Memtest86 failed advice

[Darkwaxer]

^(MB: MSI MPG Z790 EDGE WI-FI DDR4)

^(RAM: Corsair Dominator Platinums DDR4 3466 MHz - 4x16GB)

^(CPU: I5 - 13600K)

^(PSU: Corsair RMx1200 Shift GPU:)

^(Merc 7900 XTX)

^(XMP profile that wouldn't boot: DDR4 3466 MHz 16-18-18-36 1.350V)

^(Stock RAM: 2133 MHz 1.196V (CPU 3.5GHz))

^(OC that booted: 3200 MHz (CPU: 5.2 GHz))

 

I found out yesterday that my Motherboard isn't completely compatible with my RAM when enabling XMP. MB supports the 32gb 3200MHz variant across all four DIMMs but not my 64gb 3466MHz across 4 DIMMs (CMT64GX4M4C3466C16), only on two DIMMs.

 

After some advice I clocked the RAM using MSI Try It! feature in the bios and also switched on the OC (or game mode) for the CPU and got the system to boot with the CPU at [5.2.GB] and the RAM at 3200MHz.

 

I reset the CMOS and ran MEMTEST86 last night and it passed with no errors to make sure my RAM was good and as recommended by Corsair here: [https://help.corsair.com/hc/en-us/articles/4413740029709-How-to-Run-Memtest86-to-check-for-RAM-faults](https://help.corsair.com/hc/en-us/articles/4413740029709-How-to-Run-Memtest86-to-check-for-RAM-faults)

 

So today I redid the overclock and ran MEMTEST86 again and got 6 fails over the 4 passes (I didn't remove any sticks of RAM, ran the test with the four sticks). I expected there to be issues as when I used MSI Try It! it didn't have an selectable option for 3200 MHz CL16 so choose CL15.

 

Here is the results:

^(Result summary)   
^(Test Start Time    2023-06-25 10:41:09)   
^(Elapsed Time    6:25:32)   
^(Memory Range Tested    0x0 - 107F800000 (67576MB))   
^(CPU Selection Mode    Parallel (All CPUs))   
^(CPU Temperature Min/Max/Ave    45C/56C/50C)   
^(RAM Temperature Min/Max/Ave)       
^(TSOD0    39C/48C/45C)   
^(TSOD1    40C/50C/47C)   
^(TSOD2    40C/50C/47C)   
^(TSOD3    38C/47C/44C)   
^(# Tests Completed    48/48 (100%))   
^(# Tests Passed    42/48 (87%))   
^(Lowest Error Address    0x1A9E80C64 (6814MB))   
^(Highest Error Address    0xDC0BAE8E4 (56331MB))   
^(Bits in Error Mask    0000000000000010)   
^(Bits in Error    1)   
^(Max Contiguous Errors    1)   
^(CPUs that detected memory errors    { 13, 14, 15 })   
^(Bits in Error)   
^(.    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    .    x    .    .    .    .)   
^(63                                                                                                                                                                                                                                                            0)   
^(Test    # Tests Passed    Errors)   
^(Test 0 \[Address test, walking ones, 1 CPU\]    4/4 (100%)    0)   
^(Test 1 \[Address test, own address, 1 CPU\]    4/4 (100%)    0)   
^(Test 2 \[Address test, own address\]    4/4 (100%)    0)   
^(Test 3 \[Moving inversions, ones & zeroes\]    4/4 (100%)    0)   
^(Test 4 \[Moving inversions, 8-bit pattern\]    3/4 (75%)    1)   
^(Test 5 \[Moving inversions, random pattern\]    3/4 (75%)    1)   
^(Test 6 \[Block move, 64-byte blocks\]    4/4 (100%)    0)   
^(Test 7 \[Moving inversions, 32-bit pattern\]    2/4 (50%)    2)   
^(Test 8 \[Random number sequence\]    3/4 (75%)    1)   
^(Test 9 \[Modulo 20, ones & zeros\]    3/4 (75%)    1)   
^(Test 10 \[Bit fade test, 2 patterns, 1 CPU\]    4/4 (100%)    0)   
^(Test 13 \[Hammer test\]    4/4 (100%)    0)   
^(Last 10 Errors)   
^(2023-06-25 16:24:07 - \[Data Error\] Test: 9, CPU: 15, Address: 6F17E61E4, Expected: DB0D7844, Actual: DB0D7854)   
^(2023-06-25 16:08:32 - \[Data Error\] Test: 8, CPU: 13, Address: 1A9E80C64, Expected: 89284BF4, Actual: 89284BE4)   
^(2023-06-25 15:23:54 - \[Data Error\] Test: 4, CPU: 15, Address: B294F6164, Expected: BFBFBFBF, Actual: BFBFBFAF)   
^(2023-06-25 13:44:00 - \[Data Error\] Test: 5, CPU: 13, Address: DC0BAE8E4, Expected: AC13A275, Actual: AC13A265)   
^(2023-06-25 12:19:27 - \[Data Error\] Test: 7, CPU: 13, Address: ABB4F2014, Expected: FFFFFDFF, Actual: FFFFFDEF)   
^(2023-06-25 10:53:04 - \[Data Error\] Test: 7, CPU: 14, Address: 4DEE035E4, Expected: 00000080, Actual: 00000090)   

