Overclock.net banner
Cancel
Create thread New Forums

AMD DDR5 OC And 24/7 Daily Memory Stability Thread

84 reading
Jump to Latest
25,721 - 25,740 of 31,499 Posts
#25,722 ·
Veii said:
Use instead of 12-8 , -4
but 2:1 is different
Your RAS is too low & GDM is still loaded.
Click to expand...
Thank you very much for your help.

I have attached the current timings below, the previous screenshot was used to show the SCLs. I have a value of 68 for RAS. This is the value for "Optimum" from the AMD Calculator.

I wanted to ask if I have done everything well. I used 12-4 instead of 12-8. What about CCDLWR & WrBustChop? Should these stay at 24-8?

I have already done a little testmem5 run with 1usmus, and it is stable so far. Voltages are currently VDD: 1.58v, VDDQ 1.400v. I will run Anta777 Ryzen3D Config later over the Night.
Text Screenshot Font Software Number
 
Save
Reply Quote Rep+ Rep+
#25,723 ·
Br3ach said:
Is disabled better/faster than 0/1 (the minimum)?
Click to expand...
I guess it´s better off, bios tells, improves stability with sacrifise of latency --> when stable without it should have better latency
 
Save
Reply Quote Rep+ Rep+
#25,724 ·
gupsterg said:
OP of thread in my signature has been updated with section What FCLK to use?

30 runs of AIDA64, 3 runs per FCLK, FCLK from 2000MHz to 2233MHz tested.
Click to expand...
me personally I would match fclk to what memory speed you are using because that will ensure you get the lowest possible nano seconds if tweaked properly. then I would raise fclk
Number Screenshot Software Computer Personal computer

Display device Screenshot Software Multimedia Electronic signage

Font Number Screenshot Software Display device
 
#25,725 ·
iLIKE2STAYU said:
me personally I would match fclk to what memory speed you are using because that will ensure you get the lowest possible nano seconds if tweaked properly. then I would raise fclk
View attachment 2696754
View attachment 2696752
View attachment 2696753
Click to expand...
I'd be interested to see 3 runs for doing as you do vs just setting maximum stable FCLK and tuning RAM.

As all I see is you end up with exact same setup, whichever way you do it.

What RAM timings you get isn't based on FCLK, so you will be at same setup whichever method you use.
 
#25,726 ·
gupsterg said:
I'd be interested to see 3 runs for doing as you do vs just setting maximum stable FCLK and tuning RAM.

As all I see is you end up with exact same setup, whichever way you do it.

What RAM timings you get isn't based on FCLK, so you will be at same setup whichever method you use.
Click to expand...
It’s been discussed https://www.overclock.net/posts/29393827/

I have a memory of someone with a high FCLK-bin chip showing actual benchmark performance of synced vs range of FCLK at different ram speeds, but I can’t find it…
 
#25,727 ·
Luggage said:
It’s been discussed https://www.overclock.net/posts/29393827/

I have a memory of someone with a high FCLK-bin chip showing actual benchmark performance of synced vs range of FCLK at different ram speeds, but I can’t find it…
Click to expand...
No worries, look forward to share of post :) .

I spent 4 hrs doing tests, this was to know what FCLK to run and all that data points to is maximize FCLK, data ZIP.

Putting aside AIDA64, I have looked at people with high FCLK Kahru RAM Test shares and couldn't match up test speed until matched FCLK in use.
 
#25,728 ·
gupsterg said:
~5.0 Gigaflops I would deem as large deviation. I have noted even on stock setup first run can be faster then others. Just compare average with each run. Even below run, as air cooled the variation could well be down to CPU riding temperature limit.

View attachment 2696728
Click to expand...
You are 100% right, and I need to stop trying to use the computer while doing the Linpack test lol

After a fresh reboot and not doing anything after boot but start Linpack I'm getting a rock solid variance similar to yours.
 
#25,729 ·
gupsterg said:
No worries, look forward to share of post :) .

I spent 4 hrs doing tests, this was to know what FCLK to run and all that data points to is maximize FCLK, data ZIP.

Putting aside AIDA64, I have looked at people with high FCLK Kahru RAM Test shares and couldn't match up test speed until matched FCLK in use.
Click to expand...
Karhu follows FCLK like a clock, like most BW constrained tests.

Latency and "game performance" is what can be better with synced FCLK. (also depends on game...)
 
#25,731 ·
gupsterg said:
No worries, look forward to share of post :) .

I spent 4 hrs doing tests, this was to know what FCLK to run and all that data points to is maximize FCLK, data ZIP.

Putting aside AIDA64, I have looked at people with high FCLK Kahru RAM Test shares and couldn't match up test speed until matched FCLK in use.
Click to expand...
Looked through your zip.

Code: 
6200CL28
FLCK    latency           avg
2000    64.2 63.4 63.3    63.63
2033    63.4 63.1 63.6    63.36
2067    62.7 62.1 62.8    62.53    Sync
2100    62.5 62.6 63.1    62.73    Regression
2133    62.4 62.8 62.9    62.70    Regression
2166    62.1 62.7 62.1    62.30    Sync+100
2200    61.7 62.0 62.0    61.90
2233    61.7 61.6 62.5    61.93    (62.5 bad run?)
Most of these advice came from when people had a hard time running high FCLK so it made sense to stop at Sync, fx 6400 2133 rather than try for unicorn 2233...
 
#25,732 ·
Text Font Number Screenshot Software


The swing is so small, even if a bad run at top and bottom. I'll make a spreadsheet with data points, averages, etc :) .

I was also looking at read write copy as well, as looking at just ns seemed just looking at one aspect and like "splitting hairs". And even looking at ns I just couldn't see what that would give as actual performance loss/gain in normal usage. So the tldr = max FCLK.
 
#25,733 ·
Br3ach said:
Is disabled better/faster than 0/1 (the minimum)?
Click to expand...
Yes, for my 2x32 kit on 2 different Asrock boards the setting of disabled-disabled-0 is the best.. read/write/copy is 1-2% better than with 1-2-0, and nitro disabled entirely is 2-3% worse (probably because then there's no robust training). For the Rx and Tx on Asrock there's no option for 0.

I'd be curious to hear if it's also better for other memory kits and other boards.. for me it's a clear winner.
 
#25,734 ·
gupsterg said:
View attachment 2696767

The swing is so small, even if a bad run at top and bottom. I'll make a spreadsheet with data points, averages, etc :) .

I was also looking at read write copy as well, as looking at just ns seemed just looking at one aspect and like "splitting hairs". And even looking at ns I just couldn't see what that would give as actual performance loss/gain in normal usage. So the tldr = max FCLK.
Click to expand...
Yes, and lose a wee bit of read/write/copy with lower Fclk also.. enough to offset the less than 1ns better latency from 3:1 "sync" in most apps and games. I tested at 6200 and 6000... for 6000 there's not even a latency benefit to be found from 2000 vs 2200.
 
#25,735 ·
One reason I stopped trying to attain 8000MT/s (2:1), was everything in my mind said FCLK is a bottleneck, you can't make use of the MT/s. UCLK also is going to be lower vs 1:1. Your better off targeting 1:1 MEMCLK/UCLK and raising FCLK to max.

As said before to Luggage, putting aside AIDA64, Kahru test speed raises as FCLK is raised. So looking at one app, one metric, doesn't paint the right picture in my opinion.
 
#25,736 ·
gupsterg said:
One reason I stopped trying to attain 8000MT/s (2:1), was everything in my mind said FCLK is a bottleneck, you can't make use of the MT/s. UCLK also is going to be lower vs 1:1. Your better off targeting 1:1 MEMCLK/UCLK and raising FCLK to max.

As said before to Luggage, putting aside AIDA64, Kahru test speed raises as FCLK is raised. So looking at one app, one metric, doesn't paint the right picture in my opinion.
Click to expand...
The only reason for 8000 1:2 on single CCD is lower vSoc if you are power and thermal limited.
 
#25,737 ·
True I had noted SOC requirement maybe lower.

On 1:1 setups using say 6000/6200, I found usually on CPU I had maybe uplift power limit by 4-6W to not have max/sustained boost limited as SOC has eaten up some of the power budget. But I have been testing with GDM: On, where as most are not. So there is that :) .
 
#25,739 ·
anamolydetected😉 said:
Yes, for my 2x32 kit on 2 different Asrock boards the setting of disabled-disabled-0 is the best.. read/write/copy is 1-2% better than with 1-2-0, and nitro disabled entirely is 2-3% worse (probably because then there's no robust training). For the Rx and Tx on Asrock there's no option for 0.

I'd be curious to hear if it's also better for other memory kits and other boards.. for me it's a clear winner.
Click to expand...
so i will try disable disable 0 now... Nitro off works pretty fine for me
GDM off also
Font Screenshot Number Software

here i ended up atm, did a short 1000% coverage and like told before this would be a run for the weekend, exept disable disable 0 will run
at least it was y-cruncher bench stable:
White Screenshot

White

White

White

also 3 hrs of forza motorsport
 
Save
Reply Quote Rep+ Rep+
#25,740 ·
Save
Reply Quote Rep+ Rep+
25,721 - 25,740 of 31,499 Posts
About this Discussion
31K Replies
1.1K Participants
Last post:
gupsterg
Overclock.net
posts
28M
members
562K
Since
2004
The premier forum for overclocking experts and enthusiasts. Discuss hardware optimization, custom builds, benchmarking, cooling solutions, and pushing the boundaries of computing performance. From beginner guides to extreme overclocking, join our technical community to master system tuning.
Full Forum Listing
Explore Our Forums
Intel CPUs Hardware News NVIDIA PC Gaming Water Cooling