Overclock.net banner
Cancel
Create thread New Forums

[Official] AMD Ryzen DDR4 24/7 Memory Stability Thread

16 reading
Jump to Latest
21,441 - 21,460 of 25,828 Posts
#21,441 ·
Artylol said:
I would be happy to hear feedback
Click to expand...
The amount of WHEA is not really relevant (other than being annoying).
What matters is if it's slower or faster than FCLK 1900.
Test with y-cruncher benchmark pi 2.5b, compare the time.
Then download and configure the xmr-stak-rx monero miner.
Compare the hash rate running it, not the benchmark mode.
 
#21,443 ·
GhOsT662 said:
Qualche consiglio su come posso stabilizzare 4x8gb con la modalità gear down disabilitata su una scheda madre asus x470f che dalla ricerca che ho fatto ha un layout topologico,
la tensione della memoria è 1,5.
Click to expand...
Post in English only here please :)

Why not tRAS/tRC 28/42?
Would be easier with 2T GDM Off.
 
#21,444 ·
ManniX-ITA said:
Post in English only here please :)

Why not tRAS/tRC 28/42?
Would be easier with 2T GDM Off.
Click to expand...


laziness, I was testing the times until late in the evening and I started with these settings to start the tm5 extreme anta stress test. As for the command rate 2t from what I've read shouldn't it be slower in terms of writing and reading than 1t gear down mode on?


P.S.
I hope this time he managed to publish in English
 
#21,445 ·
GhOsT662 said:
As for the command rate 2t from what I've read shouldn't it be slower in terms of writing and reading than 1t gear down mode on?
Click to expand...
Yes it's slower but not by that much. While it's 10x times easier to achieve.
You need a lot of VDIMM for CL14 1T, probably even more for 4 x DIMMs.
1.5V could be not enough. I have a set of Viper SR and they need 1.56V minimum.

There's also the option to try 1T GDM Off with AddrCmdSetup to 55/56.
Slightly slower but requires less VDIMM and is lot easier as well.

GhOsT662 said:
I hope this time he managed to publish in English
Click to expand...
Yes, it's in English :)
 
#21,446 ·
ManniX-ITA said:
Yes it's slower but not by that much. While it's 10x times easier to achieve.
You need a lot of VDIMM for CL14 1T, probably even more for 4 x DIMMs.
1.5V could be not enough. I have a set of Viper SR and they need 1.56V minimum.

There's also the option to try 1T GDM Off with AddrCmdSetup to 55/56.
Slightly slower but requires less VDIMM and is lot easier as well.



Yes, it's in English :)
Click to expand...


Interesting i will try with 1T GDM Off with AddrCmdSetup at 55/56. Also is there any chance i can get command rate 1t with gdm off at 3733 cl15 at 1.55 temps shouldn't be a problem as i have alseye ram cooler kit in a corsair 7000d case.
 
#21,447 ·
Mr.Sunshine said:
Running core cycler now to check if anything has changed but system hasn't had any issue in past. I've never used TM5 to check memory just Memtest that has run for 23 hrs before I get 1 error. Was going to use TM5 to see if I can pin point that error.

Just checked in event viewer:

Log Name: System
Source: TPM
Date: 12/2/2022 7:42:20 PM
Event ID: 15
Description:
The device driver for the Trusted Platform Module (TPM) encountered a non-recoverable error in the TPM hardware, which prevents TPM services (such as data encryption) from being used. For further help, please contact the computer manufacturer.

Log Name: System
Source: Microsoft-Windows-WER-SystemErrorReporting
Date: 12/2/2022 7:42:27 PM
Event ID: 1001
Task Category: None
Level: Error
Keywords: Classic
User: N/A
Computer:
Description:
The computer has rebooted from a bugcheck. The bugcheck was: 0x0000001a (0x0000000000000411, 0xffffbd00ff1f4ff8, 0x000000019ad7d880, 0xff06bd00ff1f4ff8). A dump was saved in: C:\WINDOWS\Minidump\120222-8265-01.dmp. Report Id: bf098e40-2c36-4bcf-b0cb-4b5789fa7020.

