1

I recently built a workstation with 256GB of memory. Typically when testing a new system I like to run at least three full passes in memtest to make sure the memory is good. However, with 256GB of memory this will take upwards of three days. I'd like to avoid that if possible.

This system supports memory mirroring. I could theoretically cut the memory testing time in half if I enabled it. So I'm wondering:

  1. Is this safe to do?
  2. If the memory is tested this way will it be fully tested? Or do I run the chance of something being missed?

This system does have IPMI, so if an ECC error is hit it will be logged to the IPMI event log.

1 Answer 1

1

From what I can tell the tests wouldn't be valid. From specifications on the 5500 series processors:

The integrated memory controller in the Intel® Xeon® Processor 5500 series processor alternates between both channels for read transactions. Write transactions are issued to both channels under normal circumstances.

Data will be written to two DIMMs, but will only be read from one. This means the full space of all DIMMs won't be fully validated

Of course this will vary depending on the processor model and how it works specifically for that series.

Additionally, I found the following:

Intel Errata BT161
Mirrored Memory Writes May Lead to System Failures
In mirrored memory mode, each channel manages its memory write bandwidth resources. Due to this erratum, if a channel in mirrored memory mode is heavily utilized, it is possible for issued writes to exceed available bandwidth resulting in write failures.

Due to scenarios like this there's a chance that errors or crashing could be encountered because of memory mirroring when there's nothing wrong with the memory itself, resulting in invalid test results.

As for whether or not it's safe, at worst the system may crash or hang, but it shouldn't adversely affect any components.

1
  • 1
    Nice Errata find! Hopefully a BIOS fix will be implemented. On a system like this, with ECC, MemTest serves as the means by which to force read/writes to all RAM. The system's ECC will catch any errors including the uncorrectable ones, and you'll see them in IPMI logs. If I were you I'd disable mirroring and let it run 3-4 days as you indicated.
    – misha256
    Commented Nov 6, 2016 at 20:59

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .