be totally stable at one value, if you lower it by 5, then you can`t boot all of a sudden.

 

[Freeco]

For tRFC I got a bit confused with what the calc suggested: 630 and alt 560. When I look in the BIOS (well, I should say UEFI) it's set to 'auto' with an actual value of 312 (and tRFC2 192, tRFC4 132). So, that's good, right? Lower is better for this one?
I can try to reduce CL to 15 or 14. I haven't tried anything lower than 16 so far. Any other timings I could focus on?

About the game tests. You should know I'm not a gamer, at all. Not since the early 2000's, like my previous OC experience. To run the FC5 benchmark, I assume you need to buy the game, right? Or can you download just the benchmark somewhere?
I already had a look at AIDA64 Extreme earlier this week, but I don't want to spend $40 on a license for this kind of thing. It would made more sense to have spent it on higher-spec RAM The trial version only shows some of the values, so basically not that useful.
I'm looking for some freeware tool (at least something dirt cheap). Something like that MEMbench that comes with the calc. Or is it good enough to evaluate performance gain/loss?

 

[Freeco]

Blowing a little dust of this topic, cause I think I've reached the end of the line on tinkering with my RAM timings.
First of, next to the Ryzen DRAM Calc I also found some useful info and suggestions here and here. Maybe it can help someone else...
I kept track of my tests in a sheet. The timings in column BX is what I ended up with. Each iteration was tested with a Membench in Ryzen DRAM Calc (scores at the bottom) and 45-60mins of Prime95. If no errors popped up in those 2 tests I consider it to be stable.
Every few tests I also ran a run of TechArp's x264 HD Benchmark. The avg of 4 runs are also shown at the bottom. It's free, and you could say this is a real-world test. While Membench is more a synthetic test, I guess.
I left out some of my early tests, and all the failed tests. I tried tweaking every timing up to row 39 'power down enable', but those last ones I couldn't get to work without spawning errors.


Now, what bothers me is that the timings in column BG have the highest score for the x264 test. I did further tweak some other timings, which resulted in a faster Membench run, but lower fps for x264 encoding. That's kinda contradictory...
2ndly, I also ran MemTest86 a few times, and even while bumping up the voltage to 1.45V I can't get it to pass 4 runs successfully. And this I take quite seriously. Stability stays the prime focus.

Anyone some advice? Why I could see a negative effect on the x264 encoding speed near the end?
For MemTest86 I think I'll just backtrack a few iterations, to eg column BG, and see if it passes those timings.

 

[examen1996]

@Freeco I want to thank you, I can even remember when I last signed in here, but for all the work you done and how much you helped me with my memory overclock I searched for my account and signed in just to say THANK YOU.

I got the same corsair ram stick (i searched for the product code and that's how i reached your post) and this is what I achieved after about a day of overclocking .

You work is admirable and highly appreciated, thank you for documenting it as meticulously as you did.

 

[Zach_01]

Can you screenshot and post the following?







First is Thaiphoon Burner software (click read and select 1 module)
For RyzenMaster please have the latest version
And AIDA64 is a paid benchmark (FYI) or you can run them individually...



Have your board's UEFI to the latest version(AGESA)
Win10 to the latest v1909
and latest Chipset drivers directly from AMD (not ASUS)

 

[examen1996]

Hey @Zach_01

Above you have the requested info, also , I am running the latest bios from asus, latest chipset drivers from amd and ryzen master is also the latest iteration (I only got the machine last friday , so everything is new ) )


Please let me know if you want to see anything else

 

[Zach_01]

You said you have R5 3600 and have it on your profile specs too, but from RyzenMaster and AIDA I can see the 3700X...
Whats the deal?

 

[Chomiq]

You said you have R5 3600 and have it on your profile specs too, but from RyzenMaster and AIDA I can see the 3700X...
Whats the deal?

Last guy is not the OP?

 

[Freeco]

you're welcome examen1996 Glad it helped someone.
And no, Chomiq, I'm the OP.
Should I also post screenshots??

 

[Zach_01]

you're welcome examen1996 Glad it helped someone.
And no, Chomiq, I'm the OP.
Should I also post screenshots??

If you like...
When I saw this thread yesterday I misread it for new one... and asked for pics since you asked for help...

 

[Freeco]

Here you go then...
My trial period of AIDA64 already expired quite some time ago. These are the individual runs:
Memory Read: 52098 MB/s
Memory Write: 28730 MB/s
Memory Copy: 49448 MB/s
Memory Latency: 68.3 ns

 

[Zach_01]

From the RyzenMaster pic I see the CPU SOC voltage to 1.36V (!)
Is this correct I wonder? Did you set it or the board by itself?
From what I know the max safe 24/7 voltage for SOC is 1.25V and this 1.36V, if true, could hurt the I/O chip longterm

 

[Durvelle27]

Here you go then...
My trial period of AIDA64 already expired quite some time ago. These are the individual runs:
Memory Read: 52098 MB/s
Memory Write: 28730 MB/s
Memory Copy: 49448 MB/s
Memory Latency: 68.3 ns

View attachment 142557 View attachment 142556
View attachment 142555

Comparable to mine

 

[Freeco]

From the RyzenMaster pic I see the CPU SOC voltage to 1.36V (!)
Is this correct I wonder? Did you set it or the board by itself?
From what I know the max safe 24/7 voltage for SOC is 1.25V and this 1.36V, if true, could hurt the I/O chip longterm

You mean the value for VDDCR SOC, right? Actually you make an interesting point...
No, I didn't change that in the UEFI settings, it's still on auto. All that's non-stock here is the DRAM voltage, which is set to 1.35V by applying the XMP profile.