Log Name: Application
Source: Microsoft-Windows-CertificateServicesClient-CertEnroll
Date: 12/2/2022 7:42:31 PM
Event ID: 86
Task Category: None
Level: Error
Keywords: Classic
User: SYSTEM
Computer:
Description:
SCEP Certificate enrollment initialization for WORKGROUP\DESKTOP-C870P3A$ via https://AMD-KeyId-578c545f796951421221a4a578acdb5f682f89c8.microsoftaik.azure.net/templates/Aik/scep failed:

GetCACaps
GetCACaps: Not Found
{"Message":"The authority \"amd-keyid-578c545f796951421221a4a578acdb5f682f89c8.microsoftaik.azure.net\" does not exist."}
HTTP/1.1 404 Not Found
Date: Sat, 03 Dec 2022 00:42:31 GMT
Content-Length: 121
Content-Type: application/json; charset=utf-8
X-Content-Type-Options: nosniff
Strict-Transport-Security: max-age=31536000;includeSubDomains
x-ms-request-id: fe06380e-0431-4634-ab47-37b1e77c7aa4

Method: GET(203ms)
Stage: GetCACaps
Not found (404). 0x80190194 (-2145844844 HTTP_E_STATUS_NOT_FOUND)


Both restart are same errors. Only does this when running TM5
Click to expand...
A TPM crash.. ok weird. Try in device Manager to wipe all TPM drivers, boot to bios, wipe TPM data there as well, boot back to windows and let windows update install the driver again. Might fix it. If you run windows 10 or 11 but dont care about TPM security features just disable it.
 
Save
Reply Quote Rep+ Rep+
#21,448 ·
Imprezzion said:
A TPM crash.. ok weird. Try in device Manager to wipe all TPM drivers, boot to bios, wipe TPM data there as well, boot back to windows and let windows update install the driver again. Might fix it. If you run windows 10 or 11 but dont care about TPM security features just disable it.
Click to expand...
@Mr.Sunshine

The BSOD is about memory corruption.
Yes the TPM crashed but that's likely because the system memory was corrupted.
I'd say you have to fix the memory profile settings.
 
#21,449 · (Edited)
dimkatsv said:
Good if it works for some people. For me i cannot boot anything below 40-D-D-D, And i only have remnant chance for stability with 60-D-D-D.
Even with AddrCMD as 64+
Click to expand...
You might misunderstand the Setups actually do. 56 is a 'stronger' value than 59 or 63; and 63 is as weak a value as 1.

They are about the setup and hold windows in regard to clock edges. In a way, a setup violation means the signal is too late; and a hold violation means the signal is too eager for the next cycle.

For the actual value in the firmware menus, the Setups are an encoded thing: Check out the AddrCmdSetup description on page 103 of this BKDG, alongside the graph on page 246 of this other BKDG doc. (note: command/address for DDR4 is SDR). "UI" is one half of a clock cycle.

By default, the data change starts on the clock fall and held until the clock fall. For what AMD is doing, Setup values 1-31 increasingly helps with hold violations by delaying changing the signal, which introduces a risk for a setup violation. Moving to 32 shifts the start of the change to the beginning of the clock cycle, wildly helping setup violations by introducing a catastrophic hold violation; and then values 33-63 increasingly helps with aiding that catastrophic hold violation.

In other words, values from 1 to 31 increasingly helps hold violations, and values from 63 to 32 increasingly helps setup violations. A Setup value of 48 would take 1/4 clock of the hold and give it to the setup; conversely, a Setup value of 16 would take 1/4 clock of the setup and give it to the hold.

So what the Setups for AMD are useful for is to pull in or postpone within the clock cycle when to change the associated pins so that the voltage is stable for the clock's rising edge.

Furthermore, AMD defines 2T command rate as "One additional MEMCLK of setup time is provided on all DRAM address and control signals (not including CS, CKE, and ODT)". This can imply that if 2T brings major stability, playing with AddrCmdSetup might help.


