DRAM calculator for Zen 3

[Nekyno]

I have just attempted to tune my G Skill Flare X 3200c14 (Samsung B-die) memory on b550 tomahawk running Ryzen 5600x. My only try that booted was 3600Mhz on Safe calculation with MAX Voltages, only to find many errors in Memtest afterwards.

I have followed all the steps - 1. Get the report with Typhoon, 2. Load the XMP, 3. Enter Zen2/B450 and calculate everything....

Has anybody tried tuning RAM on Zen 3? Is it a general issue or do I have something wrong on my end?

 

[VuurVOS]

I have just attempted to tune my G Skill Flare X 3200c14 (Samsung B-die) memory on b550 tomahawk running Ryzen 5600x. My only try that booted was 3600Mhz on Safe calculation with MAX Voltages, only to find many errors in Memtest afterwards.

I have followed all the steps - 1. Get the report with Typhoon, 2. Load the XMP, 3. Enter Zen2/B450 and calculate everything....

Has anybody tried tuning RAM on Zen 3? Is it a general issue or do I have something wrong on my end?

3600MHz at which cas latency? Please provide all information

 

[Nekyno]

3600MHz at which cas latency? Please provide all information

I have set up what is in those black boxes. tRFC1/2/4 has been calculated on additional calculator. All other settings were left unchanged.

 

[TheLostSwede]

As pointed out by 1usmus, you no longer need to change the "other" tRFC settings, just the primary.
Also, I wouldn't bother with most of the settings, as they have minimal impact on the performance.
Keep in mind that this is a tool that gives you approximate settings, some tuning might be needed to make it all stable.
In my case, this is what the calculator spits out and as you can see, I run some of the timings tighter than that.

 

[tabascosauz]

You're probably going to need more than just 1.4V for 3600CL14... as usual, 1usmus' general voltage recommendations are...optimistic. I'd allow up to 1.45V. FlareX may be B-die, but it's far from Gskill's best or newest.

Have you already run 3600/14 on a prior board and CPU before, or is this your first shot?

VDDG and VDDP is a difficult topic, but DRAM calc recommendations are honestly wack, and consensus is that more is not always better. More SOC, mor VDDP can definitely harm your stability, though here the blatant test errors are pointing to insufficient DRAM voltage.

 

[Anonymous1243]

If you Google around you'll find lots of people not being able to get above 3200mhz on MSI boards anymore with new BIOS. I haven't seen MSI acknowledge it as an issue yet anywhere

 

[VuurVOS]

I have set up what is in those black boxes. tRFC1/2/4 has been calculated on additional calculator. All other settings were left unchanged.

Orginal 3600CL14 kits uses 1.45v. So 1.40v is too low.

tRFC values shouldnt be so high. You can change it to 360/267/165 (tRFC/tRFC2/tRFC4) when running at 3600MHz.

I would also set the following values to get better stability:
tRDRDSCL 4
tWRWRSCL 4
tRRDS 6
tRRDL 8
tWRTL 12

I am not sure if ZenTimings already works with the Ryzen 5000 series. If the application is working, please provide us the screenshot (https://zentimings.protonrom.com/)

 

[Nekyno]

You're probably going to need more than just 1.4V for 3600CL14... as usual, 1usmus' general voltage recommendations are...optimistic. I'd allow up to 1.45V. FlareX may be B-die, but it's far from Gskill's best or newest.

Have you already run 3600/14 on a prior board and CPU before, or is this your first shot?

VDDG and VDDP is a difficult topic, but DRAM calc recommendations are honestly wack, and consensus is that more is not always better. More SOC, mor VDDP can definitely harm your stability, though here the blatant test errors are pointing to insufficient DRAM voltage.

Nope this is a new build, I haven't had chance to test it prior with ideally Zen 2. Plus this is my first time tuning RAM.