And basically all I changed are these DRAM timings:

And further down the screen also the values for tRFC, tRFC2, tRFC4 and CR.

I'm curious why VDDCR SOC is reported to be so high. Cause it might be the reason why I hardly see the SOC power below 10W at idle.
I've built a slightly slimmed down desktop for my sister last month, also built around a Ryzen 5 3600, and on this one I see the SOC power goes down to 5W at idle.

It might just be Ryzen Master not showing the proper value. HWiNFO64 6.20 says SoC Voltage (SVI2 TFN) is at 1.087V. Or isn't that the same thing?

 

[GorbazTheDragon]

I just manually locked SoC voltage to 1.1v on my board (same model)... More voltage =/= more stability for that one past that point.

Micron rev E runs very high RCDRD, and RCD generally doesn't scale with voltage so you can't expect that to become more stable.

As far as the performance regression goes, it might be caused by GDM going on or off. I would lock that to on and retest multiple times to reduce variance. Given that you say it doesn't pass memtest86, I'm inclined to believe that it's partially stable, not enough to crash, but enough to cause the benchmark to lose performance due to repeating things.

 

[Zach_01]

SoC voltage SVI2 is the same but HWiNFO has even another value for this, in a section below that one (CPU VCORE SOC).
Around 1.05~1.1V is the default for ZEN2.
Must be wrong reading from RyzenMaster. It doesn’t show at all Memory and memVTT voltage. Is the latest version?
Make sure to have latest BIOS, latest chipset drivers from AMD (not ASUS), latest win10 version, and latest RyzenMaster version.

As for SOC power consumption mine also stays constantly around 12W but:
1. have disabled power savings for SOC (CPU Uncore/SoC OC Mode: enabled)
2. raised voltage for cLDO VDDP/VDDG

Your sister’s system what ram runs?
Ram speed, infinity fabric speed, mem controller speed... can all affect power of SOC.

Can you post a full page of HWiNFO?

 

[GorbazTheDragon]

It doesn’t show at all Memory and memVTT voltage. Is the latest version?

Doesn't show on my X470 prime pro either (in ryzen master, and there is no DRAM voltage sensor shown in hwinfo either), presumably something with the board not exposing that sensor/controller.

I am getting around 11-12w on the SoC rail at the moment while running x-plane...

 

[ocbenny]

Freeco, nice thread. I have the same memory as you do. Instead of letting ryzen dram calc to the xmp, did you try and export a html report from typhoon burner and import it to ryzen dram calc? Full instructions in the comments. I got all primary and secondary timings running today and tested out as in the video posted without error..

 

[examen1996]

Coming back to this thread I have some questions that may sound stupid.
Have you ever encountered situations where like games (AC Odyssey and Witcher 3) start and run(and when you are playing them they wont crash, no matter how long you play them), however after some time, be it in idle, or just after some youtube you are not able to open them back up ?. And only those 2 .

I don't understand, passed the passmark memtest (left it to do all the runs) and ran prime95's blend(without avx) test for 2 hours, no errors or instability.
However , these 2 games behave strange. And now I ask you guys(since i don't have experience with ram oc), should I just revert to xmp ?

Memtest passes with flying colors, prime95 and aida64 also no problem, and never did I had stutters or games crashing on me . And also even when they don't start, others such as Forza horizon 4, work without problems.

 

[Freeco]

AFAIK I have the latest Ryzen Master. When I let it check for updates, it says it's the latest version. The About page mentions 2.1.0.1424.
(it's a bit annoying AMD doesn't mention a version on the download page...)
Yeah, and I did install the latest BIOS for the board (5406 with agesa 1.0.0.4), and I have the chipset drivers from the AMD site (1.11.22.454). But I see last week 2.01.15.2138 was released. I'll install this in a moment.

Here's some screenshots from HWiNFO (I left out the bottom part for my SSD/GPU/NIC)


(the good observer will notice I'm no gamer )

The RAM in my sister's build is quite similar to mine. Also Corsair Vengeance LPX 3600MHz 18-22-22-42, but I believe it's a different SKU. If I remember well it had different IC's than my sticks. I left hers at XMP timings, as she's not the 'tinkerer' like I am

ocbenny, what good would it do to export the report from TB and import that in Ryzen Calc?

 

[examen1996]

@Freeco The reason why importing the profile is ,,better'' is that you really put in your memory timing (auto reading is a bit off, usually some values are higher , some lower ), however, the numbers you ahve already reached (the low timings) are beyond the safe values , so you wont gain anything from it .

Below you can see the values with the imported(exported with taiphoon burner) profile:

 

[examen1996]

UPDATE: Just got back on my standard xmp profile, and the games that i sometimes have issues starting are still randomly not starting.
As said before, other games work fine, this + the fact that all my overclocks passed the 2h+ passmark memtest and my own prime blend 2h + make me think that the oc was actually stable, and that the issue is with some driver or with the games itself .

What do you guys think, every oppinion is reatly appreciated.

 

[Chomiq]

Do you have the games running minimized on taskbar and they don't return to normal state or do they don't start at all? Some games act up when minimized so that's normal.