Maximus X Hero memory overclock issue / help request

[Gre4ka]

Hello, guys.

I moved some weeks ago from my old s1366 westmere rig onto the new one. I got i7-8086k, Maximus X Hero mobo, 32gb (16x2) 3200CL14 memory from G.Skill and Samsung 970Pro 1Tb SSD, other parts remained unchanged (1080Ti etc, HDD's etc)

I successfully delided my 8086 and replaced default TIM with LM using RocketCool tools. I use a custom waterloop in my setup and applied a Monoblock fullcover from EKWB to cool my CPU and mobo VRM's. First I decided to overclock and test my CPU, I got 5GHz easy with 1.3v on Vcore LLC6, AVX offset 0, CPU cache 4700. Temps never rise higher then 75-77C under LinX (linpack) AVX and Prime95 small fft's with AVX.
Then I pushed my memory to it's specifications - 3200MHz, 14-14-14-34 CR2, I haven't used XMP, I set primary timings manually and left #2-3 subtimings in auto with maximus tweak at Mode 2. It's rock stable in this mode and works fine. I'm getting such results in AIDA64:



Then I tweaked some subtimings a little and got some additional boost in performance:



It's still rock stable at this settings, passing memtest86 for hours and prime95 large fft's for about half a day.

And then... I decided to test my memory to get some higher MHz with different timings to see what I can get. And I realized that I can't move any higher.
I tried 3600 with 16-16-16-36, 17-17-17, etc. Then 3866-4000 with CL 18-19 - no way.
CR2, maximus memory tweak mode 1/2, VCCIO/SA at 1.2-1.25 and pushed up to 1.4V on mem. Just no way. All I usually get is 55 on Q-Code table. Or (depending on timings) random 4X errors or B1 or D5 or AX. On higher frequencies with tighther timings and mode2 it's usually "55" in other situations several retries with B1/D5/4x errors without reboots and then I got an "overclocking failed" and press F1 invitation.

Yesterday I updated my BIOS from 1602 to the last 1704. Things become a little better. I was able to reach 3600 16-16-16-36 CR2 but only with Mode 1. In mode 2 i getting "55" instantly. And even this settings not reliable. It's not always start from the first time getting several retries (w/o reboot) passing through B1/D5 on Q-Code and only after 2-3 autoretries - boots. I find some guides on subtimings and tried to set them manually. Nothing changed in stability but performance rised significantly. But anyway it's not stable. VCCIO/SA voltages and mem V under 1.37-1.38 doesn't help. It's not stable in prime95 and hangs/reboots in 2-4 hours.
Here are performance results:



But until it's not stable all of this is useless.
With 1704 bios I more often get B1/D5 errors instead of "55" when I try something like 3733-3866-4000 MHz and differend timings.
Several people that I know pushed their G.Skill 3200CL14 32Gb kits (16x2) at something like 4000-17-17-17 on even cheaper mobo's like 370Z Strix. Not on Apex.
I can't understand where is issue. Could someone help me? Maybe there are some "secret" settings in BIOS that should be turned on/off or adjusted? I can attach my bios settings (1704) profile. Or I may post my settings as pictures. There are a lot of memory/CPU/overclocking related settings that I neved heard before and/or I can't find any valuable info on them. For example here is screenshot of setting from guy that pushed his 3200CL14 to 4000CL17:


Is my CPU IMC so bad? Or samsung B-die's in my memory love only low MHZ with low CL? Or I just missed some valuable settings in bios that are critical?

Please help 🙂 Thanks in advance!

