Overclock.net banner
Cancel
Create thread New Forums

*Official* Intel DDR5 OC and 24/7 daily Memory Stability Thread

29 reading
Jump to Latest
14,421 - 14,440 of 33,810 Posts
#14,421 ·
7empe said:
Ok, it wasn’t easy to nail down the issue with VST at 8400c36, while less stressful tests like Karhu, tm5 and MTP were ok up to the percentage that convinced me that it’s not memory itself failing. Neither the VDD/VDDQ/SA/MC/TX voltages. VST is now running for an hour… before it was crashing in random iterations, sometimes at 1st, other time at 10th. Never passed more than 20 minutes though. I’ve been playing with ring ratio, decreasing it from regular 51x to 50x, 49x. No change. It does not look like signal disintegration (ODTs) either. It looks like it was Ring’s PLL. I’m always running all PLLs at 915 mV. Bumped up Ring PLL to 1.020 V and voila.

Currently vdd/vddq 1.50, tx 1.30, sa 1.25, mc 1.425. Timings like for 8200, but increased tRFC/tRFCpb to stay at 120 ns as it works best for me.

Btw. If VST doesn’t pass for you, and you were stable at 8200, try 8266. You may notice the problem already there but it is easier to troubleshoot. I’m doing 8266c36 at vdd/vddq 1.47.
Click to expand...
Trying this atm. I was absolutely unable to run y-cruncher vst with 8200mts for more than 8 minutes before... Wish me luck.
BTW. I had ring, ring svid, ring pll on auto which my Mainboard translates to "F* u, I ain't doing s*" :censored:
 
#14,422 ·
stevenodellrogers said:
@Frank3764 -- is this like all ddr5? oh I see - it tells a-die, m-die, ... excellent, thanks!!
sheesh! I would think a lot of work went into this!
Click to expand...
May help to identify the ICs on the desired RAM brand - for me the SK Hynix 5600 OEM (greens) work best. I still did not have my hands on the KLEVV 5600, but should be also a very good option running it on the Apex.
 
#14,423 · (Edited)
elbramso said:
Trying this atm. I was absolutely unable to run y-cruncher vst with 8200mts for more than 8 minutes before... Wish me luck.
BTW. I had ring, ring svid, ring pll on auto which my Mainboard translates to "F* u, I ain't doing s*" :censored:
Click to expand...
I can reproduce VST issue by changing Ring PLL only. Ring PLL 915 mV VST crashes at iteration 0. Ring PLL 1020 mV VST goes forever (?). Ring ratio does not do anything in this case. What's most important is that changing PLLs does not retrain memory (if fast boot = enabled), so root cause has been found I think...

EDIT: Looking for voltage floor that helps rn. 1020 mV seems to be too much for 8400.
 
  • Rep+
  • Helpful
Reactions: tcm, sugi0lover, MickJones and 1 other person
Save
Reply Quote Rep+ Rep+
#14,424 ·
7empe said:
I can reproduce VST issue by changing Ring PLL only. Ring PLL 915 mV VST crashes at iteration 0. Ring PLL 1020 mV VST goes forever (?). Ring ratio does not do anything in this case. What's most important is that changing PLLs does not retrain memory (if fast boot = enabled), so root cause has been found I think...

EDIT: Looking for voltage floor that helps rn. 1020 mV seems to be too much for 8400.
Click to expand...

That's interesting. I have mine set to .96. I corrupted everything trying to get 8600 there other day. Simple bios clear fixed it but I went right back to 8400 as suggested and started tightening things up. Just been working a lot lately but I'm pushing it in the background.
 
#14,425 ·
7empe said:
I can reproduce VST issue by changing Ring PLL only. Ring PLL 915 mV VST crashes at iteration 0. Ring PLL 1020 mV VST goes forever (?). Ring ratio does not do anything in this case. What's most important is that changing PLLs does not retrain memory (if fast boot = enabled), so root cause has been found I think...

EDIT: Looking for voltage floor that helps rn. 1020 mV seems to be too much for 8400.
Click to expand...
UPDATE on Ring PLL:

900 mV: no bueno
915 mV: no bueno
930 mV: no bueno
945 mV: really, really no bueno
960 mV: 👍 bueno
 
  • Rep+
Reactions: sugi0lover, owikh84, FreeSpeechIsKnowledge and 2 others
Save
Reply Quote Rep+ Rep+
#14,426 ·
7empe said:
You’re touching very interesting topic of reliability. Do you know how reliable the software running on critical infrastructure is? I mean software running in military, space etc. It’s a rule of 6x9. If the probability of stable work is 99.9999% then it is considered as stable. But there is always 0.0001%… thus spectacular failures. Stability can be considered only with given probability. Same is for all overclocks Buildzoid is presenting. The fact that his VST runs through the night does not mean that it won’t fail minute later after it was shut down. Longer it takes, higher reliability you get. Nothing more.

