NEED HELP WITH MEMORY OVERCLOCKING VOLTAGE ON Z690 HERO

[ir_cow]

I'd be happy to collaborate with you on these testing efforts also. Was hoping to nab a 12100F for some benching in the very near future, but I could also nab a 12600K (13600K) as well. Just let me know where I can help, and I'll do my best!

Makes me want to get a 12400F or something locked. I've known about the limited SA on those, but haven't bought one yet. But yes, lets collaborate I need answers lol.

 

[Deleted member 185158]

Makes me want to get a 12400F or something locked. I've known about the limited SA on those, but haven't bought one yet. But yes, lets collaborate I need answers lol.

Well, ifbI can get a chance, I'll order a cpu and just send you my 12400F. I've only air cooled this chips, so it's only been at 5ghz+ with core reduction. Most of the time, 4.6ghz is no problem. I should mention, using the stock air cooler, but with an AMD fan from a wraith Prism for extra cfm.

 

[Dr. Dro]

Sorry for bringing this thread up but seems like the most productive discussion on CPU SA voltages that I've found looking around.

Hey @ir_cow I've finally got my 13900KS going, and the 2x16GB DDR5-6800 Trident Z5 Hynix A kit you've reviewed before. I'm having a bit of trouble mastering its XMP on the Z690 ACE with its newest BIOS installed. I'd be perfectly content if I got this stable as it is, though I'm ready to accept my fate if I have to lower the speed to 6400 or so.

Apparently my CPU really likes low SA voltage, I've tested ranging from 0.95 to 1.3 and the higher it goes, the quicker that it errors out even on a basic tool such as memtest86+. 0.95 has been the most stable by far, even 1.0V will error out within 20 minutes of the random sequence tests running.

As for the CPU VDDQ and VDD2, I've set both to 1.35V. Toyed around but it seems that also the higher they go the more prone to errors the memory becomes, but they don't work if they're much lower than this realm (for example 1.2V doesn't boot, and 1.25v errors out fast).

Any ideas? It would be greatly appreciated. Thanks brother

 

[ir_cow]

@Dr. Dro sucks you are having issues with that nice 13900KS. I only have one 13th Gen to compare. Besides looking at XOC forums / threads for clues and what "normal", I don't have the answers.

I'm leaning towards the motherboard being the issue here. Not like its bad unit, but not built for 6800+. I re-tested the Z690 Carbon WiFi in December. DDR5-7000 was the maximum I could go and I had to enter all the voltages for that too.

The reason why I am thinking its motherboard related is because I noticed the SA voltage requirements changes per MB and the maximum before errors is different too. Like for example, GB Z690 Elite AX will be 1.2V, while EVGA Z790 Classified is okay with 1.4V (XXX voltage - don't do that). I'm using the ASUS Z790 Apex for memory stuff now, It likes 1.3V, where as previous GB Z690 Tachyon, often refused to post with 1.3V.

I'm guessing auto XMP without changing voltages doesn't work either?

 

[Dr. Dro]

Yeah, auto XMP doesn't work, it fails memtest. I just completed a run of 4h memtest86+ and two 30 minute passes of OCCT memory, it passed with no errors at 6400, using:

SA voltage 0.9
CPU VDDQ 1.35
CPU VDD2 1.35
DRAM 1.4
DRAM VDDQ 1.4

And the XMP preset's timings. I think this is the hill I'll die on ultimately, I'm just doing some finishing touches now. I seem to have an issue with Windows 11 being exceptionally laggy on this setup though. I have no idea why. I'm suspecting the SSD, but I'll get there. Cheers

 

[ir_cow]

SA 1.0 will get your to about 6400 tops. which is why the none-K series sucks for memory. Intel locks them to 1V.

For reference. SA 1.25-1.3, VDD2 1.4 and VDDQ_TX 1.4-1.45V will get you to 7600-8000 with the right MB. This is independent of the DRAM voltage.

Edit: This might seem silly, but I've come across weird stuff before. The DIMMs are in the correct slots right?