[bass_junkie_xl]

hello , ya x hero wifi her with bios 1704 has been the best bios for me for memory over clocking .
my kit if g.skill rgb trident z 16gb 2x8gb for 3000 mhz cl-14-14-14-34 1.35v xmp
oc to stable 4000 mhz cl-17-19-19-39 1.4v vccio =1.21v vcssa=1.21v 4000 % pass in hci mem test /prime/asus rb

i can run them up to a stable 4200 c19 but thats this kits max before it pewps.

here is my scores for 4000 c-17-19-19-139 1.4v , 8700k @ 5.1 ghz avx offset - 1 , @ 1.344v adaptive , cache @ 4.4 ( havent got it it yet )



you could try under extreme tweaker page @ the bottm set eventual dram volts to 1.39v and try vccio = 1.2 - 1.25 v , vcssa 1.2 - 1.25v maximus tweak mode 1
your 3200 c 14 kit im scoreing the same latency but higher read/write @ 4000 if u need help i could send you my 4000 mhz profile and you could test it ?

[Gre4ka]

bass junkie xl
Thx for your reply. Unfortunately you use 16Gb kit (8x2). Such mem much easier in overclock.

I spent yesterday about a half of day and was able to reach 3466 CL15 fully stable with minimal subtimings.

But anything higher results in instability. I was able to start at 3733 CL16, but mobo usually comes through several retries and memory "training" giving me D5 / B1 and other Q-Codes 2-3 times before it starts. For 3733 CL16 I had to set ALL subtiming with hands, especially IO-RTL. Because a huge fluctuations happens from one restart to another.
Anyway I get a lot of errors in Memtest86 test #6 (block move).
I tried different voltages on VCCIO/SA - realised that 1.15-1.20 range giving me best chance to start and minimal retries/Q-Code errors, also minimal #6 test errors in memtest (2-10 per cycle). Anything less than 1.10-1.15 - just no start and exotic Ax Q-Code errors. Anything higher than 1.2-1.25 also exotic Q-Code errors and if it starts - memtest giving me more errors. I tried to increase Vcore a bit - to 1.33-1.35v - nothing changes. The most confusing thing - voltage on RAM. If I set even a bit less than 1.35v (1.33) - no boot or massive quantity of errors in #6 test and others. And.... if I set anything higher I'm not able to boot. Different Ax errors in Q-Code. Tried 1.37-1.40.
This really confuses me...
I can't realise what's stopping me - not good IMC in CPU or very very specific memory kit that don't like high MHZ, but likes good low timings. Or maybe both.
I don't think that it's mobo issue, because a lot of people overclock their mem on much cheapier boards like Strix etc. Setting are mostly the same. And I don't think that all this overclocks done with setting manually all secondary/tretiary settings by hands. This can improve performance or stability, but random Q-Code errors, nor non-stable starts (first indicator that something is unstable and will not work well) and sensless reaction on ctitical memory OC voltages...

Here is my results with 3466 CL15 and maximum titened timings:



used asrock mem configurator as it's more suitable to show all timings/subtiming on one page.

[bass_junkie_xl]

try vccio 1.21 / vcssa 1.21v in bios go to extreme tweaker under eventual dram volts set 1.39v
asus tweak mode 1 ,

found this beauty page ! it has helped me get better oc

follow problem # 1 ( code 55 with 4 sticks )
- basicly chnage timings in bios under dram timings go to timings # 3
-change these 4 to match your rams cl ( example mine is cl -17 so change all 4 to 17 or what ever your first cl # is )
-tRDWR_sg= your cl first #
-tRDWR_dg=your cl first #
-tRDWR_dr=your first cl #
-tRDWR_dd=your first cl #

link: https://community.hwbot.org/topic/147255-guide-skylake-memory-timings-on-asus-motherboards/

read through it try my settings but chnage those 4 settings to your cl level example 4000 cl-19-19-19-39 you change all 4 of those i mentioned above to ( 19 )

this little guide helped me get around 4 thousand more read write and around 4 ns less just changeing a few things .

my kit is 3000 mhz cl 14-14-14-34 1.35v xmp it scores around thr 40k read write range and ns=45
my 4000 tuned profile ( tuned a bit not heavily tuned ) scores the picture above with 41.8 -42.2 ns

[Gre4ka]

After a days of research and testing different patterns/settings, I realised that most of modern kits are already pushed to their limits on the factory.
Also very many people think that they made a stable overclock, using unappropriate soft/utilities to test their mem overclock & overall stability.

