DRAM overclock complete Noob dumb question

[Zach_01]

"One truly irritating thing is that there are so many different names for the same thing. There is no "fCLK" "mCLK" "uCLK" switch that I can find." -- Me

"DRAM multi X38 with MEMCLK/UCLK in auto mode and the board run it on 2:1 ratio (probably this)
You need to find the MEMCLK/UCLK mode in there and set it to MEMCLK==UCLK for them to run on same 1900MHz speed." --Zach_01

"UCLK DIV1 MODE
If UclkDiv1Mode==1, UCLK==MEMCLK
If UclkDiv1Mode==0, UCLK==MEMCLK/2" --AMD CBS/NorthBridge/XFR

Not ONLY do they have three names for the same thing, they also have the UPPER and lower case all effed up.
SHEESH!

Allthough all vendors have in their BIOS/UEFI multiple names and sections for same things, I hear that AsRock BIOSs are kind of messy and clanky more than others.
But you found it...

 

[moproblems99]

Allthough all vendors have in their BIOS/UEFI multiple names and sections for same things, I hear that AsRock BIOSs are kind of messy and clanky more than others.
But you found it...

I didn't think their UEFI could be as bad as it is.

 

[NoJuan999]

Next time tighten TCKE 1st, then do the secondaries. I bet you like it!

I did, I just forgot to list it.
I have tCKE set at 1.

 

[pcwolf]

Well, well, well ... into the battle again, tonight. Entered a few cautious, minor changes in Ryzen Master, hit the "Apply and Test" button ... sit through yet ANOTHER reboot cycle.
BOOM! Wha?

At DRAM 3600Mhz, didn't matter what numbers I entered, or none at all, it just flat ran MEMCLK=UCLK=FCLK. Like the DIMMS were just happy to be there.

3800Mhz? Uh-uh, no way. 3733Mhz? Ha-hah-ha-haha. Nope. Not gonna give you any North Bridge past 950Mhz 1::2
In my own particular back-asswards way, I was entering the DRAM Calc timings in over and over, never hit the legendary 1::1::1 sweet spot.

My partial success method tonight: set laughably flat and loose timings, then try varying the voltages to the DRAM.
Voltage first; timing later. Back to the original reason for my original post: I don't know what I am doing.

Big thanks to all, especially Zach and The Swede. You guys know your stuff!
Onward to 3800, next!

 

[Zach_01]

Well, well, well ... into the battle again, tonight. Entered a few cautious, minor changes in Ryzen Master, hit the "Apply and Test" button ... sit through yet ANOTHER reboot cycle.
BOOM! Wha?

At DRAM 3600Mhz, didn't matter what numbers I entered, or none at all, it just flat ran MEMCLK=UCLK=FCLK. Like the DIMMS were just happy to be there.

3800Mhz? Uh-uh, no way. 3733Mhz? Ha-hah-ha-haha. Nope. Not gonna give you any North Bridge past 950Mhz 1::2
In my own particular back-asswards way, I was entering the DRAM Calc timings in over and over, never hit the legendary 1::1::1 sweet spot.

My partial success method tonight: set laughably flat and loose timings, then try varying the voltages to the DRAM.
Voltage first; timing later. Back to the original reason for my original post: I don't know what I am doing.

Big thanks to all, especially Zach and The Swede. You guys know your stuff!
Onward to 3800, next!

View attachment 146247View attachment 146248View attachment 146249View attachment 146250

You could tweak timings a little to bring that high latency down to 60ies...
Is your RAM a 3200 XMP rated?
For changing speeds and timings you should have XMP disabled from BIOS.

 

[pcwolf]

1.) Memory is Team Group 4x8Gb 4133Mhz rated 93% by DRAM Calc. Single rank.
2.) All week long stuck at 933/950 on uCLK. Used XMP to set the system to the DRAM defaults, then removed it.
3.) I think it was the tweak down on VDDCR SoC voltage from 1.2 to 1.125 that locked in coupled mode. Auto set VDDG and VDDP to 1.09v

Next step: Walk down the primary timings as you suggest. When it won't boot, walk up one step.
Then research your posts again to find the little tweaks down into the secondaries you have suggested.

By the way ... raw memory read and write speeds are both MUCH higher on Aida64 running the RAM at 4133 and uncoupled 2:1.
So, I am basically just chasing latency for bragging rights

 

[lorry]

MB manufacturers seem to take delight in renaming things, why I have no idea. Took me a while last night to find the AMD-V switch on my Gigabyte board last night, they renamed it SVM mode for whatever reason

 

[Zach_01]

1.) Memory is Team Group 4x8Gb 4133Mhz rated 93% by DRAM Calc. Single rank.
2.) All week long stuck at 933/950 on uCLK. Used XMP to set the system to the DRAM defaults, then removed it.
3.) I think it was the tweak down on VDDCR SoC voltage from 1.2 to 1.125 that locked in coupled mode. Auto set VDDG and VDDP to 1.09v

Next step: Walk down the primary timings as you suggest. When it won't boot, walk up one step.
Then research your posts again to find the little tweaks down into the secondaries you have suggested.