Regarding your next-day-boot-issue. It’s very common situation and if couldn’t be corrected you could hit the hardware limitation or bios issues. If things are running at their limit, then even humidity matters.
Click to expand...
Interesting facts there! Thanks for sharing.

I'll definitely have to mess around with the ring PLL voltage. I do think the ring can be an easy target for a source of instability. I'll give that a whirl tonight and see if anything changes. So you're basically saying the PLLs don't matter for anything but Ring, in your observations?

7empe said:
UPDATE on Ring PLL:

900 mV: no bueno
915 mV: no bueno
930 mV: no bueno
945 mV: really, really no bueno
960 mV: 👍 bueno
Click to expand...
Do you know what is "Stock" for PLLs when they are on auto?
 
Save
Reply Quote Rep+ Rep+
#14,427 · (Edited)
MickJones said:
Interesting facts there! Thanks for sharing.

I'll definitely have to mess around with the ring PLL voltage. I do think the ring can be an easy target for a source of instability. I'll give that a whirl tonight and see if anything changes. So you're basically saying the PLLs don't matter for anything but Ring, in your observations?



Do you know what is "Stock" for PLLs when they are on auto?
Click to expand...
Stock is 900 mV which is the lowest they can get. I always work with elevated PLLs to 915 mV because I expect vdroop there as well. And so far I could notice improvements only by increasing ring PLL itself.

It was just lucky hit. Imagine that I had to have everything else set perfectly correct to be able to find this guy. The correct action could be falsely hidden behind some other instability issues…
 
  • Rep+
Reactions: owikh84 and MickJones
Save
Reply Quote Rep+ Rep+
#14,428 ·
MickJones said:
I'm interested to know how many people truly "stabilize" VST on 13th gen memory controllers. In Buildzoid's latest video he is complaining about how sporadic and random DDR5 is on Intel platforms.... and the crux of the video is that even when things "seem" stable, you can pass every test, reboot and pass tests - but SOMETIME in the future, it will not be full stable and VST or something will just crap the bed randomly.

In my exploration so far I find this to be true, even on the Apex. I thought I had 8200 fully stable - rebooted several times both full shut downs and restarts,and passed VST and other stressful test no problems or errors in multiple hours of various tests.... Next day, reboot again just like before - fail on the first iteration VST.

I'm going to try your ring PLL suggestion too - thanks for sharing that. But I think Buildzoid is on to something and I'm not convinced ANY intel machine is stable about 8000 at all times regardless of how much testing you do around initially dialing in your settings - something will eventually crap the bed. I'm definitely not saying you or anyone else have "unstable" settings. Nor am I saying that that a machine that has these random failures is "unstable" - in most normal work loads it wouldn't be. But I think the Zoid id onto something, in that AMD systems can be fully stabilized reliably (despite their much lower speeds), and Intel cannot this generation.
Click to expand...
I think there just isnt a hard cut off between stable and unstable frequencies, for example 7800 may be rock stable, 8000 variable/flaky stable and 8200 unstable, or even a 400mhz range that is flaky (sometimes pass stability/trains sometimes not). But most will push past the rock stable settings as understandably want highest frequency with tightest settings.

For example z690/12900k, 6800c32 easy to get stable, every setting seems to train, ran 6 months not a single issue, even would install windows/updates/software. 7400 difficult and time consuming to pass stability tests, but when finally passed knew it wasnt really stable from prior variable training issues.

Same with z790 hero, I might could get 8000 to run stability tests but would know it isnt really stable from variable booting, change one setting and fails to train, change it back fails to train again, etc. 7800 trains every time across a wide range of settings.

I just stay in the easy to get stable range which seems to be the non flaky range, so far never have problems doing that. And if I had a mobo/cpu combo that had a very large flaky range, I would change both.
 
  • Rep+
Reactions: tcm, 7empe and MickJones
Save
Reply Quote Rep+ Rep+
#14,429 ·
PLLs also sweet spots. More voltages can stabilize but can also create distortions of the generated clock signals. Proceed with caution :)
 
  • Rep+
Reactions: FreeSpeechIsKnowledge
Save
Reply Quote Rep+ Rep+
#14,430 · (Edited)
I suspect the random stability issues are related to ODTs and skew control but without seeing what gets trained, can't fix it...

If anyone has a MB that can see the values trained automatically I'd be really interested to see what values it comes up with.

The ODTs resistance values are probably why we have such disparity between the apex and dark when it comes to IMC voltages.
 
#14,431 ·
iamdjango said:
I suspect the random stability issues are related to ODTs and skew control but without seeing what gets trained, can't fix it...

If anyone has a MB that can see the values trained automatically I'd be really interested to see what values it comes up with.