 

[Dr. Dro]

SA 1.0 will get your to about 6400 tops. which is why the none-K series sucks for memory. Intel locks them to 1V.

For reference. SA 1.25-1.3, VDD2 1.4 and VDDQ_TX 1.4-1.45V will get you to 7600-8000 with the right MB. This is independent of the DRAM voltage.

Edit: This might seem silly, but I've come across weird stuff before. The DIMMs are in the correct slots right?

Yes, in A2 and B2 as indicated on the manual, which are the ones further away from the socket. Honestly the Z690 ACE is an E-ATX 4 dimmer, so while a rich motherboard it's also prone to that kind of trouble. 6400 might as well be around what it can reasonably do. Once I have it fully set up I'll explore optimizing the timings.

I've spent all day yesterday figuring out the SA voltages, but from 1.0V and up it begins to gradually lose stability. Seems the sweet spot is 0.9 to 0.95 here.

The unresponsive OS thing, I installed Windows 10 instead and it's gone. My Win 11 USB stick must have been corrupted somehow. Can't wait to see the final result once I manage to tweak the subtimings, I'm impressed with how fast things install.

 

[glennt]

Yeah, auto XMP doesn't work, it fails memtest. I just completed a run of 4h memtest86+ and two 30 minute passes of OCCT memory, it passed with no errors at 6400, using:

SA voltage 0.9
CPU VDDQ 1.35
CPU VDD2 1.35
DRAM 1.4
DRAM VDDQ 1.4

And the XMP preset's timings. I think this is the hill I'll die on ultimately, I'm just doing some finishing touches now. I seem to have an issue with Windows 11 being exceptionally laggy on this setup though. I have no idea why. I'm suspecting the SSD, but I'll get there. Cheers

I actually just came to say thanks.

I have an msi pro z690-a wifi board, with a 12700k cpu and 2x16Gb g.skill 6400cl32 ram kit, that I have never been able to run at xmp, because it's always produced errors. After reading this post, I checked the voltages and saw that the auto settings after enabling xmp were setting sa voltage to 1.28V. Ive now set my voltages as you have here, 0.95V SA, 1.35V for both cpu vdd voltages and 1.4V for both dram voltages.

Well, I don't want to speak too soon, but it's passed the 30 minute occt memory test, and I'm about an hour into testmem5 absolut profile, without a single error.

Even if I don't pass, this will still be the closest I've ever come to getting the ram stable at 6400..... and this is with the sub timings from buildzoids easy ddr5 hynix timings video.

I'm on windows 11, but haven't noticed any issues with the os lagging. So, sorry you couldn't get your ram up to its 6800, but at least feel good that you might have helped me finally get my advertised speed

 

[Dr. Dro]

I actually just came to say thanks.

I have an msi pro z690-a wifi board, with a 12700k cpu and 2x16Gb g.skill 6400cl32 ram kit, that I have never been able to run at xmp, because it's always produced errors. After reading this post, I checked the voltages and saw that the auto settings after enabling xmp were setting sa voltage to 1.28V. Ive now set my voltages as you have here, 0.95V SA, 1.35V for both cpu vdd voltages and 1.4V for both dram voltages.

Well, I don't want to speak too soon, but it's passed the 30 minute occt memory test, and I'm about an hour into testmem5 absolut profile, without a single error.

Even if I don't pass, this will still be the closest I've ever come to getting the ram stable at 6400..... and this is with the sub timings from buildzoids easy ddr5 hynix timings video.

I'm on windows 11, but haven't noticed any issues with the os lagging. So, sorry you couldn't get your ram up to its 6800, but at least feel good that you might have helped me finally get my advertised speed

Sweet! It was giving me a lot of grief at first but I also managed to improve on the timings since that, required a little more voltage than this but in general, my 13900KS likes very low SA voltage. It's very sensitive.

