AMD Ryzen Memory Tweaking & Overclocking Guide

[Remarc]

the guide is very informative,good job

 

[Alpi]

Think lower latency is what You should forceing with Ryzen. Bandwith is pretty good, eeven better than You can achieve with Intel platforms c2c. (measured @ Aida benchmark)
My 3466c14 24/7 setup performance can be seen in the attached file. t's stable and also gives nice system performance !

 

[Zubasa]

I have R5 2600, MSI B450 GAMING PLUS and 2x8GB Hynix AFR. It stable on 3200MHz with 16-17-18-18-34 timings, but don't work on 3466 with any settings. I've tried all the options that Ryzen DRAM calculator offers, but always get an error in Prime95. What can help?

There is always silicon lottery with your particular CPU / Ram, and the motherboard design / bios also makes a difference.
Some set up will just not do 3466, maybe try 3400. I find it significantly easier to stabilize compare to 3466.

Think lower latency is what You should forceing with Ryzen. Bandwith is pretty good, eeven better than You can achieve with Intel platforms c2c. (measured @ Aida benchmark)
My 3466c14 24/7 setup performance can be seen in the attached file. t's stable and also gives nice system performance !

Given Ryzen's L3 is a victim cache, I agree that generally latency is what you want.

 

[lgustavomp]

I'm curious to see the impact of high memory frequency on cpu temperature, if there is some.

 

[FlanK3r]

One of the best OC guide ever! Many thanks for this article and hard work with software!

 

[Lindatje]

I have now this stabiel.

 

[Zubasa]

View attachment 121314

I have now this stabiel.

That is a quite high voltage on the SOC, I would try to lower it.
You shouldn't need more than 1.1V for DDR4 3466, my TR does it at around 1.06V.

 

[Lindatje]

That is a quite high voltage on the SOC, I would try to lower it.
You shouldn't need more than 1.1V for DDR4 3466, my TR does it at around 1.06V.

Its on 'auto' for the SoC voltage.

 

[Zubasa]

Its on 'auto' for the SoC voltage.

Becareful with Auto voltage, many motherboards are overly aggressive.
They just blast a ton of volts into things and hope it work stable.

 

[FlanK3r]

depends on chip also, if he has Summit Ridge or Pinnacle Ridge (I do not see it on screen)

 

[Jism]

Becareful with Auto voltage, many motherboards are overly aggressive.
They just blast a ton of volts into things and hope it work stable.

As AMD already said, it's not really the voltage that degrades chips, but the current. The Auto setting is pretty fine. Undervolting requires lots of testing and i dont think people are going to use a CPU for more then 10 years. At least i was'nt planning to use my 2700x for 10 years.

 

[smamo]

1usmus - thank you for sharing all of your learnings with us! The world of memory tweaking is full of both opportunity and frustration, and comprehensive guides like this one from experts in the community are invaluable.

Unfortunately I have Hynix MFR (F4-3000C15-8GRBB) and I am attempting to tweak on my early Ryzen 7 1700. I have found several different sets of speeds and timings that are close to stable and I can get away with using them, but an infrequent BSOD during stress testing means it's not 100% stable. I'm still searching for the optimal combination of frequency and timings.

When using your recommended TestMem5 v0.12, sometimes the program will stop running. The timer will continue but none of the tests are actually running. I believe this only happens when my RAM is not 100% stable. Is this an expected behavior?

 

[Jism]

Try a earlyer version of memtest. I woud'nt go with any memory tweaking if it's not 100% stable. You could experience data loss in the worst case.

 

[smamo]

Thanks for the suggestion Jism but it looks like the prior version of this testmem software is from 2001 and is for DOS. I am pretty sure that this behavior of the test stopping is a symptom of an unstable memory overclock. If I am truly stable, the test runs fine.

I have finally been able to achieve what looks to be a fully stable overclock. For anyone else with crappy Hynix MFR like me and a 1st gen Ryzen, maybe this will help you. DRAM voltage is 1.40v. SoC voltage is automatically set by my motherboard and I cannot change it. tRC and tRFC seem to greatly affect both my ability to boot and my stability. I think that tRDRDSCL and tWRWRSCL also have a large impact on stability. I really wanted to get these to 4 but I don't think I can do that at this frequency.