GhOsT662 said:
Any advice on how I can stabilize 4x8gb with gear down mode disabled on an asus x470f motherboard which from the research I've done has a topological layout,
memory voltage is 1.5.
Click to expand...
I suggest to read above too.

I'm running a tri-rank 2x8+2x16 Rev.E a hair over 3800 with GDM off:
My images there are slightly out of date. I now have ClkDrvStr to 20 ohms and ProcODT to 34.3 ohms; I also have tWTRS to 2.

I'm in a casual process of trying to get tRDRD DD down from 5 to 4; playing with the command-address bus (specifically CsODT) allowed me to get tRDRD SD to 4, but different-dimm is notably harder.
 
#21,450 ·
could i get some feedback here? I know that there is a sheet explaining what the various errors mean, sometimes the explanation is still a bit broad when trying to track down the root cause. not getting any crashes, and hwinfo doesnt even register these errors...system seems to be running great, but clearly there are more than a few issues with my timings
anything jump out as being seriously wrong just based on my screen shots? DIMM temps reached 50c during testing but typically stay under 45c during long gaming sessions. Voltage is set at 1.48v but my EVGA Dark X570 always overvolts, based on hwinfo and bios monitoring my sticks are getting 1.517v
Sleeve Art Font Fashion design Human anatomy

Font Screenshot Technology Software Multimedia
 
#21,452 ·
You are getting all those errors even with setup 56, strange.

For my post to be more helpful you will have to dial back the timings and find a stable point and then work on tightening them up till it errors again. It will take alot of time but you will get there. And an easier way is use DRAM calculator to guide you first. That's what I done and I'm no expert.

You should also watch it while it's running the test and if it errors out which it will do most likely near the start unless you have good timings just close the TM5 app. No point in running it for so long and get as many errors as you did.
 
Save
Reply Quote Rep+ Rep+
#21,453 ·
ManniX-ITA said:
The amount of WHEA is not really relevant (other than being annoying).
What matters is if it's slower or faster than FCLK 1900.
Test with y-cruncher benchmark pi 2.5b, compare the time.
Then download and configure the xmr-stak-rx monero miner.
Compare the hash rate running it, not the benchmark mode.
Click to expand...
I actually do lose performance but it seems like due to TDP and EDC constraints. Raising them allows to push scores on 2000 highers than 1900. I now understand that I don't necessarily test IF OC in the same condition for different IF Clock because higher IOD power consumption eats up CCD budget. In order to make proper test, I guess I need to lock CPU speed at something reasonable like 4GHz and test different setups.

2000 results were quite a bit worse than 1900, but now 2000 makes new records.
 
#21,454 ·
See anything grossly wrong here guys or? Outside of it being tPHYRDL 26/28 which I don't seem to be able to correct no matter what I change. 1,450v vDIMM btw. Running TM5 1usmus as we speak.

Font Screenshot Technology Electronic device Software
 
Save
Reply Quote Rep+ Rep+
#21,455 ·
Imprezzion said:
See anything grossly wrong here guys or? Outside of it being tPHYRDL 26/28 which I don't seem to be able to correct no matter what I change. 1,450v vDIMM btw. Running TM5 1usmus as we speak.

View attachment 2586736
Click to expand...
I bet if you lowered procOdt I bet you could get it back to tphyrdl 26
May also need to mess with iod and soc voltage
2T seems to get 28 tphyrdl more often than 1T
Good timings, and impressive latency considering your CPU
 
#21,456 ·
Imprezzion said:
Outside of it being tPHYRDL 26/28 which I don't seem to be able to correct no matter what I change.
Click to expand...
iirc, 26/26 with 2T would prefer an even tCL.

It's likely that you can get 1T-off 26/26.


tRDWR and tWRRD loose; try 9/1? if tWRRD can't be 1, you should be able to get it to 1 if you further lower tCWL from tCL (for their definitions, tRDWR has +(tCL-tCWL) and tWRRD has -(tCL-tCWL)).

tRCDWR can be 8. tRP might be able to come down a tick or two.



