Maximus XI Hero: struggling to get DRAM stable @ 3600

[vadimzuev90]

Hi everyone, just as the title suggests I'm having some hard time with my DRAM and would like to ask the folks here for some advice.

The main components in question (complete specs can be found in my profile) are a Maximus XI Hero (WiFi), an i9 9900KF and 2 sticks of G.Skill's TridentZ Royal memory (F4-3600C16D-32GTRG to be precise).
BIOS version is 1401, dated 26 Nov 2019, I updated via Internet before tweaking any settings.

After going through some trial an error I kept the following main settings for the CPU: core mult x51, cache mult x45, vCore 1.33V. As for the RAM, at first I went with the "XMP I" profile ("ASUS optimized") but I was getting code 55 on boot quite often, so I switched to "XMP II" (Intel default) which seemed more stable.
Once into Windows, everything (including the DRAM @ 3600) looked OK, I ran a couple of hours of Prime95 in both CPU and RAM stress modes, as well as Cinebench, User Benchmark and HCI MemTest with up to 1000% coverage. Nothing crashed, nothing overheated.

Weird thing about the DRAM @3600: the BIOS seems to always overwrite whatever I enter for DRAM, VCCIO and VCCSA voltages.
When set to "Auto", I immediately get code 55 on boot, PC hangs for half a minute, then restarts and I get into UEFI seeing the following values entered into the input fields: DRAM 1.400V (actual 1.414), IO and SA both 1.35000V (actual 1.360).
Lowering the voltages manually (even just a bit) results in multiple unsuccessful start attempts, then the system eventually boots into UEFI and I see the same stuff as above.
Despite the obviously high voltages, I had no thermal issues, the temps seemed quite reasonable for the clocks, averaging 30°C idle, 45° gaming, 75° Cinebench and 85° Prime95.

The real problem: with those settings I still randomly get codes like 55 and 32 as well as multiple unsuccessful start attempts on (re)boot, after which I get into UEFI, exit without changing anything and eventually it boots normally (sometimes AURA settings are reset). Very annoying stuff. As previously said, once I get into Windows everything is rock-solid, no issues after extended gaming and working sessions (I use Android Studio with multiple emulators, it produces considerable CPU / memory load). Lowering CPU clock seems to have no impact on the issue.

My current solution: I set the DRAM to 3200 Mhz @ 1.35V, the rest to "Auto" which translated to 1.312V for VCCIO and 1.264V for VCCSA. Running those settings for several days, had no issues so far. Right now I only see 2 options:

Option 1: Live with that

Pros:

• no more headache
  
• possibly no significant performance penalty, except for benchmarks flagging the RAM as underperforming
  

Cons:

• knowing that G.Skill has some of the best DIMMs out there, can't avoid that bad feeling of not getting what I paid for
  

Option 2 (suggested by a colleague): contact the vendor and try to get DIMMs replaced as "not working/unstable/whatever"

Pros:

• potentially this may work out
  

Cons:

• even more headache
  
• the vendor might not agree
  
• no guarantee that replacement DIMMs will be any better
  

What are your thoughts guys, did I miss something? Thanks.

[mdzcpa]

Using XMP will always overwrite your manual settings. XMP sets the speed, memory timings and voltages.

What specific memory kit do you have (the gskill part #)??

Keep in mind that although the sticks may be rated for a certain speed, the memory controller on the CPU must be up to the task. Anything over 2666mhz on the 9900 CPU is technically overclocking.

I would try keeping your CPU at default speeds and then set your ram to XMP and see what happens. Sometimes an overclock or non default setting on the CPU can affect the memory controller. See if your sticks can hit 3600 with your CPU at default Auto settings.

[vadimzuev90]

mdzcpa wrote:
Using XMP will always overwrite your manual settings. XMP sets the speed, memory timings and voltages.

but I can at least use "Auto" with my DRAM running at 3200 (XMP II), while at 3600 the entry "Auto" is automatically replaced with the values mentioned above, here's an image to illustrate that (top -> bottom: DRAM, VCCIO, VCCSA)

mdzcpa wrote:

What specific memory kit do you have (the gskill part #)??

I have the F4-3600C16D-32GTRG kit, also made sure that my board is on the QVL before buying.

mdzcpa wrote:

I would try keeping your CPU at default speeds and then set your ram to XMP and see what happens. Sometimes an overclock or non default setting on the CPU can affect the memory controller. See if your sticks can hit 3600 with your CPU at default Auto settings.

I'll try your suggestion when I get home today, though indeed one wouldn't buy an i9 9900K(F) just to run it on stock.

[mdzcpa]

vadimzuev90 wrote:
I get home today, though indeed one wouldn't buy an i9 9900K(F) just to run it on stock.

I completely concur. The idea is to see if the memory will run at rates speed. If so you will know your cpu overclock is affecting your memory stability. *If they run at stock cpu speed, press your cpu overclocked up in steps.

[vadimzuev90]

It seems the CPU memory controller ist not the reason, I have reverted all settings to stock, then activated XMP I, on reboot I got code 55 and POST in safe mode...

UPDATE May 22th 2020: so, I didn't bother to return the DIMMs, the max stable speed I got to is 3300 Mhz. However it seems I was able to compensate a bit for the missing 300 Mhz doing 2 things: overclocking the cache to 5Ghz and loading the memory preset called "2x16 GB Samsung B-Die D-Side" - this applied quite decent timings which seem to be stable:



At least UserBenchmark now reports the RAM as "performing as expected" despite the fact the modules are compared with identical DIMMs running at 3600. I think I can live with it 😉