MemTest86 not consistent ?

[leclod]

So I've been running MemTest86 a lot lately and I noticed inconsistencies,

The first ones I noticed were those L3 Cache and Memory speed reports.

Then I noticed a full successfull Memtest lasted sometimes less than 4h and sometimes more than 5h.

You can see here for example, one time 54% of test 5 in 3mn25s, the next time 56% of test 5 in 5mn8s

I'm pretty sure those 2 tests were done using the same settings (which seem stable so far)

Any ideas ?

Specs :
Intel 12400
MSI B760M MAG Mortar DDR4 Wifi
RTX 3060
PNY 2*16GB XLR8 at 3466MHZ 16 18 18 36 1T Gear 1 1,35V

[David (PassMark)]

Yes, this is a bit strange. I also note that the benchmark GB/sec numbers for the RAM speed were pretty different.
(but the clock speed of 3466MT/Sec was the same).

Can you post the debug logs (they should have already been stored on your USB drive)
https://www.memtest86.com/tech_debug-logs.html

[leclod]

Can you post the debug logs (they should have already been stored on your USB drive)
https://www.memtest86.com/tech_debug-logs.html[/QUOTE]

Those should be it

(current setting wasn't stable, I got errors last night, now I'm at 3466 17 18 18 38 2T Gear 1 1,35V)

[David (PassMark)]

From the two log files.

2023-06-27 14:03:30 - MtSupportRunAllTests - Test execution time: 329.909s (Test 7 cumulative error count: 0, buffer full count: 0)

2023-06-27 12:52:13 - MtSupportRunAllTests - Test execution time: 26.296s (Test 7 cumulative error count: 0, buffer full count: 0)

So there certainly is a big difference in the test execution time. A similar pattern is seen in the other tests as well. So it isn't random and isn't just one test.
Logical conclusion is that your system's performance isn't consistent. Either some settings were different, or you have thermal or power throttling.

[leclod]

From the two log files.

2023-06-27 14:03:30 - MtSupportRunAllTests - Test execution time: 329.909s (Test 7 cumulative error count: 0, buffer full count: 0)

2023-06-27 12:52:13 - MtSupportRunAllTests - Test execution time: 26.296s (Test 7 cumulative error count: 0, buffer full count: 0)

So there certainly is a big difference in the test execution time. A similar pattern is seen in the other tests as well. So it isn't random and isn't just one test.
Logical conclusion is that your system's performance isn't consistent. Either some settings were different, or you have thermal or power throttling.

I understand, you must be right. But it sure isn't obvious.
The cpu temps are far from thermal throttling level.
Power throttling ? I'm far from power throttling too, with my frugal 12400 (I think).
I think my settings were the same but I'm not 100% sure.
(I remember reading somewhere about clearing CMOS between each different setting, which I didn't do. This would be too much of a hassle in my config.)

[leclod]

Can you tell me which are the right speeds, is it L3 Cache 27.6GB/s and Memory 3978MB/s or is it L3 Cache 10.3GB/s and Memory 18.5GB/s ?
So I know where I'm at when I test... (the first set makes more sense to me, cache should be faster than Ram)

[leclod]

Sorry to insist, but I might have further interesting enough information.
My system is otherwise stable.
I did undervolt my cpu by -0.020 before but I've reduced to -0.010 according to your instability claim.
I had one error in Memtest so I tried increasing Ram voltage to 1.37V.
I got the first successful test in 4h43 (T1 0h47 T2 1h22 T3 1h22 T4 1h12)
The next day I got back to an undervolt of -0.020 and reduced the Ram voltage to 1.36V.
I got the second successful test in 3h41 (T1 0h47 T2 1h06 T3 0h54 T4 0h54)
So the more stable system takes the longest.
Can the Intel standard energy saving features explain such differences ?
Thank you

[David (PassMark)]

is it L3 Cache 27.6GB/s and Memory 3978MB/s or is it L3 Cache 10.3GB/s and Memory 18.5GB/s ?

The 2nd faster set of numbers is more typical, but still low.
Cache speeds of around 50GB/s are possible. And L3 cache should be faster than main RAM.
(3978MB/s is actually slower than some current hard drives)

So neither set of numbers is OK in the end.

[leclod]

For information only,
I came across a more logical display. The speeds shown, although probably still low for you, at least make sense.