• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Ryzen 7000 / AM5 Memory guide (DDR5 6000+ stability, slow boots, memory training etc)

Status
Not open for further replies.

Mussels

Freshwater Moderator
Joined
Oct 6, 2004
Messages
58,412 (7.74/day)
Location
Oystralia
System Name Rainbow Sparkles (Power efficient, <350W gaming load)
Processor Ryzen R7 5800x3D (Undervolted, 4.45GHz all core)
Motherboard Asus x570-F (BIOS Modded)
Cooling Alphacool Apex UV - Alphacool Eisblock XPX Aurora + EK Quantum ARGB 3090 w/ active backplate
Memory 2x32GB DDR4 3600 Corsair Vengeance RGB @3866 C18-22-22-22-42 TRFC704 (1.4V Hynix MJR - SoC 1.15V)
Video Card(s) Galax RTX 3090 SG 24GB: Underclocked to 1700Mhz 0.750v (375W down to 250W))
Storage 2TB WD SN850 NVME + 1TB Sasmsung 970 Pro NVME + 1TB Intel 6000P NVME USB 3.2
Display(s) Phillips 32 32M1N5800A (4k144), LG 32" (4K60) | Gigabyte G32QC (2k165) | Phillips 328m6fjrmb (2K144)
Case Fractal Design R6
Audio Device(s) Logitech G560 | Corsair Void pro RGB |Blue Yeti mic
Power Supply Fractal Ion+ 2 860W (Platinum) (This thing is God-tier. Silent and TINY)
Mouse Logitech G Pro wireless + Steelseries Prisma XL
Keyboard Razer Huntsman TE ( Sexy white keycaps)
VR HMD Oculus Rift S + Quest 2
Software Windows 11 pro x64 (Yes, it's genuinely a good OS) OpenRGB - ditch the branded bloatware!
Benchmark Scores Nyooom.
This ones going to be short and sweet, fortunately.
I've done a LOT of repetitive testing with a 7900x here to replicate why some users have slow boot issues of what they mistakenly call 'memory training', and i did not want this lost hundreds of pages into the Zen Garden clubhouse thread where no one would see it.

It's usually down to one of two things

1: A physical issue like a too-tight or imbalanced CPU cooler causing poor contact. Don't tighten one side of a cooler all the way and shift over, do them evenly. Clean the pads on the CPU from fingerprints.
2: Incorrect BIOS settings, probably copied from what "everyone says is best"


DDR speeds are 'doubled' so 6000 MT/s RAM (incorrectly sold as 6000 MHZ) is actually 3000 MHZ. This is what confuses beginners.

MCLK is the true memory clock, double it for DDR speeds.
FCLK is the Infinity fabric, that seems to need to be within a certain multiplier range of the MCLK - any value dividable by 0.25x (1.5x, 2.75x 3x, etc) This is critical.
People suggest 3000/2000/3000 is the dream settings for DDR5 6000, and it is. However, there's very little information on what to do if your PC has issues with that.


1. Higher is not always better - Look at the insane latency on the first result.

2. Lower is not always slower. 1800Mhz was only 0.009% slower than 2000, while being faster than 2067.

3. You can only run UCLK at half or full speed, halving has around a 3ns penalty. If you fix stability issues by halving it? Do so!
1698661452531.png


Put simply, not every CPU and motherboard are going to handle the Fclk at 2000 - lower it to 1800 and see if your slow boot issues go away.


Use AIDA64's cache and latency test and check your own system - sometimes the math is not perfect due to internal mysteries and rounding errors.

As an example DDR5 7200 runs extremely easily on this Asrock Taichi Carrara with Uclk halved, and the latency is lower than the DDR5 6000 testing when it's commonly believed that's not possible.
3600 MCLK divided by 2066 IF maths out to 1.74x and gives great results for XMP on with no manual tweaking.
1698662108382.png


I Also have some DDR5 8000 i'll be testing in the coming days, knowing how these values need to be in sync is more complex than the old 1:1 days but also more freedom as long as you do some basic math first!
 
You will ever get a true reading with aida64 there is simply to much variance between runs, I've tried with my am5 system many times.....

You are so right regarding CPU mount, I have found am5 is extremely sensitive to imbalanced pressure on the socket, too little or to much causes a plethora of
issues, as an early adopter I went through this for days till I figured this out.....

as for MCLK/FCLK/UCLK....

There is no magical formula regarding these, I've tried most combinations, and have always settled on 3200 MCLK/2133 FCLK/3200 UCLK,

why may you ask, because it shows the most repeatable results within margin of error....

the only time there has been substantial improvement is with the fclk bug.....
 
Last edited:
A contact frame is a good and sensible buy for AM5 for this reason.
 
DDR speeds are 'doubled' so 6000 MT/s RAM (incorrectly sold as 6000 MHZ) is actually 3000 MHZ. This is what confuses beginners.
Sorry for the side question but... How many cycles/clocks does it take to send one transfer?
 
A contact frame is a good and sensible buy for AM5 for this reason.
not required, never has been, Educating people about the am5 socket contact issues is free.......
 
Last edited:
Don't think I said it was required, did I?
 
No you didn't say it was required, you said it was a sensible buy,
at the end of the day a contact frame would not mitigate people tightening up a cooler unevenly......
 
There is no magical formula regarding these, I've tried most combinations, and have always settled on 3200 MCLK/2133 FCLK/3200 UCLK,
FCLK should be MCLK/1.5 to stay in sync.

So for 6200MT it's 3100MCLK/1.5=2067

2133 FCLK would show higher latency despite theoretical improvement.

57ns Zen4.png
 
You will ever get a true reading with aida64 there is simply to much variance between runs, I've tried with my am5 system many times.....