I have been suspicious about those low voltages that has been thrown by the DRAM calculator. Would you suggest to do 1.45V RAM and leave motherboard voltages to AUTO?

 

[thesmokingman]

I have set up what is in those black boxes. tRFC1/2/4 has been calculated on additional calculator. All other settings were left unchanged.

View attachment 174777

lol, you're not gonna run that unless you have very highly binned b-dies. And 1.4v... The other major thing is you cannot trust dram calc verbatim, use it as a starting point but in no way is it guaranteed.

 

[tabascosauz]

Nope this is a new build, I haven't had chance to test it prior with ideally Zen 2. Plus this is my first time tuning RAM.

I have been suspicious about those low voltages that has been thrown by the DRAM calculator. Would you suggest to do 1.45V RAM and leave motherboard voltages to AUTO?

At 3600 or below, leave VSOC at 1.1 and leave VDDP and VDDG alone. Generally these board vendors know what they're doing, there's no reason to mess with them unless you're pushing beyond 3600, at which point VSOC might need to exceed 1.1V (and is a bit more complicated).

As long as you're keeping the DIMMs below 50C in a heavy memory test, you can go all the way up to 1.5V and you'll be fine. 1.45V should be about right.

The good thing about B-die is that if you have cheaper and worse binned B-die, you can generally still manage the same "good" freqs and timings under 4000MT/s if you give it more DRAM voltage.

 

[Nekyno]

Orginal 3600CL14 kits uses 1.45v. So 1.40v is too low.

tRFC values shouldnt be so high. You can change it to 360/267/165 (tRFC/tRFC2/tRFC4) when running at 3600MHz.

I would also set the following values to get better stability:
tRDRDSCL 4
tWRWRSCL 4
tRRDS 6
tRRDL 8
tWRTL 12

I am not sure if ZenTimings already works with the Ryzen 5000 series. If the application is working, please provide us the screenshot (https://zentimings.protonrom.com/)

I will try that! Thanks.

 

[thesmokingman]

At 3600 or below, leave VSOC at 1.1 and leave VDDP and VDDG alone. Generally these board vendors know what they're doing, there's no reason to mess with them unless you're pushing beyond 3600, at which point VSOC might need to exceed 1.1V (and is a bit more complicated).

As long as you're keeping the DIMMs below 50C in a heavy memory test, you can go all the way up to 1.5V and you'll be fine. 1.45V should be about right.

The good thing about B-die is that if you have cheaper and worse binned B-die, you can generally still manage the same "good" freqs and timings under 4000MT/s if you give it more DRAM voltage.

If you have to go over 1,1v on SOC you will invite WHEA errors into the mix.

And the thing with bins, there's that dram revision drop down box. Imo a lot of errors come from people exceeding the reality of their dram chip's binning. Most should be using the drop down that picks the slower timings...

 

[Nekyno]

So I have just tested 3400 14-14-14-14-28 1T at 1.4V and it survived Memtest86 for 15 minutes, one run of Cine R20 and Aida RAM test.

It seems that SOC Voltage left on AUTO is already doing 1.12V. Should I limit the SOC Voltage to 1.1V when going to 3600CL14 at 1.45v DRAM?

 

[thesmokingman]

It seems that SOC Voltage left on AUTO is already doing 1.12V. Should I limit the SOC Voltage to 1.1V when going to 3600CL14 at 1.45v DRAM?

No, if it's on auto leave it alone. The voltage isn't a one way street it depends on the load and current. Cpu voltage could show 1.45v which it will but again it is misleading until you know what the load and current is.

 

[tabascosauz]

So I have just tested 3400 14-14-14-14-28 1T at 1.4V and it survived Memtest86 for 15 minutes, one run of Cine R20 and Aida RAM test.

It seems that SOC Voltage left on AUTO is already doing 1.12V. Should I limit the SOC Voltage to 1.1V when going to 3600CL14 at 1.45v DRAM?

