cancel
Showing results for 
Search instead for 
Did you mean: 

z790 Apex Encore Owners - Voltage Question

cthree
Level 9

Hello,

I've spent a lot of time over the past month learning how to properly memory overclock.  I'm still a rookie, but I've learned a lot in a short amount of time.  It's been incredibly time consuming, stressful but highly rewarding.  I ended up ditching my z790 hero and switched to the 2-dimm Apex Encore board and I'm glad I did.  I'm at the point where I can successfully stabilize higher frequencies with tighter timings in a reasonable amount of time.  I still have a ton of questions regarding voltages however...

For starters, I noticed that IVR VDDQ-TX (CPU VDD2 I believe) likes to run really low on this board.  I've also noticed that VDDQ-TX is not linked to Mem VDDQ like on other Asus boards.  I've been able to stabilize 8000cl38 w/ (VDDQ-TX @ 1.25 and SA @ 1.18).  Any higher than this, both TX and SA needs to go up.  I've also stabilized 8200cl38 w/ fairly tight timings w/ (VDDQ-TX @ 1.35, SA @ 1.25, Mem VDD 1.45, Mem VDDQ @ 1.51).  Which leads me into my question....

Is there an ideal methodology for tuning voltages when overclocking memory?  For example, should I focus on tuning SA/TX/IMC VDD, or Mem VDD/VDDQ first?  As of now I typically find a lower SA & TX voltage and then begin tuning Mem VDD/VDDQ.  If I can't stabilize after tuning Mem VDD/VDDQ, I then begin increasing my SA and TX.  Does it make more sense to tune SA/TX/IMC VDD first, and then proceed to Mem VDD/VDDQ?  I wonder what the general approach is regarding tuning voltages, if one even exists.

Looking forward to some feedback!

 

940 Views
22 REPLIES 22

Silent_Scone
Super Moderator

hI @cthree,

Continue you what you are doing and use a methodical approach from a stable range. 

Typically from my experience with 13th & 14th gen, you shouldn't need to adjust VCCSA once you've found an appropriate voltage. As you increase frequency, MC VDD and IVR TX may need to be increased, as with VDD and VDDQ if insufficient for the applied settings. I find for every ratio moving from 8200, IVR required a 20mV increase in order to maintain stability. The same is often found for IMC VDD if the voltage is insufficient. Some CPUs also fail to POST if using more than 1.5v IMC VDD.

Like with anything related to signal IO and overclocking, often sharing what works for one CPU to another is all one can do.

13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

Appreciate the reply.  I wish I got to this message the other day... would have saved me a good hour.

I recently began attempting to stabilize an 8400 OC and it has been by far the most difficult to stabilize.  It took me awhile to realize I needed to start upping TX 10mv at a time to begin seeing solid results.  Increasing VCCSA absolutely destroyed stability, even as little as 10mv.  I guess at these frequencies the margin of error becomes increasingly small.

What really throws me for a loop is when I can reach stability at different VCCSA voltages.   On my setup, 1.24 or 1.25 V are stable in all my OC profiles.  I stick with 1.24 V because at 1.25 V I get noticeable input delay in games.  This is the part that drives me mad.  

Is there a particular stress test that's really good for testing IMC/VCCSA voltage stability?  Should I rely on benchmark results at this point and pick the voltage w/ the better score?  When I reach this point and I'm passing all stability tests, I resort to what 'feels better' in game.  

Glad it helped.

VCCSA has always been sensitive to voltage changes. This is easier to replicate the higher we push frequency as the signal margins are much tighter. How much voltage is required depends on the CPU sample, memory kit used and the board. The latter is because if traces are precisely length-matched with an appropriate impedance for both the memory controller and memory module output stages it can enhance the available overhead for overclocking. This is partly why less optimisation is needed targeting lower frequencies on the motherboard's memory QVL list.

For memory stability, I normally stick to Karhu RAM Test. If the VCCSA voltage is an issue, the system will often throw an error fairly quickly.