Once you get a fully stable result you may be able to pursue an improvement such as tightening the timings or raising the frequency. My MEG Z690 Ace does not like more than 6400 sadly, but I was able to tighten it to CL30, which I'm very happy with

I settled on SA 1.0, VDDQ 1.3, VDD2 1.35 and DRAM 1.48 for my final timings. You will have some room to play with for sure!

 

[ir_cow]

@Dr. Dro it seems either a given cpu likes SA voltage a lot or not at all. I cant even runn 6000 without 1.2V lol.

 

[Dr. Dro]

@Dr. Dro it seems either a given cpu likes SA voltage a lot or not at all. I cant even runn 6000 without 1.2V lol.

This platform's been insane figuring out. It's super intolerant to voltages, if I have something off by 0.05 it's going to generate errors, and it's like, it just errors out or programs crash, never the full-on BSOD with the glaring hint to where things went wrong. Still it only took around 2 weeks to perfect my settings, the only thing I haven't figured out, like I mentioned on the memory kit thread, is why after tightening, I can't seem to use sleep mode anymore, and shutdown actually fails (the motherboard will power down the CPU and everything, but forgets to turn itself off). Minor inconvenience, but I'd like to get that eventually fixed. Changing memory training settings seems to have had no impact on resolving this issue for me - I might try to raise DRAM VTT a little later and see if it resolves, normally this is half the DRAM voltage but influences POST, I wonder if this has an impact on the low-power states that I'm having trouble with. Either way, I haven't found where to change that setting on my board, I'm sure it's buried somewhere though, or maybe that's a DDR4 only thing. I haven't toyed with it in some time.

I've also noticed that using slow training (can be found in the Memory Fast Boot section in MSI motherboards) really helpful with getting the board to POST reliably when testing tight timings. Another feature that may come in handy is the "Memory Try It!" (the mem auto-OC) feature, even if it gives you garbage timings and wrong voltages, it seems to do some specialized form of training targeting that frequency which will greatly improve your chances of POSTing successfully, even after you changed the subtimings. I've read something that claims the same.

 

[ir_cow]

Memory Try It type functions never had worked well for me. I have seen the memory training options but don't use it really, It did help me reach DDR5-7000 for the 12900K when it was a hard thing to do. Maybe I should try it for 13900K and DDR5-8400.

If I had money, I would offer to buy that CPU/MB off of you. I'm fascinated by your combo and being stuck at 6400.

 

[Dr. Dro]

Memory Try It type functions never had worked well for me. I have seen the memory training options but don't use it really, It did help me reach DDR5-7000 for the 12900K when it was a hard thing to do. Maybe I should try it for 13900K and DDR5-8400.

Aye, I'm not using it either. It didn't like the timings I settled on. I'm just puzzled with the sleep mode/shutdown thing. No lead at all on that. I've found a thread on MSI's forum about it, but seemed to be like an early BIOS bug that was supposedly patched. Well, still happens to me once the memory is tightened, just my luck.

 

[jaszy]

SA seems very RNG.

MSI Z690-A PRO WIFI defaults 0.90 with 4800 Sam B JEDEC on the 12700k I have, but will auto set 1.24-1.25 with anything over 5600.

My 8 layer corsair 6000 CL36 Samsung B kit errors out on XMP, but a 5600 10 layer t-create JEDEC (likely A die) team group kit will do 6000 cl30-36-36-77 like nothing and wont bother erroring at all.. Voltage are all low.1.35v ram/IMC, etc.

Make it make sense.. lol

 

[glennt]

mine still turned out to not be stable in the end. Passed every stress test I threw at it, worked fine for games for about a week, but then got bsod on boot after that.

my whole system seems to be losing stability on the daily though. when I first built it, I had ram at 5800CL30 and CPU undervolted with adaptive/offset -0.100V. Seemed stable for over a year, never had issues in games at all.....then over the last couple of months I started getting MSFS crashing mid flight. Did more testing and I can't get anything stable beyond 5600CL36 on the ram (although I can drop the DRAM voltage to 1.25V at this level) and CPU I can't go anything beyond -0.035V without getting a random chance of BSOD at boot.