Those tests mean next to nothing. MT86 is far too slow for overclocking purposes (useful for testing a brand new kit you just bought, not so much for everything else), Cinebench literally does not touch memory, and AIDA is way too weak to be put in the same sentence as "stability testing".

Not sure why you're testing a frequency that you ultimately don't want to be at, with a voltage that won't work at the 3600/14 you mentioned...

Go directly to 3600/14 with 1.45V. Run HCI memtest (by opening enough concurrent 2-3GB instances to fill almost all your memory), or use Karhu, or anta777's configuration for Testmem5. If you can pass 30min-1hr of any of these, then we'll talk about the other timings.

If you still can't make 3600/14 with 1.5V, then relax it to 16-16-16 and run 1.35-1.4V and call it a day.

Uou can relax tRCD and tRP; you don't need to run flat timings.

 

[Nekyno]

Those tests mean next to nothing. MT86 is far too slow for overclocking purposes (useful for testing a brand new kit you just bought, not so much for everything else), Cinebench literally does not touch memory, and AIDA is way too weak to be put in the same sentence as "stability testing".

Not sure why you're testing a frequency that you ultimately don't want to be at, with a voltage that won't work at the 3600/14 you mentioned...

Go directly to 3600/14 with 1.45V. Run HCI memtest (by opening enough concurrent 2-3GB instances to fill almost all your memory), or use Karhu, or anta777's configuration for Testmem5. If you can pass 30min-1hr of any of these, then we'll talk about the other timings.

If you still can't make 3600/14 with 1.5V, then relax it to 16-16-16 and run 1.35-1.4V and call it a day.

Uou can relax tRCD and tRP; you don't need to run flat timings.

I just booted with 3600CL14 at 1.45V and I am about to launch HCI memtest, wish me luck.

After roughly 50 minutes there are no errors. One of the RAM reached 40C and RAM Voltage reached momentarily 1.456V.
Many thanks for help!


Extreme profile in DRAM calculator with my XMP loaded is showing following changes:

tWR: 26 -> 14
tRFC: 351 -> 288
tRTP: 10 -> 8

Top revision of B-die on SAFE calculation is a bit more adventurous:




If you have any suggestions on timings to try, or better yet, a good material to study so I can actually understand what I am doing, I would be grateful.

From what I've read 1.45V is pretty safe on B-die, with some suggesting 1.5V is fine still. What do you think?

Currently, I am unsure whether 3800Mhz and above would be worth sacrificing latency.

 

[Anonymous1243]

Did you actually gain performance? Nobody else is having success: https://www.reddit.com/r/Amd/comments/jpya2n

 

[Nekyno]

Did you actually gain performance? Nobody else is having success: https://www.reddit.com/r/Amd/comments/jpya2n

What helped me is to leave all motherboard voltage settings to AUTO, when I used to override them to specific voltage I often couldn't boot either. It's fair to say that I have been trying with lower DRAM Voltage before so it might've been due to that.

Stock with PBO:


3600CL14 OC:

Orginal 3600CL14 kits uses 1.45v. So 1.40v is too low.

tRFC values shouldnt be so high. You can change it to 360/267/165 (tRFC/tRFC2/tRFC4) when running at 3600MHz.

I would also set the following values to get better stability:
tRDRDSCL 4
tWRWRSCL 4
tRRDS 6
tRRDL 8
tWRTL 12

I am not sure if ZenTimings already works with the Ryzen 5000 series. If the application is working, please provide us the screenshot (https://zentimings.protonrom.com/)

So far I haven't been able to open ZenTimings v1.2, not sure where the issue is.

 

[tabascosauz]

@Nekyno looks reasonable for just getting the basics dialed in. Though I suspect it will take some time to understand exactly how much latency difference there should be between Zen 2 and 3 with the exact same memory.