13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

Thanks again for the info.  All of this is very helpful as I wrap my head around all of intricacies of DDR5 overclocking.  At this point I'm maining a 8000 profile I tuned just due to the significantly lower MEM VDD required to stabilize it.  I also found that I had to manually tune the CPU voltage for default settings.  Default settings were jamming way too much voltage into my CPU.

I've used pretty much every motherboard brand there is and it seems Asus really cranks up the voltage @ default settings.  I wonder if this is overdone?  It's definitely a bit frustrating when your expectation is that default settings should just work, but that just wasn't the case in my situation.  Sure, I passed stability tests, but everything felt absolutely awful so it's moot.

 

Silent_Scone
Super Moderator

You're welcome.

It's not so much "too much voltage", just the vendors need to weigh up what's going to work best for a larger proportion of CPUs. Unfortunately, there's no magic bullet with these things - so auto values aren't always going to be the best fit for some CPU samples. Some of these things aren't avoidable when running components outside of specification.

13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

Yeah, that makes sense.  I guess there's only so much you can do when so much of it relies on the CPU's IMC which is completely out of the vendor's hands. 

Kind of related to this topic... my 8000cl38 OC Profile is no longer stable after yesterday.  At first I thought it was due to my CPU possibly not being stable under default settings due to the fact that I rigorously stress tested my manually tuned 8000cl38 profile.  I passed an overnight run of tm5 1usmus_v3 and 3+ hrs of VST, 1+ hr of karhu.  I figured w/ the overnight 8-10/hr run of TM5 1usmus_v3 I was good to go... but sure enough, I retested again w/ VST and it failed in 8m.  I just don't understand how this is even possible, it's incredibly frustrating.  The profile that is now failing had these settings (VCCSA 1.2 / Mem VDD/VDDQ 1.44/1.35 / TX 1.3 / IMC VDD 1.406).  I am now re-tuning this profile and now I need (VCCSA 1.24 / TX 1.34 / IMC VDD 1.418).  I just began re-tuning yesterday so I hope to finish the 're-tune' by today.

This kind of behavior drives me absolutely mad.  Is this normal or is this simply due to the fact I'm pushing high frequencies on a platform with poor signal integrity issues?  The fact I was able to pass stress/stability/memory tests consistently makes this even more confusing.  I wonder if my methodology needs to be reworked.  It also begs the question whether the stability tests I'm using are flawed in some way.  I began incorporating OCCT AVX2 & SSE tests as well since I've read that AVX2 is good for testing DRAM voltages and SSE is good for IMC voltages.

There’s numerous reasons an OC can become unstable. Best advice I can give is don’t hammer the system too much with things such as VST. There will be a short bed in period with these things depending on how close you are too the edge of stability.

For memory, on my gaming system I  usually stick to Karhu RAM Test between 4000-10,000% coverage, and never had any stability problems. 

13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090

Okay, that's great to know!  I'll change up my stress testing methodology a bit because I keep getting mixed results when relying ultimately on passing VST in the end.

After about another 1/2 day of tuning, I found my existing 8000 profile I had tuned (SA 1.2 / TX 1.3 / IMC VDD 1.408 / MEM VDD/VDDQ 1.44/1.35) was in fact not stable.  I deemed it stable initially due to passing VST and an overnight TM5 1usmus_v3 and 1hr+ of karhu but it must have been on the edge of stability.  I dialed up voltages across the board and I'm currently at (SA 1.26 / TX 1.36 / IMC VDD 1.412 / MEM VDD/VDDQ 1.48 / 1.38) and I appear to be approaching stability again.  Games feel great and responsive again so that's a huge plus.

I went in to this w/ the notion that finding the lowest possible VCCSA was ideal and I think this probably made things more difficult for me.

Majority of 14th gen samples prefer around 1.2 to 1.25v SA

13900KS / 8000 CAS36 / ROG APEX Z790 / ROG TUF RTX 4090