So I'm basically resigned to just keeping both stock now. 4800CL40 ram, and no undervolt on the cpu (12700K). The -0.035V on the CPU made basically no difference to temperatures, and the peformance with the ram at 5600 seemed intermittent....sometimes it made no difference, sometimes it actually seemed to make it worse :s

 

[jaszy]

mine still turned out to not be stable in the end. Passed every stress test I threw at it, worked fine for games for about a week, but then got bsod on boot after that.

my whole system seems to be losing stability on the daily though. when I first built it, I had ram at 5800CL30 and CPU undervolted with adaptive/offset -0.100V. Seemed stable for over a year, never had issues in games at all.....then over the last couple of months I started getting MSFS crashing mid flight. Did more testing and I can't get anything stable beyond 5600CL36 on the ram (although I can drop the DRAM voltage to 1.25V at this level) and CPU I can't go anything beyond -0.035V without getting a random chance of BSOD at boot.

So I'm basically resigned to just keeping both stock now. 4800CL40 ram, and no undervolt on the cpu (12700K). The -0.035V on the CPU made basically no difference to temperatures, and the peformance with the ram at 5600 seemed intermittent....sometimes it made no difference, sometimes it actually seemed to make it worse :s

I have a JEDEC non XMP Hynix A die kit running at 5600 cl46 that defaults for 1.25 SA/VDDQ/TX 1.2.

I was considering using it in another build for someone with a 12700k/12900k, but they need 100% stability.. wondering if setting higher voltages by default will hurt longevity for someone who only upgrades once every 10 years.

i know some boards forced 1.1+v with 4800 JEDEC early on, but I would think microcode changes "fixed" that. Hard to tell.. newer chips could also have better tolerances out of factory.. who knows.

Could just be that our IMC's are weak... I think this is the case when i try to run Samsung B at 6000 36-36-36-76 on my 12700k. Its borderline unstable regardless of similar kits on QVL... Not sure if higher IMC voltages could fix that..

I think Sam B on DDR5 prob just sucks similar to older Hynix dies on Early Zen/Zen+

Could be temp sensitive.. I have no clue.

 

[chrcoluk]

I guess you shouldn't use any Z690/Z790 MB ever lol. Default is generally 1.3-1.35V. If you go to low like 1.1, it simply won't boot depending on your memory freq and configuration.

I'm at SA 1.3, TX 1.5 and VDD2 1.45V for DDR5-8000 right now. Go much lower and it just doesn't boot.

You providing some useful info on here in a few posts I found now, I for one appreciate it.

My defaults with auto.

SA circa 0.976, but it fluctuates and seems to be in dynamic voltage mode, peaks at 0.992, motherboard allows an offset to be set like other dynamic voltages.

VDDQ fixed at 1.2v set auto in bios.

So when I was trying to beat the 3200mhz ceiling on my ram, I didnt touch VDDQ (or if I did I dont think I bumped it much), and from your info 1.2 isnt that high, interesting. It is a IMC ceiling I think as gear 2 made it stable at 3400+.

 

[ir_cow]

@chrcoluk Depends on the MB too. Now with this ASUS Z790 Apex after a few months of playing around, I settled on 1.15 SA, 1.25 TX and 1.25 VDD2 for DDR5-8000. 8200 needs a nice bump to 1.3 SA, 1.35 TX and 1.4V VDD2. Why the big jump? Idk but that's what it takes for y-cruncher and memtest5 to pass if I'm going for long uptime. Like it good for a few hours at 0.05v lower, but it will eventually error out.

DDR5-8400 is passed my IMC limit for ambient cooling. Good for about 30min from a cold boot before it starts to fail y-cruncher. 1.3 SA and 1.4-1.45v for TX/VDD2 is the limit of my CPU. It will error out regardless of te frequency if it's too high.

 

[Dr. Dro]

You providing some useful info on here in a few posts I found now, I for one appreciate it.