I'm not too sure what to do with this data. Presume the overclock isn't secure so should:

* I put my system back down to stock 2133 MHz;
* Remove two sticks of RAM and OC to 3466 MHz
* Try to adjust the latency to be CL16
* Just leave it?

Thanks for any input.

[RONOTHAN##]

What this type of this is generally referred to as is retrain instability. It's the most annoying thing about working with memory on 11th gen and newer Intel CPUs. I don't think I've seen anyone report this happening in Gear 1 though, so that part is a bit weird. Double check to make sure you're in Gear 1 and not Gear 2. 

 

A couple things you can try are setting some of the IMC voltages manually or checking a couple different BIOS revisions. IMC voltages are very useful for getting higher memory frequencies to work, especially with higher density configurations. The two most important for DDR4 are System Agent (usually abbreviated as SA or VCCSA) and VDDQ TX (MSI calls this CPU VDDQ). With DDR4, higher is usually better with System Agent, so setting this to 1.4 or 1.45V is probably a good idea (1.45V is the highest I'd run daily), though there have been some reports of it not scaling past 1.3V with some CPUs so your mileage may vary. As for VDDQ TX, the optimal value changes depending on the specific CPU, BIOS revision, and the exact motherboard you're using so you will need to mess around with it, though usually 1.35V is about where this needs to be and where I would start.

 

It would be helpful to know the exact timings that the 3466 CL15 preset has, since it could be that the timings are a little too tight for your memory, though there's usually 1-2 ticks available in each timing as a safety margin from the manufacturers to help make sure the XMP works (not always though, a lot of 3600 CL18-22-22-42 kits will not do any lower than 18-22-22-42). Also, knowing the memory IC present on the sticks themselves would be helpful, can you report back with the version number printed on the sticks themselves (it will look something like "ver. 4.31" or "ver. 5.43").

[leclod]

64GB is a lot, you need it ?

I'm currently involved in a similar process,

If I were you i'd try setting XMP, then changing frequency to 3200.

Then you should have DDR4 3200 MHz 16-18-18-36 1.350V

If this works then you can look further...

(I'm at DDR4 3200 MHz 15-17-17-36 1.370V)

Edited June 25, 2023 by leclod

[Darkwaxer]

3 hours ago, RONOTHAN## said:

What this type of this is generally referred to as is retrain instability. It's the most annoying thing about working with memory on 11th gen and newer Intel CPUs. I don't think I've seen anyone report this happening in Gear 1 though, so that part is a bit weird. Double check to make sure you're in Gear 1 and not Gear 2. 

