[coreboot] KGPE-D16 and Samsung DDR3L memory sticks

[Daniel Kulesz]

Hi folks,

here's a short update regarding the beforementioned DDR3L Samsung memory sticks on the KGPE-D16 running coreboot.

Thanks to help and hints from Timothy, I experimented with different cpu/memory configurations and have done some stress and memory testing. So far, I have a few interesting findings to report for a 1-CPU-package configuration (using Opteron 6276):

(1) When running coreboot master with 4 DIMMs (in orange slots), the DIMMs are reported in dmidecode to be running at 800MHz and no ECC errors show up in dmesg during memtester runs. Same goes for the vendor bios.
(2) When running coreboot master with 8 DIMMs, the DIMMs are reported in demidecode to be running at 667 MHz and MC4 errors show up in dmesg during memtester runs.
(3) When running the same as in (2) but with the *vendor BIOS*, dmidecode reports 800MHz and memtester runs without MC4 errors.
(4) When running the same as in (2) but with *Coreboot 4.3*, dmidecode reports 667MHz but memtester runs *without* MC4 errors as well.

It is still on my list to validate (using benchmarks) whether dmidecode just reports non-sense or if the DIMMs really run at only 667MHz. However, the issue with the MC4 errors is much more severe and likely to be a regression between Coreboot 4.3 and the current master version. It will require further bisecting to identify the cause.

@Other KGPE-D16 users: Does dmidecode in coreboot report your memory sticks to be running at 800MHz (provided you have PC12800 ones)? And are seeing MC4 failures when running memtester? (For me, it happens at stage 2 with the "random testing" - usually after less than 7 minutes after starting the test and even if I just use 10G of memory for testing).

I also tested using two cpu packages and other memory populations, but as Timothy has obligations regarding the proper dimension of my PSU, it makes more sense for me to test in 1-CPU-configuration first to find the cause of these MC4 errors in coreboot before proceeding further.

Cheers, Daniel



Btw.: I had some "false negative" kernel oopses when running stress-ng with all tests enabled on both vendor bios and coreboot. Seems to be a known software issue (Ubuntu bug 1654073).