My conclusion:
my kit can work only on specified speed & timings. 3200-14-14-14-34-2N 1.35v
MTM mode 2 seems to work (need to test several hours more)

memtest86 will help to find bad sticks (defective) or absolutely unstable profiles and prevent you from OS installation / data corruption, but not more.
Linx (Linpack) worked for me on 3600CL16 on which I got a lot of errors using special utilities and even able to boot 50% of all times.
Prime95 in custom mode (512-4096 with 80-85% of amount of RAM to test) will show unstability well, but for me it worked for hours w/o any error.

Now I use a MemTest from HCI Design https://hcidesign.com/memtest/
as recommended I start # of copies that equal to my CPU cores/HT cores (12 in my case) and set a part of memory for each copy, leaving about 2-2.5Gb for the system. I use 2380Mb in my setup, this leaves 2.2Gb of RAM for windows.
Free version works perfect, but it will stop after finding first error. This will not allow you to check the "progress" of adjustments and see if errors amount decreases or not.

This will also test your VCCIO/SA required voltages very well. I got 1 error after 500% of test coverage with VCCIO 1.1v and SA at 1.2v, now retesting with VCCIO 1.15v

also never follow the advices to lower tRFC and tREFI timings to minimum/maximum. In Mode2 motherboard sets tRFC to 450, I get a massive number of errors in 360-400. The same with tREFI - 65535 no go. Defaults to 11xxx. Yes, they make a massive (~10%) amount of performance boost, but...

So, most "overclocks" are not 24/7 stable and will show mem errors in memtest / prime95 after a hours of test. I prefer to tune to mem specs that tested by manufacturer.
The rest are for benchmark purposes only.

[NemesisChild]

I've tested 3200 & 3600 kits, the 3600 kits hands down can overclock a whole better.

I can easily hit 4000 (same CL16 timings) with my 3600 G.Skill set simply by bumping the DRAM voltage to 1.385v. Mode 1 or 2 aren't needed.

Having a good CPU IMC obviously helps too.

[bass_junkie_xl]

NemesisChild wrote:
I've tested 3200 & 3600 kits, the 3600 kits hands down can overclock a whole better.

I can easily hit 4000 (same CL16 timings) with my 3600 G.Skill set simply by bumping the DRAM voltage to 1.385v. Mode 1 or 2 aren't needed.

Having a good CPU IMC obviously helps too.

G.skill.tridnet z non rgb 3600 cl 16 16 16 36 got to 4000 cl 16 @1.385 v stable in hci ?wow my rgb trident z 3000 cl 14 can do 4000 17 19 19 38 1.38v 4133 cl 19 .
I might swap to a higher end kit soon i think

[HiVizMan]

NemesisChild - the IMC on processors is so varied it is really difficult for folks to truly understand that fact and it frustrates them. When I bin CPUs for benching it really strikes a nerve. IMC is for me as important as how well the CPU can OC. In normal everyday life maybe more so.

[Gre4ka]

NemesisChild
HiVizMan

thanks for your opinion
but question was about 32gb (16x2) dual ranked modules
of course 8x2 (16gb) setups overclocks better

but imho 16gb is too small for modern pc & games
32gb allows to fully turn off windows swap & etc and never think about any loading/unloading from memory during gaming or work

[HiVizMan]

I hear you Gre4ka - I can confirm your findings on the lower tRFC and tREFI timings by the way. Had the same thing.

Lastly for benching or running benchmarks you only need your system to be stable enough to give you the result. I would never consider any system I was benching at the very edge or performance as being stable. But then what is stable in electronics/PC's components can just die for what seems like no reason or cause. Be working one moment and then stopping.

32MB and 64MB memory is always going to put more workload onto the IMC and be harder to overclock. You are correct I think that for most memory they are right at the upper limit as sold. But we keep trying to get more performance for our money...

Enjoy your weekend.