Partial checking of RAM a good idea?

[Syntax2001]

Hi,

I have 128 GB of RAM and I want to check it with MemTest86.
As I need this machine the whole day, I can check the RAM only over night and so I have nightly time windows of about 10 hours.
I want to check all the RAM with 4 passes but to check the whole 128 GB in one test run would take approximately 24 hours.
So, I had the idea to check the RAM partially in 32 GB chunks (with overlaps) over four nights by using the Address range features.
I.e. the first night I check 0 - 34 GB; second: 32 GB - 66 GB, third: 64 - 98 GB, fourth: 96 - 128 GB.
Each of these runs (with 4 passes) should take about 8 hours, which is acceptable for me.

Is this procedure sound? Are the any noteworthy disadvantages compared to the long big single test run?

By the way, I would like to suggest an improvement:
Currently only the total elapsed time of the complete test run is displayed.
I would find it handy if we could see additionally:

1. The elapsed time of the current pass.
2. The elapsed time of the current test.
3. The ETAs for the current test run, pass and test.

I know that the calculation of the ETAs can be quite tricky, but rough estimations would be nice.

[David (PassMark)]

That should work.
Are you running multi-threaded testing? As that can make testing much faster.

Time estimates aren't easy as there are various user settings and hardware capabilities that can change the test time.

[Syntax2001]

Thanks for your answer!
And yes, I'm running multi-threaded testing because MemTest86 told me to do that (to produce more load on the RAM) and I also observed that multi-threaded testing is running must faster.
But even with multi-threaded testing all 128GB at once would take a loooooong time.
So, I'm going to use the partial checking approach over four nights.

Regarding estimates:
Yes, I perfectly understand that it is next to impossible to make any estimations before the run.
Perhaps the following approach may be feasible:
Let's assume for simplicity that a pass only consists of 4 tests (not 11 or 13).
Now, if we can tell from experience/guesstimates that the relative durations of the tests is (as they are all check the same amount of RAM):

• Test 1: 100
• Test 2: 150
• Test 3: 500
• Test 4: 250

So, in total the pass will take up to 100+150+500+250=1000 relative "time units".
Now, at the very beginning the ETA for the whole pass is unknown.
But when let's say 5% of Test 1 is done (i.e. Test 1 tested 5% of the Address Range) in, for example, 4 seconds, we could calculate the ETA for Test 1 as 4 s / 0.05 - 4 s = 76 s.
To calculate the ETA for the pass:
5% of Test 1 are according to the relative test durations from above equal to 0.05 * 100 / 1000 = 0.5 % of the pass. ETA for the pass is then 4 s / 0.005 - 4 s = 796 s.
Further, if it's possible to estimate that the relative duration of a fast first pass is about let's say half the duration of a full second/third etc. pass (and all the full pass take roughly the same time), an ETA for the whole test run could be calculated by applying the same type of logic.
Of course, this simple linear logic may have to be extend -- for example for the Bit Fade test were some constant time waits are there.
(I've implemented this logic in many projects to calculate ETAs for long running (sub)processes (Data Warehousing, ETL), and in most of the cases it yielded not perfect but still useful results.)

Another topic: The installation of MemTest86 on a USB stick.
I found that this can be done very easy, without formatting the stick.
I just took memtest86-usb.img and inside I found the EFI folder (that contains the BOOT folder with the EFI executables). Then I copied this complete EFI folder to the root of a FAT32 formatted USB stick (which contained a lot of other stuff). Then I rebooted and in the Boot Menu I could select the USB stick and it booted into Memtest86!
I tried this with a current Asus motherboard and an older (2015) Gigabyte motherboard and it work nicely on both of them.