MemTest86 v9.5 Beta 1 Release (2022-09-30 Update - Beta testing is now closed)

[keith]

(Sep 30/2022 Update) MemTest86 v10.0 has been officially released. As a result, MemTest86 v9.5 beta testing period is now over. Thank you to our beta testers for the bug reports and feedback.

Downloads
(Sep 30/2022 Update) MemTest86 v10.0 Pro and Site Edition can now be purchased from our sales page. As usual, the Free Edition is available for download on the normal MemTest86 download page. Beta downloads are no longer available.

New Features

• Added new experimental memory test as Test 14 [DMA test]. This test exercises the disk controller's DMA hardware to perform memory access, bypassing the CPU. The motivation for this test came from discovering a defective RAM module that did not produce errors when accessed via the CPU, but failed when files were read from disk via DMA.

Fixes/Enhancements

• Fixed hammer test incorrectly running in single-sided mode in Free version
• Fixed clock speed measurement failure for ARM chipsets due to cycle count register not being enabled
• Fixed bug in reading ECC error count registers for various Intel/AMD Ryzen chipsets
• Fixed reading ECC error status register for Intel Tiger Lake-H and Alder Lake chipsets
• Fixed ECC detection on Intel Ice Lake-SP chipsets
• Fixed ECC support for Intel Rocket Lake chipset variant
• Added ECC support for Ryzen Zen 3 50h-5fh chipset
• Added support for retrieving Intel Ice Lake-SP CPU info
• Added support for retrieving Ice Lake-SP RAM SPD data
• Added support for retrieving Ice Lake-SP RAM temperature data
• Updated blacklist

How to report problems

Either make a post here in the forum, or send us an email at the address listed on our contact page. When reporting an error please provide as much details as possible. If you are running on a USB drive, there should be a log file that has been generated in the EFI/BOOT directory called MemTest86.log. Sending us this will be of great help. Additionally a photograph of the problem would also be useful if possible/applicable.

[g126]

Great idea - DMA test!

I've downloaded it and I'm running it now, but I find it to be very SLOW... It's 3h55m and it's about 34% though the 3rd pass.

It usually takes around 4/5 hours to go through the whole cycle of tests 4 times.

Specs: 12900KS with 32GB of DDR5@6000MHz.

[g126]

Great idea - DMA test!

I've downloaded it and I'm running it now, but I find it to be very SLOW... It's 3h55m and it's about 34% though the 3rd pass.

It usually takes around 4/5 hours to go through the whole cycle of tests 4 times.

Specs: 12900KS with 32GB of DDR5@6000MHz.

Took 7h36m to do 4 iterations of this test (see report attached).

[David (PassMark)]

Thanks for the feedback.

What type of USB drive was this? (Was it a USB3 drive in a USB3 port?)

I just noticed the SPD RAM details are wrong as well for your DDR5 RAM, "2GB DDR5 PC5-4800"

Can you Email us the debug log.

[Jakob Rößler]

Is there an estimation available, when version 9.5 will be finally released that this is also usable with the pro-version?

[RedCrab]

ASUS B550m-C / Ryzen 5600x / DDR4-3200 32GB
8GB USB-3 flash drive on USB 3.1 port.
Time to complete 3 full passes was 6:55.

What are you using as the source device for the DMA test?
Looks as though it may be the flash drive.

[David (PassMark)]

Looks as though it may be the flash drive.

Correct. Can't be sure if it is safe to write to the internal hard drive. It would risk data loss.

[nicknack]

Two HP notebooks, one older (Folio 9470m) the other current.

On the older one, Test14 hangs hard immediately -> reset necessary.

On the newer device, the test runs with 8GB Ram with an HDD via USB2.0 in about 1h 45min. Protocol attached.
(Later correction: I wanted to send, but pdf is rejected).

One question: How much, how often and where is written to the connected drive (normally a stick)?

Sorry for the possibly bad English. Machine translation.

[David (PassMark)]

Can you post the log files for both cases, MemTest86.log. See 1st post for details.