 

A couple things you can try are setting some of the IMC voltages manually or checking a couple different BIOS revisions. IMC voltages are very useful for getting higher memory frequencies to work, especially with higher density configurations. The two most important for DDR4 are System Agent (usually abbreviated as SA or VCCSA) and VDDQ TX (MSI calls this CPU VDDQ). With DDR4, higher is usually better with System Agent, so setting this to 1.4 or 1.45V is probably a good idea (1.45V is the highest I'd run daily), though there have been some reports of it not scaling past 1.3V with some CPUs so your mileage may vary. As for VDDQ TX, the optimal value changes depending on the specific CPU, BIOS revision, and the exact motherboard you're using so you will need to mess around with it, though usually 1.35V is about where this needs to be and where I would start.

 

It would be helpful to know the exact timings that the 3466 CL15 preset has, since it could be that the timings are a little too tight for your memory, though there's usually 1-2 ticks available in each timing as a safety margin from the manufacturers to help make sure the XMP works (not always though, a lot of 3600 CL18-22-22-42 kits will not do any lower than 18-22-22-42). Also, knowing the memory IC present on the sticks themselves would be helpful, can you report back with the version number printed on the sticks themselves (it will look something like "ver. 4.31" or "ver. 5.43").

Thank you for you very detailed reply.

So I’m definitely in Gear 1.

 

So just to check before I change anything:

 

Change:

CPU SA Voltage to 1.3 to 1.45V

CPU VDDQ to 1.35V

 

My timings are: 16-18-18-36

Version: 4.31

 

Thanks once again.

 

[Darkwaxer]

3 hours ago, leclod said:

64GB is a lot, you need it ?

I'm currently involved in a similar process,

If I were you i'd try setting XMP, then changing frequency to 3200.

Then you should have DDR4 3200 MHz 16-18-18-36 1.350V

If this works then you can look further...

(I'm at DDR4 3200 MHz 15-17-17-36 1.370V)

Honestly.. extremely likely not. I bought the kit in 2018 when I think there must’ve been a pricing error, they were normally supposed to be £650 and they were on offer for £320. It’s just taken me this long to get around to building a computer, they’ve been sat around waiting for years.

 

Thanks I’ll give that a try also.

[RONOTHAN##]

1 hour ago, Darkwaxer said:

Change:

CPU SA Voltage to 1.3 to 1.45V

CPU VDDQ to 1.35V

Yup, that's what to change. 

 

1 hour ago, Darkwaxer said:

My timings are: 16-18-18-36

Version: 4.31

So you've got Samsung B die. Good news is that if you want to overclock your RAM, that's usually what you want as that does the tightest timings and usually scales with voltage. Bad news is it's the hardest IC to drive and has the hardest time clocking high, especially in quad rank configurations like you are. If you pull two sticks out you'll have a significantly easier time hitting 3466MT/s or even higher. 

[Darkwaxer]

13 hours ago, RONOTHAN## said:

Yup, that's what to change. 

 

So you've got Samsung B die. Good news is that if you want to overclock your RAM, that's usually what you want as that does the tightest timings and usually scales with voltage. Bad news is it's the hardest IC to drive and has the hardest time clocking high, especially in quad rank configurations like you are. If you pull two sticks out you'll have a significantly easier time hitting 3466MT/s or even higher. 

Thanks for that.

 

So overnight I enabled the standard XMP Profile but dialled down the RAM to 3200 (CPU SA 1.144V - CPU VDDQ 1.2V) and ran Memtest86 again. It passed surprisingly.

 

Keeping these settingsI have just ran:

15 min Stability test (for memory) on OCCT with no errors. (testing 80% but running 100% amount of memory currently)

15 min CPU Stability test: Extreme mode - data set large - no errors

15 min CPU Stability test: Extreme mode - data set small - test stopped after 10 seconds with CPU hitting 92c (had OCCT to stop test if it went above 90c)

 

‘Warning message - CPU Package sensor overheat’

Restarted the computer and checked the bios settings in prep for this post. Booted up and ran OCCT again:

15 min CPU Stability test: Extreme Mode - data set small - manually stopped the test after CPU hit 100c…

Checked the setting to make sure the 90c stop limit was on.. it was.

15 min CPU Stability test: Normal Mode - data set small - manually stopped the test when the CPU hit 98c after a few seconds.

 

Not sure why the safeguard stopped working. Going to try to undervolt the CPU and try again.