These settings give me a latency as measured by AIDA64 of ~77ns. Reducing latency is my primary target in this whole adventure. FYI my Ryzen 7 1700 is overclocked to 3.8GHz using p-states.

I want to keep tweaking but I've already spent so many hours going down this rabbit hole! I don't think there is much more than I can gain from this hardware.

 

[Jism]

Have you tried the latest? https://www.memtest86.com/download.htm

DDR4 is so much more advanced related to testing then back in 2001 was, writing single digits and expecting a clean result. The latest offers hammering and all, very suitable for rough testing. I woud'nt take 0 errors after 5 completed runs minimum.

Edit: Hynix is'nt so bad. I still managed a clean 3400Mhz out of 3200Mhz on stock settings / voltages. I have'nt bin playing yet with timings because i'm using Primocache. I cant afford any memory errors going on.

 

[smamo]

I don't think that's the same software. I am using the testmem5 0.12 software recommended by 1usmus with his custom test script.

 

[Jism]

From 2001? And what do you think memtest of this year does?

 

[smamo]

No, the prior version of testmem is from 2001. The one I am using is much more recent. I'm not an expert, I am just using the recommended stability testing program recommended by 1usmus, the author of this wonderful guide. He may be able to shed more light on it.

https://www.overclock.net/forum/27577522-post2594.html

 

[_larry]

TBH Samsung E-Die is not that bad for daily use, you should be able to get it running @3200 using the Calculator.
Because even when running @3200, the Calculator gives you more optimized values compare to the XMP settings anyway.
You could still see some gains, and 3200 shouldn't be hard to stablize with the right settings.

Finally have some time to try it now. Would be awesome to have the 3200mhz stable. I will play around with it and see what happens...I have NEVER been able to get it stable above 3066mhz...this guide is a huge help though.

 

[CityCultivator]

I'm getting a Dell Inspiron 7375. It has a ryzen 2700u.
Here are bios images, showing that memory clock and timings can be modified.
Reading this article, it says that multirank dimms are locked to 2400MHz max. Are there any single rank high clock sodimms? Does these exist? Does someone have any link, or know how to recognise those?

 

[Voluman]

Great guide and article, thank you.
Any chance to test some memory intensive application too? Winrar, 7zip, superpi 32m? I guess your gaming selection represent results to bigger audience.

 

[Caring1]

I'm getting a Dell Inspiron 7375. It has a ryzen 2700u.
Here are bios images, showing that memory clock and timings can be modified.
Reading this article, it says that multirank dimms are locked to 2400MHz max. Are there any single rank high clock sodimms? Does these exist? Does someone have any link, or know how to recognise those?

Start your own thread and ask away.

 

[CityCultivator]

Start your own thread and ask away.

Yeah, I shall start that, when I get the laptop.
I was just asking, because this thread appears, by the title, to be about ryzen memory overclocking and the laptop I am getting is a ryzen one, with memory overclocking support.

 

[Papaulus]

Thank you very much 1usmus.
Hi all, One question, What about LoadLineCalibration of Soc Voltage? Its important for ram satbility? I have various options, Auto,Regular,Medium,High or Extreme.

Thanks

Hi all, One question, What about LoadLineCalibration of Soc Voltage? Its important for ram satbility? I have various options, Auto,Regular,Medium,High or Extreme.

Thanks

Finally I tested all options and the best for my is Auto, the other give more errors than Auto.

 

[Papaulus]

Finally Stable 3200 CL14 on my R5 1600+Strix b350f! The key for my is the tRDWR+tWRRD values (after testing 6-1,6-2,6-3,6-4,7-1...) changed from 6-3 (Calculator Default) to 7-4 Thanks 1usmus! Thanks you very much Power Down Disabled. 1.35 Ram Vol and 1.050v for the Soc Vol with recomended procODT+RTT.