[nicknack]

I was wrong.
The older notebook was was probably NOT frozen. Maybe I was impatient...

But now the newer notebook froze during another run. (this time for real)
Attached both logs.

To my questions above: In the meantime I found the additional partition for the DMA test. So that's cleared up. But I would still be interested in the amount of data written to the disk during test 14.

[jama]

I was testing 9.5beta for last couple days (my workstation is showing 1 bit errors, but only when all 4 modules are used, each with 32GB - testing 2 and other 2 modules separately didn't show an error).

Here are some things I've noticed:
1) Running memtest from USB drive is _significantly_ slower compared to running it from EFI partition on NVME drive, I had 9.4 version in NVME and 9.5beta on USB drive an I was assuming that the beta version has some extra debugging enabled, but now I've tried with 9.4 version on USB drive and it's also very slow. It's mostly visible when it's starting (like detecting memory controller and SPD information, before the main screen is even shown, it even looks like it got frozen, before it shows the first line, then it takes couple minutes, compared to maybe half minute when running from NVME. Doesn't it read the memtest binary once from the disk and then it should run from memory?

2) When running from NVME (grub loading memtest as gentoo ebuild installs it https://github.com/gentoo/gentoo/blo...bin-9.4.ebuild + https://github.com/gentoo/gentoo/blo...t86-bin-grub.d) it doesn't allow Benchmark results, test results, screenshots or MemTest86.log files to be written in EFI partition (while this works when running from USB). Am I or gentoo doing something wrong? It looks similar to https://www.memtest86.com/tech_configuring-grub.html

3) F12 for screenshots is great (much better than me taking photos of the screen with phone), but it's unfortunate that it works only in some menus (doesn't work when test is running or when there is "press any key to continue" message, I would rather press just Enter or Space and let F12 bound to screenshots at any time).

4) If I start memtest with monitor (LG OLED TV) connected to my nvidia GPU over HDMI and shutdown the TV over night, then in the morning when I turn the TV on, it doesn't receive any signal over HDMI (so cannot see the results from over night test run). If I connect different monitor (LG LCD) over DP, then it still doesn't show anything. But if I switch from HDMI to DP before starting the memtest (before going to bed) then in the morning it works fine and I can see the output.

5) It would be great to show if the memory is running in single/dual/quad/octa channel configuration, it's possible to figure out from the shown memory info about populated slots (if you know your motherboard or have the motherboard manual at hand), but clear indication in the results protocol would be useful.

Attaching 3 MemTest86.log files (2 from 9.5beta and 1 from 9.4, all from USB drive, because of issue "2)").

Thanks for great tool! Hopefully it will help me RMA my modules successfully .

[jama]

Also

6) it shows only 16 cores on 32c/64t AMD 3970X with these warnings in the log file:

2022-08-02 17:00:46 - This platform has 64 logical processors of which 64 are enabled.
2022-08-02 17:00:46 - Warning - The number of processors (64) is greater than the maximum supported (16). Forcing the number of processors to 16.
2022-08-02 17:00:46 - Number of hyperthreads detected: 0
2022-08-02 17:00:46 - Warning - the number of hyperthreads detected (0) is not half the total number of processors (16)
2022-08-02 17:00:47 - Testing MP support
2022-08-02 17:00:47 - MPSupportTestMPServices - AP dispatch test
2022-08-02 17:00:47 - MPSupportTestMPServices - Starting AP#1
2022-08-02 17:00:47 - MPSupportTestMPServices - Starting AP#2
2022-08-02 17:00:47 - MPSupportTestMPServices - Starting AP#3
2022-08-02 17:00:47 - MPSupportTestMPServices - Starting AP#4
2022-08-02 17:00:48 - MPSupportTestMPServices - Starting AP#5
2022-08-02 17:00:48 - MPSupportTestMPServices - Starting AP#6
2022-08-02 17:00:48 - MPSupportTestMPServices - Starting AP#7
2022-08-02 17:00:48 - MPSupportTestMPServices - Starting AP#8
2022-08-02 17:00:48 - MPSupportTestMPServices - Starting AP#9
2022-08-02 17:00:48 - MPSupportTestMPServices - Starting AP#10
2022-08-02 17:00:49 - MPSupportTestMPServices - Starting AP#11
2022-08-02 17:00:49 - MPSupportTestMPServices - Starting AP#12
2022-08-02 17:00:49 - MPSupportTestMPServices - Starting AP#13
2022-08-02 17:00:49 - MPSupportTestMPServices - Starting AP#14
2022-08-02 17:00:49 - MPSupportTestMPServices - Starting AP#15
2022-08-02 17:00:49 - MPSupportTestMPServices - Waiting for AP's to complete execution
2022-08-02 17:00:50 - MPSupportTestMPServices - CPU #1 finished execution
2022-08-02 17:00:50 - MPSupportTestMPServices - CPU #2 finished execution
2022-08-02 17:00:50 - MPSupportTestMPServices - CPU #3 finished execution
2022-08-02 17:00:50 - MPSupportTestMPServices - CPU #4 finished execution
2022-08-02 17:00:50 - MPSupportTestMPServices - CPU #5 finished execution
2022-08-02 17:00:50 - MPSupportTestMPServices - CPU #11 finished execution
2022-08-02 17:00:51 - MPSupportTestMPServices - CPU #12 finished execution
2022-08-02 17:00:51 - MPSupportTestMPServices - CPU #13 finished execution
2022-08-02 17:00:51 - MPSupportTestMPServices - CPU #14 finished execution
2022-08-02 17:00:51 - MPSupportTestMPServices - CPU #8 completed but did not signal (event wait time = 1001ms)
2022-08-02 17:00:52 - MPSupportTestMPServices - CPU #9 completed but did not signal (event wait time = 1169ms)
2022-08-02 17:00:52 - MPSupportTestMPServices - CPU #10 completed but did not signal (event wait time = 1335ms)
2022-08-02 17:00:52 - MPSupportTestMPServices - CPU #6 completed but did not signal (event wait time = 1504ms)
2022-08-02 17:00:52 - MPSupportTestMPServices - CPU #7 completed but did not signal (event wait time = 1671ms)
2022-08-02 17:00:52 - MPSupportTestMPServices - CPU #15 completed but did not signal (event wait time = 1166ms)
2022-08-02 17:00:52 - MPSupportTestMPServices - AP dispatch test complete
2022-08-02 17:00:53 - MP test failed. Setting default CPU mode to SINGLE
2022-08-02 17:00:53 - Applying configurations
2022-08-02 17:00:53 - [CONFIG] This platform has 16 logical processors of which 16 are enabled.
2022-08-02 17:00:53 - [CONFIG] MAXCPUS is not valid

it would be nice to show some information in the UI as well (so that users don't wonder what happened with other cores)

[keith]

1) Running memtest from USB drive is _significantly_ slower compared to running it from EFI partition on NVME drive

Looking at the log timestamps, it looks like writing each log entry to the log file on disk is extremely slow. We've seen similar issues with users using older/slower USB 1.0/2.0 flash drives. Try using another known fast USB 3.x flash drive, if you aren't already using one.

2) When running from NVME it doesn't allow Benchmark results, test results, screenshots or MemTest86.log files to be written in EFI partition (while this works when running from USB).

Does this happen when configuring it manually according to https://www.memtest86.com/tech_configuring-grub.html?

3) F12 for screenshots is great (much better than me taking photos of the screen with phone), but it's unfortunate that it works only in some menus (doesn't work when test is running or when there is "press any key to continue" message, I would rather press just Enter or Space and let F12 bound to screenshots at any time).

Thanks for the feedback. We'll look into this.

4) If I start memtest with monitor (LG OLED TV) connected to my nvidia GPU over HDMI and shutdown the TV over night, then in the morning when I turn the TV on, it doesn't receive any signal over HDMI (so cannot see the results from over night test run). If I connect different monitor (LG LCD) over DP, then it still doesn't show anything. But if I switch from HDMI to DP before starting the memtest (before going to bed) then in the morning it works fine and I can see the output.