You can follow the Memtesthelper guide to dial in some secondaries and tertiaries. It's on github. If you want, you can use HCI to help you test and shave off unnecessary volts that you dont need for stability. Less volts = less heat = more good

To finalize your overclock run some Prime95 Large FFT setting for about an hour (to stress the memory controller and see how your sticks do under near-max heat), and run HCI overnight at near full capacity. If no errors, you should be set.

As for the random firmware freakouts like the insane latency on Reddit, it's going to take some time for firmware to iron out its bugs.

 

[Zach_01]

@Nekyno why did you set B450/X470 on DRAMcalculator? Dont you have your 5600X on B550/X570 board?

This is me with 1.47V for DRAM:

And that HWmonitor is for museum in front of HWiNFO

 

[Nekyno]

You can follow the Memtesthelper guide to dial in some secondaries and tertiaries. It's on github. If you want, you can use HCI to help you test and shave off unnecessary volts that you dont need for stability. Less volts = less heat = more good

Thanks a lot. I have found the guide helpful. Link below:

MemTestHelper/DDR4 OC Guide.md at master · integralfx/MemTestHelper

C# WPF to automate HCI MemTest. Contribute to integralfx/MemTestHelper development by creating an account on GitHub.

github.com

As far as shaving off Voltages, I am already ahead of you. 1.42V was unstable, so I am running 1.43V on HCI.

@Nekyno why did you set B450/X470 on DRAMcalculator? Dont you have your 5600X on B550/X570 board?

This is me with 1.47V for DRAM:

View attachment 174797

View attachment 174798

And that HWmonitor is for museum in front of HWiNFO

View attachment 174800

I tested switching motherboard, it changed literally nothing. Nice timings, what RAM are you running?

 

[Zach_01]

You should set DRAMcalculator on X570, even though you have a B550. B450/X470 might be much different from 500series boards and the DRAM oc requirements.

I have these Corsair Vengeance LPX:

 

[Nekyno]

So apparently the best OC on zen 3 is to actually run 4 memory sticks.

This just landed at gamers nexus.

 

[Space Lynx]

So apparently the best OC on zen 3 is to actually run 4 memory sticks.

This just landed at gamers nexus.

im starting to regret everything in life.

 

[Nekyno]

Brief update. I further knocked down some secondary and tertiary timings. Results below:


I am gonna test it first at 1.43V DRAM . 1.44V should work for sure.

This is starting to be an obsession for me so I am gonna have to quit fast before it consumes even more of my time and energy.

Interestingly enough, RAM OC has nearly no effect on Cinebench R20 or Passmark CPU bench compared to stock 3200CL14. Maybe something is bugged right now, given those 4stick setups are showing much better performance.

I thought I would post sources I found helpful.

OC Guide:

MemTestHelper/DDR4 OC Guide.md at master · integralfx/MemTestHelper

C# WPF to automate HCI MemTest. Contribute to integralfx/MemTestHelper development by creating an account on GitHub.

github.com

Timings explained:

RAM timings knowledge base.

DDR3 Memory Timings Explained | MSI HQ User-to-User FAQ

<p>An explanation of DDR3 and memory timings.</p>

forum-en.msi.com

DDR4 SDRAM - Understanding Timing Parameters - systemverilog.io

Explanation to several of the key timing parameters in DDR4 memory like CL, CWL, tREFI, tRFC, tDQSS, tCCD_S, tCCD_L and many more

www.systemverilog.io

Visualization:
https://users.ece.cmu.edu/~koopman/lectures/ece348/08_bus_memory_handouts.pdf Page 18 and further.

Book (1k pages):

Cache and Memory Hierarchy Design

An authoritative book for hardware and software designers. Caches are by far the simplest and most effective mechanism for improving computer performance. This innovative book exposes the characteristics of performance-optimal single and multi-level cache hierarchies by approaching the cache...

books.google.cz

Available via Scribd subscription: https://www.scribd.com/book/282518929/Memory-Systems-Cache-DRAM-Disk