You are so right regarding CPU mount, I have found am5 is extremely sensitive to imbalanced pressure on the socket, too little or to much causes a plethora of
issues, as an early adopter I went through this for days till I figured this out.....

as for MCLK/FCLK/UCLK....

There is no magical formula regarding these, I've tried most combinations, and have always settled on 3200 MCLK/2133 FCLK/3200 UCLK,

why may you ask, because it shows the most repeatable results within margin of error....

the only time there has been substantial improvement is with the fclk bug.....

FCLK:MCLK ratio definitely matters, there is a point however in which faster flck out paces 1:3 ratio.

6000:2000
6200:2067
6400:2133

Will be standard, and optimal generally achievable options when it comes to 1:1 memory mode on AM5.

It generally takes an additional 100mhz on the fclk to make up for de-sync the FIFO buffer by not running 1:3; which is generally only viable for improvement at 6200mhz for memory.

In testing 6400 at 2133/2200 fclk, there is little to no difference in latency while there is a moderate improvement in memory read and write bandwidth.
 
FCLK should be MCLK/1.5 to stay in sync.

So for 6200MT it's 3100MCLK/1.5=2067

2133 FCLK would show higher latency despite theoretical improvement.
1.4ns vs lower bandwith..........
 

Attachments

  • cachemem 6200.png
    cachemem 6200.png
    91 KB · Views: 3,701
  • cachemem6400.png
    cachemem6400.png
    90.8 KB · Views: 3,618
Last edited:
6 and 2 3`s.....

I must admit I find mussels very first image intriguing, something very off with that image as I have personally never seen memory latency that high.......
 
I don't think the contract frame does anything for AMD. Intel has the IHS issue.
It does exactly one thing. It looks really nice.
 
I use this with my 7600X system, my only gripe with it is its a PITA to fit properly especially if one has BIG fingers. When I built my second AM5 system, I did not take the backing paper of it & it made fitting a hundred times easier - still does the same job in the end so its win win. So that cheap accessory helps keep a clean fit for the HS & CPU.

From what I've experienced in reference to the OP for months now with my AM5 systems is this - do not use beta bios, always use official non beta even if its a few months old. I'm on 1.28 for my Asrock board now & its not giving me any issues with slow boots (of course the first boot with a given config is going to be slow) I never copy anyone else's setting cause of the silicon lottery issue. Unfortunately one has to be focused & put in the time to find the ideal config for their combo of cpu/ram/board & bios version. There is no easy fit for everyone. Another thing too, try & use EXPO ram if you can - it aids in taking the guess out of setting primaries & subs to the right levels for Zen 4 IMC uniqueness.

I have a new set of G.Skill EXPO ram playing with atm but have no interest in running anything less than 1:1 cause 7600X is a single CCD & the extra bandwidth from 1:2 can benefit dual CCD chips more so.
 
a64benchTR7960XquadchannelDDR5.png

:) for me, this run´ best
 
As an example DDR5 7200 runs extremely easily on this Asrock Taichi Carrara with Uclk halved, and the latency is lower than the DDR5 6000 testing when it's commonly believed that's not possible.
3600 MCLK divided by 2066 IF maths out to 1.74x and gives great results for XMP on with no manual tweaking.
View attachment 319483

I came across this thread trying to get my new 7900X box setup. I have Teamgroup 700MHz 24x2 DDR5 (34-42-42-84). Initially I was ~77000 MB/s mem read speeds with ~87ns latency with my rams primary XMP profile. I tried the timings above and my system is still rock stable I am getting ~81000 MS/s mem speeds and ~77ns latency. Nice little improvement!

1729444374916.png
 
well i'm using ddr6000 with d-die micron and its cl36-44-44-96 with 1.35v dcop profile.
should i sell and buy a better ram?
or can i squiz any preformance from it?
 
well i'm using ddr6000 with d-die micron and its cl36-44-44-96 with 1.35v dcop profile.
should i sell and buy a better ram?
or can i squiz any preformance from it?
No, just change your graphics card if want to get more performance which you really can feel in real world.
 
No, just change your graphics card if want to get more performance which you really can feel in real world.
lol i keep my gtx 960 for oldy games basicly
and yep i know my gpu is really bad this days :(
i want to grab me 7700xt next month...
 
I want to share some settings. These are on Asus Intel z790 board, but many of these settings may be applicable to AMD systems.

This helps with quicker cold post times for me, results may vary!!

EDIT: Want to mention, with full screen logo disabled, you'll watch the VGA boot, some cards are slow.
 

Attachments

  • 20241020_130702.jpg
    20241020_130702.jpg
    3.7 MB · Views: 548
  • 20241020_131001.jpg
    20241020_131001.jpg
    4.6 MB · Views: 546
  • 20241020_131209.jpg
    20241020_131209.jpg
    2.3 MB · Views: 541
Last edited:
1729633024622.png
1729633048673.png

Here are my results, isnt' the 74.8 ns too high?
Compared to @dgianstefani 's 57.9 with a similar module looks a bit bad.
Any suggestion?
 

Attachments

  • 1729632783117.png
    1729632783117.png
    328.7 KB · Views: 486
  • 1729632789025.png
    1729632789025.png
    39.7 KB · Views: 380
Hi @Mussels I recently got myself an AMD Ryzen 5 7500F running on an ASUS prime b650m k. I understand that the mobo is maybe not the best, but it does the job. My question is, when I try to set the FCLK freuency in the bios to be 3000 my system will not start. I tried 2100 and it worked but 2500 will not work. I am not sure if there is other settings I need to adjust. Memory is DDR5 6000 which I can run 6400 with lower timings, but not 100% stable. Your help would be greatly appreciated.
 
Status
Not open for further replies.
Back
Top