This is likely an issue with the UEFI BIOS firmware, as MemTest86 has no control over monitor output.

5) It would be great to show if the memory is running in single/dual/quad/octa channel configuration, it's possible to figure out from the shown memory info about populated slots (if you know your motherboard or have the motherboard manual at hand), but clear indication in the results protocol would be useful.

Yes, we agree this information is useful. Practically, this requires implementation specific code for each CPU chipset, which is a non-trivial amount of work. We are happy to prioritize this feature if a person or organization can fund such development.

6) it shows only 16 cores on 32c/64t AMD 3970X with these warnings in the log file:

The maximum number of CPU threads available for memory testing is 16 for the Free version, and 512 for the Pro version.

As for the warnings, it is an indication of a possible bug in the UEFI multiprocessor implementation. Though we agree that it would be better to display this warning to the user in the UI.

[jama]

Thanks for all your kind answers.

We've seen similar issues with users using older/slower USB 1.0/2.0 flash drives. Try using another known fast USB 3.x flash drive, if you aren't already using one.

Yes it was old small 2GB Kingston DataTraveler 2.0 (which I'm keeping just for these small tools when need, oh well time to switch to ssd ).

Does this happen when configuring it manually according to https://www.memtest86.com/tech_configuring-grub.html?

Actually I've discovered, that it was writing the .log file and screenshots, but into different EFI partition, not the one from where grub is installed and from where it started memtest86 (I've more drives and 3 of them have EFI partition (2x Linux, 1x Windows11), the files where written to the first EFI partition found (happens to be the one with Window11 loader) as confirmed in the .log file. This is good enough for me (now when I know where to look for them), but might be confusing for other people with more EFI partitions.

In the end I've added startup entry to EFI directly without grub and it seems to work fine and I can easily switch between multiple versions as well, e.g.:

efibootmgr --create --gpt --disk /dev/sdd --part 1 --label "MemTest86 v9.5-beta1" --loader "\\EFI\\memtest86\\memtest86-bin-9.5-beta1.efi"
efibootmgr --create --gpt --disk /dev/sdd --part 1 --label "MemTest86 v9.4" --loader "\\EFI\\memtest86\\memtest86-bin-9.4.efi"

Yes, we agree this information is useful. Practically, this requires implementation specific code for each CPU chipset, which is a non-trivial amount of work. We are happy to prioritize this feature if a person or organization can fund such development.

Fair enough, didn't realize that it's chipset specific.

The maximum number of CPU threads available for memory testing is 16 for the Free version, and 512 for the Pro version.

I don't mind the limit (my RAM is failing even in single CPU test), but it's missed opportunity for you to show real number of CPU threads available and a hint to buy Pro version to use them all.



Two more things I've noticed when testing saving the output files to EFI partition:

1) The RAM benchmark .ptx files aren't written (the UI shows it as well) when the Benchmark directory is missing (either moved by user or non-existent on found EFI partition), log shows:

2022-08-02 16:40:12 - Test has completed successfully. Saving results to Benchmark\mt86Bench-20220802-164012.ptx
2022-08-02 16:40:13 - Unable to open Benchmark\mt86Bench-20220802-164012.ptx for writing (Not Found)

it would be nice to create the directory if it doesn't exist - If I create it before starting memtest then it saves the benchmarks fine.

2) When it shows "<Save an HTML report to (MemTest86-Reporte-date-time.html)? ((y)yes>)" it's not really clear if e.g. pressing ENTER will save it (as "y" looks like the default option here), but looks like only pressing "y" key works here, easy to miss when many other screens are "any-key-to-continue" (e.g. when you pres F12 here in hope of a screenshot ).

[David (PassMark)]

a hint to buy Pro version to use them all

More threads actually has a negative performance impact at some point. So the feature isn't as great as it might seem. RAM isn't quick enough to feed data to 16+ cores typically.