By the way ... raw memory read and write speeds are both MUCH higher on Aida64 running the RAM at 4133 and uncoupled 2:1.
So, I am basically just chasing latency for bragging rights

In the end, only through benchmarks you will find what speed combinations would be best for you. AIDA64 scores are only for memory performance alone. What will benefit you the most in your daily usage or maybe your special usage(if you have any), can be determined only be testing.

Keep in mind, that if your RAM is rated for 4133 CL18-18-18-18-38 when you run it below that speed, like 3600~3800 the primary timings should be set to at least something like CL16-16-16-16-32/34. This will bring down latency and may be increase bandwidth a little. I mean what you have tested so far its not in the optimal way possible.

 

[pcwolf]

Closing in ...

 

[oxrufiioxo]

Your latency looks stupidly high

 

[Zach_01]

Closing in ...

View attachment 146311

Something is very off. Is FCLK same as those above?
And most importantly... did you disable XMP? You definitely must because XMP is for 4133. If you disable it the board will set any timings on auto more accordingly to the speed you run.
Can you give us a home page of RyzenMaster with the above AIDA settings and results?

May also want to try with those primaries:
tRFC: 351
tRFC2: 261
tRFC4: 160
tFAW: 16
TCke: 1

 

[pcwolf]

XMP OFF
Used your suggested settings. tFAW 16 throws memory errors. 1usmus recommends tFAW between 4 and 6 times tRRDS so 24 removed the errors.
Ryzen DRAM Calculator memory tester is giving me a clue: inter-CCX latency is twice the innerCCX latency.

 

[Schmuckley]

TFAW 16 is too low.

 

[Zach_01]

XMP OFF
Used your suggested settings. tFAW 16 throws memory errors. 1usmus recommends tFAW between 4 and 6 times tRRDS so 24 removed the errors.
Ryzen DRAM Calculator memory tester is giving me a clue: inter-CCX latency is twice the innerCCX latency.

View attachment 146364View attachment 146363View attachment 146366

Still memory latency is unreasonably high... by now it should have been under 70ns. About 66-67...

 

[lorry]

Still memory latency is unreasonably high... by now it should have been under 70ns. About 66-67...

Remember though I also had serious problems getting down to 65

 

[Zach_01]

Remember though I also had serious problems getting down to 65

Yes but 75 is still too high for the given settings.
Hey maybe you can make suggestions!

Take a look at this, with lower settings:

Bandwidth is significant lower because of the significant lower speed and core count. But memory latency is where it suppose to be.

 

[lorry]

Yes but 75 is still too high for the given settings.
Hey maybe you can make suggestions!

Take a look at this, with lower settings:
View attachment 146372

Bandwidth is significant lower because of the significant lower speed and core count. But memory latency is where it suppose to be.

Thing is though I've still not got much of a clue as to what we were doing back then, I was pretty much simply following your lead & asking why you suggested what we did next, to try and learn.

 

[Zach_01]

Thing is though I've still not got much of a clue as to what we were doing back then, I was pretty much simply following your lead & asking why you suggested what we did next, to try and learn.

I can’t remember what I eat yesterday, let alone what we came up for your ram 2-3 months ago. Last 5 days I changed so many settings, and bench so much, that all are now a blend in my mind. I was saying, like you can just tell your settings... although your ram and pcwolf’s is not the same...

 

[moproblems99]

Bandwidth is significant lower because of the significant lower speed and core count. But memory latency is where it suppose to be.

Perhaps there is a translation issue but core count doesn't really matter. CCXs and ranks matter.

 

[lorry]

I can’t remember what I eat yesterday, let alone what we came up for your ram 2-3 months ago. Last 5 days I changed so many settings, and bench so much, that all are now a blend in my mind. I was saying, like you can just tell your settings... although your ram and pcwolf’s is not the same...

For what my timings are worth, here ya go, you set em though lol

 

[Zach_01]

For what my timings are worth, here ya go, you set em though lol

View attachment 147035 View attachment 147036

Oh, I forgot that we've made such a good job!
Yeah... 65ns is very nice!!

 

[lorry]

This wouldn't be a temperature related issue, would it? Asking as I don't think anyone has mentioned that.

 

[Zach_01]

This wouldn't be a temperature related issue, would it? Asking as I don't think anyone has mentioned that.

Well done @lorry Its not temp related but its Speed related... What you told reminded me to check his speed and there it is....!!!

@pcwolf why your CPU is stuck on base clock?

 

[Waldorf]

dont just crank up voltage. you want to run stock/xmp/spd max for start, and v later to see if you can get the next lower timing "stable".
your better off switching ram to 3600 (so 1:1 clocks) and 1.35v, as most ryzens will only do 1800 on IF.
most boards will add a bit of voltage. i can do 3600-16/19, when spd is 3600-18/22@1.35v, so dont raise it from the start.
this will get you the lowest timings without pumping lots of power thru the modules by brute force.

lower primaries to best numbers (you can get), before touching any secondaries,
raising V to check if a timing thats can be "made" stable.

 

[Zach_01]

First he needs to get his boost up and running again. The CPU is stuck on base clock. Until #29 I see that boost was working, but on the last screenshots its not...