My defaults with auto.

SA circa 0.976, but it fluctuates and seems to be in dynamic voltage mode, peaks at 0.992, motherboard allows an offset to be set like other dynamic voltages.

VDDQ fixed at 1.2v set auto in bios.

So when I was trying to beat the 3200mhz ceiling on my ram, I didnt touch VDDQ (or if I did I dont think I bumped it much), and from your info 1.2 isnt that high, interesting. It is a IMC ceiling I think as gear 2 made it stable at 3400+.

I discussed this with cow on Discord a couple of days back but the latest BIOS (ver 1.B0) on my MSI Z690 Ace has completely changed the memory training behavior with the 13900KS CPU. With this updated BIOS, my motherboard is now very particular to tREFI and tRAS values, if either of these are too high or tight, the board no longer POSTs. I had 131072 tREFI on the 1.A0 BIOS, but this new one instantly causes it to fail POST if either value is at an extreme range now.

However, this isn't a big deal - it would seem targeting a tREFI value of 65535 clocks is the optimal window for reliability and latency reduction, and a tRAS of 28 clocks is unrealistic and will be ignored by the IMC regardless, the correct formula for tRAS is tCL + tRCD and you should add two clocks to that for optimal latency, so take my target settings of DDR5-6400 with a CAS latency of 30 and a a tRCD of 38, that amounts to 68 clocks + 2 for the DRAM cycling window and this value is set to 70 clocks.

30-38-38-70-2 primaries

It seems that the voltages I had set before - 1.0 SA, 1.3 VDDQ. 1.35 VDD2 remain stable, but I'm willing to bet that they can be changed, optimized somehow.

Also, regarding:

Aye, I'm not using it either. It didn't like the timings I settled on. I'm just puzzled with the sleep mode/shutdown thing. No lead at all on that. I've found a thread on MSI's forum about it, but seemed to be like an early BIOS bug that was supposedly patched. Well, still happens to me once the memory is tightened, just my luck.

BIOS 1.B0 has resolved the issue which caused the motherboard to enter this zombified state after a shutdown command and the low-power state (sleep mode) crashes. I can only presume this occurred due to the memory training settings.

I'll play more with this once my 4080 arrives but, I think I got it down pretty good rn.

 

[chrcoluk]

Yeah I think thats good advice.

I made another discovery on my bios. It offers up TRC as a tunable value and defaults it to 132 in the UEFI which I noticed as a really high default, I originally didnt notice what was actually happening because this timing is not shown with asrock timing configurator, however hwinfo64 does show it, and it appears even though the UEFI offers it, it is still forced using the old algorithm which I think is something like TRAS plus one of the other primaries. My main 3 (4 now in UEFI) primaries are all 14, TRAS is 30, so TRC is 44. So yeah another useless ASRock UEFI setting that does nothing.

 

[Dr. Dro]

Yeah I think thats good advice.

I made another discovery on my bios. It offers up TRC as a tunable value and defaults it to 132 in the UEFI which I noticed as a really high default, I originally didnt notice what was actually happening because this timing is not shown with asrock timing configurator, however hwinfo64 does show it, and it appears even though the UEFI offers it, it is still forced using the old algorithm which I think is something like TRAS plus one of the other primaries. My main 3 (4 now in UEFI) primaries are all 14, TRAS is 30, so TRC is 44. So yeah another useless ASRock UEFI setting that does nothing.

tRAS is at least the sum of tCL + tRCD, if this value is set below the memory controller will ignore and apply that formula regardless. This is why some motherboards allow it to be set to anything. Another setting that I've learned does nothing is tWR, which should be left on Auto - this value is generally not applied when set directly in the BIOS and is derived from the tertiary tWRPRE - and for it to be applied correctly, the value set in tWRPRE must be identical to tWRPDEN.

Try Mem TweakIt, it allows you to change the timings with the system live as well:

ASUS MemTweakIt 20221101 Ver. | bianbao.dev

bianbao.dev