Merciless_Rick said:
could i get some feedback here? I know that there is a sheet explaining what the various errors mean, sometimes the explanation is still a bit broad when trying to track down the root cause. not getting any crashes, and hwinfo doesnt even register these errors...system seems to be running great, but clearly there are more than a few issues with my timings
anything jump out as being seriously wrong just based on my screen shots? DIMM temps reached 50c during testing but typically stay under 45c during long gaming sessions. Voltage is set at 1.48v but my EVGA Dark X570 always overvolts, based on hwinfo and bios monitoring my sticks are getting 1.517v
View attachment 2586700
View attachment 2586699
Click to expand...
Do you achieve stability with GDM on or 2T? If so, AddrCmdSetup with a tSetup bias tends to hate ClkDrvStr; I suggest to drop clock to 20 ohms and raise addrcmd to like 30 ohms. I also suggest messing with CsODT as you are indeed rank selecting.

RTT_Nominal shouldn't do anything for you as you have 1 DIMM per channel; disable it for a (probably immeasurable) benefit in temperature/power.
 
#21,457 ·
Artylol said:
I actually do lose performance but it seems like due to TDP and EDC constraints.
Click to expand...
It just needs more for higher FCLK.
What matters is the total power budget and that's something you can't control.
You can raise voltages to let it pump more but there's a breakeven point.
After that, the performances will start to drop.

Artylol said:
In order to make proper test, I guess I need to lock CPU speed at something reasonable like 4GHz and test different setups.
Click to expand...
Don't. Static OC needs different voltages and settings than PBO.
Always compare PBO vs PBO or Static vs Static.

Artylol said:
2000 results were quite a bit worse than 1900, but now 2000 makes new records.
Click to expand...
That's promising but if you really want to know test as I told you.
Another good test is Linpack Xtreme; but the result can be thermally limited.
If you top the thermal limit, you don't know for sure if it's the FCLK lowering the score or the thermal throttling.
 
#21,458 ·
ManniX-ITA said:
It just needs more for higher FCLK.
What matters is the total power budget and that's something you can't control.
You can raise voltages to let it pump more but there's a breakeven point.
After that, the performances will start to drop.



Don't. Static OC needs different voltages and settings than PBO.
Always compare PBO vs PBO or Static vs Static.



That's promising but if you really want to know test as I told you.
Another good test is Linpack Xtreme; but the result can be thermally limited.
If you top the thermal limit, you don't know for sure if it's the FCLK lowering the score or the thermal throttling.
Click to expand...
With pbo you are always thermally limited though. (I say, re-testing limits with 5C water because I rebuilt my loop from last winter)
 
#21,459 ·
Luggage said:
With pbo you are always thermally limited though. (I say, re-testing limits with 5C water because I rebuilt my loop from last winter)
Click to expand...
The problem is specific to Linpack Xtreme.
If you are thermal limited, y-cruncher and xmr-stak-rk will give you consistent results.
You can easily spot if the lower scores are due to the FCLK.
If you raise the FCLK, even thermal limited, you'll get better results.
Not with Linpack Xtreme; you'll hit earlier the thermal limit, you'll get inconsistent results and raising the FCLK will likely lower even more the scores.
 
#21,460 · (Edited)
ManniX-ITA said:
If you are thermal limited, y-cruncher and xmr-stak-rk will give you consistent results.
Click to expand...
Idk what limits are you, guys, using for PBO that you are so easily themally limited everywhere. Unless you mean soft limit at about 82-85 degrees (and that's only in heaviest stress tests).
For example i can hit only soft thermal limit with 5600X at about 115W power consumption (and that's really a lot of energy for 5600X... Far beyond efficiency point, ngl). Ofc, lower temps - higher frequency = welcome inconsistencies
And at temperature heavy load even if i will set power limit to 120, my CPU just won't take it. So it won't heat.
Well, unless i force OC it with constant voltage and try to run AVX on 4750+ mHz frequency. Then it will instantly throttle.
 
21,441 - 21,460 of 25,828 Posts
About this Discussion
26K Replies
1.4K Participants
Last post:
UNOE
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