Memory speed is incorrect when running Memtest86

[Anthony.Lee]

Hello,
We are using the following configuration to run the MemTest86 memory test. We noticed that on the first boot, the memory speed is correct and the test runs normally. However, after rebooting the system and running MemTest86 again, the reported memory speed drops. Interestingly, other tools and burn-in test do not show this issue.
Have you seen a similar problem before?

Configuration:
CPU: Core Ultra 7 165H, 5.00GHz, 16-Core
Memory: Micron MT62F2G32D4DS-023 LPDDR5X-8533 64GB ( Memory Down -on board)

[David (PassMark)]

You are using an old release of the software. V11.3 is the current release at the moment.

Modern memory can run at variable clock speeds. Dynamic Memory Boost (on some platforms) and Dynamic Voltage and Frequency Scaling (DVFS), allow the memory controller to adjust the RAM's clock speed and voltage dynamically based on workload demands.

MemTest86 only reports the clock speed at a certain instance in time (when it is booting up). Might be interesting to compare the GB/sec figures to see if it is really running slow all the time, or just for a short period.

[Anthony.Lee]

Thank you, We will use the new version and saw the real speed

[Anthony.Lee]

We used the new version 11.3, and the problem still occurred.
But the GB/S values in the actual test do not seem to be different.

[David (PassMark)]

It is also worth noting that MemTest86 never sets the memory speed. It reports on the current value, but never changes it.

So it should be ALWAYS using the speed set in BIOS, even if what is displayed appears incorrect (obviously we would like to display the correct current value however)..

BIOS determines the clock speed (based on the firmware programming, the CPUs capabilities and the RAM's SPD data).

The GB/s value should be significantly different if the RAM was really running at 5586 instead of 8533Mhz.

You can post the 11.3 debug log if you like and we can take a look at it.

[Anthony.Lee]

I attached the log file.
Thank you

[David (PassMark)]

I assume that log was from a 2nd incorrect run?
Can you also post a log from when the memory speed looks "correct". As the 5586Mhz value is totally plausible as the correct value.
It is probably also worth asking, what is it you consider to be correct and what values do you have set in BIOS?

[PWSR]

Can I jump in with my speed issues?

I tried to figure this out myself, but am raising the white flag. Can you see what you can discover, please?

Scenario
Testing SODIMM DDR4 3200 modules in laptops
Purchased and downloaded newest V11.3 yesterday
Found a couple laptops that can handle 3200: 11th gen plus generally
Learned that Memtest will run at system speed regardless of what is displayed
Verified 3200 speed in BIOS for both systems
One system displays 1066 MT/sec (SLOW) and the other displays 3198 MT/sec (FAST and ideal)
But aren't they supposed to be running at the same speed? The slow takes a LOT longer to complete.
Is CPU clock factored in? If so, why is it running so slow? (1.4 GHz) And is there any way to make it faster?

Fiddled with BIOS settings - Tried limiting cores. Turned off all power saving options.
I have suspicions that the key here is the i5-11500H in the fast system. Pretty sure those don't have E cores. But not sure why or how that can be explained.

Ideal solution looks like...
- Completed tests in as little time as possible
- Confidence that modules are being tested at 3200
- optional: ability to use CHECKMEMSPEED=JEDEC parameter

FYI - I'll be out of the office about 30 minutes after posting this. I'll likely review and respond tomorrow morning.

Thanks for the assist!

[David (PassMark)]

PWSR, those two laptops don't appear to be identical. Different CPUs, different number of CPU cores, different RAM.....
So there is no reason to think they will run the same.

Also in the fast log, it found a memory error at Test 3 and stopped after ~20 seconds. In the slow log it found no errors and did a complete pass (obviously taking much longer). This appears to be the major difference in the logs.

There was this in the slow log however (the base clock frequency being an invalid value of 0xFFFFFFFFFFFFFFFF hexadecimal)
So we will take a look at that.

2025-05-07 23:16:46 - poll_timings_tigerlake - BCLK_FREQ=FFFFFFFFFFFFFFFF
2025-05-07 23:16:46 - poll_timings_tigerlake - invalid BCLK_FREQ, using default BCLK 100.0MHz

In the fast log it appeared like this (a valid value)
2025-05-08 10:41:32 - poll_timings_tigerlake_h - BCLK_FREQ=000185AC000185AC (99.756MHz)
2025-05-08 10:41:32 - get_mem_ctrl_timings - [0-0-0] 3192 MT/s (22-22-22-52)

[Anthony.Lee]

I attached the PASS log file. Thanks.

[PWSR]

David, thanks for the quick response. It's impressive how you look after your forum.