The ODTs resistance values are probably why we have such disparity between the apex and dark when it comes to IMC voltages.
Click to expand...
There was a bash script that could read odts set by bios. It was working just fine with z490 boards. I need to check if it is possible to adjust it for z690-790.
 
  • Rep+
  • Helpful
Reactions: zimatz, iamdjango, Jor3lBR and 1 other person
Save
Reply Quote Rep+ Rep+
#14,433 · (Edited)
Mem kit just arrived, working on a low latency 8000, before moving to 8200. Simply trying to improve my work pc daily oc. :unsure:

If I cannot achieve some amazing results beyond the 7600 kit - I'll ship it back on Monday.
Musical instrument accessory Font Audio equipment Material property Automotive exterior

Light Office supplies Automotive tire Office equipment Hood

Rectangle Font Screenshot Parallel Number

Rectangle Font Screenshot Parallel Number
 
  • Rep+
Reactions: just20sent and kill_a_wat
Save
Reply Quote Rep+ Rep+
#14,434 ·
Grigoriv said:
Hey guys, i got TeamGroup 7600mhz ddr5 2x16 kit with Aorus Master z790 and i9 13900k. I installed it with thermal grizzly contact frame and first i had many errors in Memtest86+ after few minutes and almost instant crashes in cinebench on default settings with XMP enabled.

I remounted contact frame with original screws as i used longer ones that came with the frame previously. Now I have over 60 minutes of stable Memtest86+ with XMP enabled however, When using cinebench Multithread test i still got Kernel BSOD after 12 minutes and during next test Cinebench crashed with Application error. Can you help me find the issue? Thanks a lot!
Edit: Temps are stable, peaking 88C in Cinebench, 40-41 max temp on DDR5 during Memtest86+.
Click to expand...
Hi again! I again readjusted contact frame with precision screwdriver with 0.05Nm setting. Im still getting application crashes in Cinebench with XMP profile enabled and i must say I'm devestated... Could anyone of you guys tell me what could be the issue with my PC here? Thank you!
 
#14,435 · (Edited)
Grigoriv said:
Hi again! I again readjusted contact frame with precision screwdriver with 0.05Nm setting. Im still getting application crashes in Cinebench with XMP profile enabled and i must say I'm devestated... Could anyone of you guys tell me what could be the issue with my PC here? Thank you!
Click to expand...
I'd start with a blank slate: run the processor at a low speed (5.40Ghz all core with no boost and cache at 4Ghz, maybe even disable e-cores) and the memory at 6000Mhz with XMP timings and see if you can pass your stress tests (I'd use y-cruncher VST and TM5 Extreme1@anta777) without errors.

2) Then once you've omitted any other stability issues, test again 7600Mhz with very lose timings primary timings (e.g 40/53/53/100) and a low tREFI of 10000 (to avoid stability issues related to heat) but bump up your IMC , SA and VDDQ TX voltages (a good starting point is 1.4v IMC, 1.3v SA, 1.35v TX) and see if you're stable at 7600Mhz. Keep increasing you IMC voltages by 0.010V if you're not. You might find your CPU's IMC or MB aren't good enough to run 7600Mhz at any IMC voltage though. If that's the case try 7400Mhz and so on.

3) If you get 7600Mhz stable with loose timings try the XMP timings again. If the XMP timings fail try increasing you DIMM's VDD/VDDQ voltages while monitoring DIMM temperature (via hwinfo for example).

4) If you get 7600Mhz stable with XMP try testing stability again with stock CPU settings.

5) Profit!

Well that's what I would do anyway, might not be the best approach but hopefully helps you get started :)
 
#14,436 ·
#14,437 ·
  • Rep+
  • Love
Reactions: Jor3lBR and WebsterRKL
Save
Reply Quote Rep+ Rep+
#14,440 ·
So I just got my Apex, and TG 7600 kit. Put the system together, updated to 0903 bios including ME FW, fresh OS and latest drivers, so a pretty clean state.

Running OCCT CPU Stress (large AVX), I have noticed the ram runs EXTREMELY hot compared to my previous 6800 Dominators
Image


This is after maybe 8 mins of that running, same thing happens in TM5 (although it stays a tad cooler, maybe 68C). I presume this isn't normal, as even the Gskill 7200 kit doesn't run this hot in XMP and that thing has no thermal pads ffs...

Any help is highly appreciated :)

PS: BIOS is Default settings, XMP I enabled and Removed all limits on CPU.

Edit: I forgot to mention I do not have active cooling on the sticks, however case airflow is superb and I have never had such high stick temps before, so it's really puzzling. Maybe because they are so close together on the Apex? idk..
 
14,421 - 14,440 of 33,810 Posts
About this Discussion
34K Replies
1.2K Participants
Last post:
remittor
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