Asus X99-E WS/USB 3.1 parrallel test hang, strange elapsed times.

[MysteryGuy]

Hi;

( I presume everything but the elapsed time issue is related to the topic at http://www.passmark.com/forum/memtes...election-modes )..

I tried out memtest86 7.2 free on my new system, and noticed the following with my setup:

1) I can't UEFI boot the CD. The white on black text gets as far as the Multiprocessor checking part before it hangs (before the GUI starts). A hard reset is required.

2) I made a USB stick and I can get into the GUI when I UEFI boot it. When I let it run the tests with all processors, it hangs after about 24 seconds in Test #2..Hard reset is required.

It seems to work with Round robin, sequential CPU, or single CPU though.

3) Booting the old version in BIOS mode allows me to specify parallel processing, but says it is using '6' of the 12 (6 + 6 Hyperthread) cores.

4) Sometimes the elapsed test time being displayed makes no sense. Sometimes it's correct (say a few minutes), other times it say it's been running 71 hours+, or even 801 hours+.

This is with an 'Asus X99-E WS/USB 3.1' motherboard running the latest 3402 BIOS. My processor is an E5-1650 V4 Xeon, with 64 GB of ECC 2400 DDR4 Memory.

Am I missing much by not being able to run in parallel CPU mode?

Thanks;

[keith]

Hi;

( I presume everything but the elapsed time issue is related to the topic at http://www.passmark.com/forum/memtes...election-modes )..

I tried out memtest86 7.2 free on my new system, and noticed the following with my setup:

1) I can't UEFI boot the CD. The white on black text gets as far as the Multiprocessor checking part before it hangs (before the GUI starts). A hard reset is required.

2) I made a USB stick and I can get into the GUI when I UEFI boot it. When I let it run the tests with all processors, it hangs after about 24 seconds in Test #2..Hard reset is required.

It seems to work with Round robin, sequential CPU, or single CPU though.

Yes, the UEFI BIOS for the Asus X99 boards in particular have issues running in parallel mode. We have contacted Asus regarding this issue but we have not been given a timeline of a possible fix. As a workaround, we are in the process of implementing a 'blacklist' for known motherboards with multiprocessor issues. For 'blacklisted 'boards, the CPU mode will be set to 'Single' by default.

3) Booting the old version in BIOS mode allows me to specify parallel processing, but says it is using '6' of the 12 (6 + 6 Hyperthread) cores.

The older MemTest86 v4 does not use hyperthreads when running in parallel mode

4) Sometimes the elapsed test time being displayed makes no sense. Sometimes it's correct (say a few minutes), other times it say it's been running 71 hours+, or even 801 hours+.

Are you seeing this in the latest MemTest86 v7.2 or the older MemTest86 v4? If this is for the latest MemTest86 v7.2, can you send a screenshot and/or the MemTest86.log file under EFI\BOOT\ of the USB drive.

Am I missing much by not being able to run in parallel CPU mode?

Running in parallel allows you to complete the tests significantly faster, and uncover any issues with simultaneous access of memory from multiple CPU cores (eg. cache coherency, shared memory bus, etc.).

[MysteryGuy]

HI;

I've attached the zipped log (from the UEFI boot directoy), as well as some photos (I had to chop them down and I left out one with a normal time because of the limitations on what it will let me post).

The log seems to have entries from yesterday and today.

It appears that the elapsed time being right or wrong sometimes changes with what core is being used. I tried round-robin, sequential, and single.

I saw it correct with a single core, wrong with a single core, and sometimes change between plausible and clearly wrong as the core # changed in the rotation.

Thanksl

memtest-log.zip

[keith]

Thanks for the logs/screenshots. We believe the clock issue is due to the timestamp counter (TSC) not being synchronized between the CPU cores.

Can you give the following build a try and let us know if the issue still persists:
http://www.passmark.com/ftp/memtest86-usb-7.3.2.zip

[MysteryGuy]

Hi;

I gave this a try, and these were the results:

1) I noticed that of the 12 listed cores, Now only 6 were selected (I presume matching the 'real' ones).
0,2,4,6,8,A

2) I did not see the elapsed time issue. I tried about a half dozen times for about 2 minutes each. So,
it appears this problem has been fixed.

3) Trying to run all cores still hangs, but I guess that was expected. I was able to try single core, round robin and sequential
for a short while. They all seemed to work, although I did have one hang in sequential in test 2 on core #6 after about 44 seconds.

But a subsequent try worked for a few minutes. Maybe I'll try a longer test when I have more time.

As a side question, I noticed that when I boot the USB stick, I'm given a choice of two different UEFI partitions to boot. Is there a difference between them?

Thanks;

[David (PassMark)]

There is no difference between the boot options.

After some testing we decided that there was no advantage to using the hyperthreading cores. So future versions will only use real (full) CPU cores. In fact there was a slight speed disadvantage when all cores where trying to access RAM at the same time.