RE: Minimizing Test Times - I understand that different laptops will result in different completion times. I was mostly interested in the why. (Which make/model/specs should I choose to get the best results?) Combining your answer and further testing got me to a place I'm satisfied with. Understanding these two variables helped.

1) Intel Speedstep BIOS Setting - Turning this back ON in the laptop with the i5-1135G7 was the most noticeable difference. It didn't change displayed CPU clock speed or MT/sec, but memory GB/sec almost doubled and test time went down dramatically. I had thought that this feature was just a way to make your system use less power and go slower. I was clearly wrong. Surprisingly, changing this setting made no difference on the LT with the i5-11500H.

2) The performance increase moving from 1Rx16 to 2Rx8 is significant.

RE: Confidence in Test Speeds and Using CHECKMEMSPEED - I believe you that they are running at system speed, but when the logs actively say otherwise, it's hard to squash that doubt.

- If one out of 50 tests fail, Murphy's Law says that would be the log I pick to send, right? I should have checked it.

- Today I was trying a different HP laptop that also has an i5-1135G7. CPU clock speed was displaying higher and test times were lower than yesterday's Dell. Maybe there's some kind of Dell-specific issue? Verified BIOS reporting 3200, but same result with timings: invalid BCLK and 1066 MT/sec. Attached new log here. I'm pretty sure it's good.

- Also noticed the HP sets aside 15.3GB to test versus the Dell's 15.7GB. Is there a quick and easy explanation for that?

- Also tried a laptop with DDR5. I didn't dig into the results, but attached the log in case it helps. For reference, modules are 5600, but max speed on that CPU is 4800. Memtest reported 2194 MT/sec.

- Looking forward to an answer to the BCLK error and/or improvements to the display of timings.

[David (PassMark)]

For fastest testing

1. Use a fast CPU
2. Use the highest RAM clock speed available (e.g. turn on XMP).
3. Use dual channel mode (so 2 x 8GB is faster than 1 x 16GB)
4. Try and avoid some of the older machines with UEFI bugs that prevent multi-tasking (i.e. testing with only 1 CPU core)
5. Only run 1 pass of MemTest86 instead of 4 (but this is a trade off, you might miss some errors)
6. Turn off any power saving modes (+run on mains power instead of battery for a laptop)

Also noticed the HP sets aside 15.3GB to test versus the Dell's 15.7GB. Is there a quick and easy explanation for that?

Not all RAM is available for testing. The UEFI BIOS (which is a mini-operating system) and other devices in the system use some RAM.

Also tried a laptop with DDR5. For reference, modules are 5600, but max speed on that CPU is 4800. Memtest reported 2194 MT/sec.

It is often the case that the advertised max speed for a RAM module isn't used by default. So the reported numbers might be correct.
Either the CPU can't run as fast as the RAM, or the BIOS selects a slow conservative JEDEC speed (like 2194) instead of the max speed, which you might only get after turning on XMP profiles (or in some cases manually overclocking the RAM).

[Anthony.Lee]

Hi David,
Did you notice any differences between the Failure log and the PAS log?"

[PWSR]

It is often the case that the advertised max speed for a RAM module isn't used by default. So the reported numbers might be correct.
Either the CPU can't run as fast as the RAM, or the BIOS selects a slow conservative JEDEC speed (like 2194) instead of the max speed, which you might only get after turning on XMP profiles (or in some cases manually overclocking the RAM).

User error, huh? Ok, maybe this one will help. Modules are 4800. i5-1235U max mem speed is 4800. BIOS reports 4800. Memtest reports 1994. No need to comment on this log specifically. I'm only providing it with the hopes that it helps you find and develop a solution.

I have workarounds in place for now. I look forward to improvements in a future release. Thanks for you help.

[David (PassMark)]

The clock speed values appear to be correct, in that they correctly display the current clock speed (which varies up and down to save power). In the log you can see the clock speed and timings varying.

2025-02-15 15:18:01 - get_mem_ctrl_timings - [0-0-0] 1600 MT/s (22-22-22-52)
2025-02-15 15:18:54 - get_mem_ctrl_timings - [0-0-0] 1066 MT/s (15-15-15-35)

and in separate log (the HP DDR5 one) it does in fact sometimes log 4788 MT/s
2035-01-14 19:33:55 - get_mem_ctrl_timings - [0-0-0] 1994 MT/s (22-16-16-31)
2035-01-14 19:35:56 - get_mem_ctrl_timings - [0-0-0] 4788 MT/s (40-39-39-76)

It would obviously be nice to display the max value possible, (and this is what already happens on most systems). Maybe your setup is particularly aggressive in down-clocking for electrical power savings. We can probably update the UI to always display the max value if we see higher values post boot.