FEATURED Samsung 2x4GB DDR3-1600 C11 ECC - M391B5273DH0-YK0

[Woomack]

Today something slightly different so ECC version of pretty popular ( at least couple of monts ago ) Samsung based on HYK0 IC.

You can find this memory in small, single processor servers from DELL, IBM and other manufacturers. Lately all were switched to Hynix but I still got two 4GB sticks.

Since ECC ( unbuffered ) memory is working also on Z87 boards then I wanted to check how it's overclocking

Memory comes as standard size DDR3 without heatsinks.







As it's server memory, there is no XMP profile and we can only see SPD settings of 1600 11-11-11-28 1.35V - pretty standard for HYK0 chips.
Memory product number is M391B5273DH0-YK0

All tests were performed on i5 4670K and ASRock Z87M OC Formula. Timings were not optimally tweaked so I bet that results can be better.

Below we can see SPD profile and results using these settings.

 

[Woomack]

Overclocking results

1600 7-8-8-21 1N 1.50V




1866 8-9-9-21 1N 1.50V




2133 9-10-10-21 1N 1.60V




2400 10-11-11-21 1N 1.65V




2600 10-12-12-21 1N 1.70V




I couldn't make it stable @2666 but will check that again other day. 2800 couldn't boot up to 1.85V but of course I will check that later

 

[Johan45]

Looks good Woomack, not bad for something no one wanted. There it is again that 40+ write but the read really dropped off when you hit 2600

 

[Woomack]

Thanks Sub timings are the same at 2400 and 2600 so I'm not sure why performance is dropping at 2600. I have to check it one day. Now I have 2 other kits to test but I will probably back to these Samsungs in couple of days.

 

[Woomack]

Looks good Woomack, not bad for something no one wanted. There it is again that 40+ write but the read really dropped off when you hit 2600

I'm not sure why but at 2600 I have to enable high performance memory option on ASRock Z87M to get as high ( or higher ) memory read bandwidth as 2400 clock. Even if 2400 and 2600 have exactly the same timings set manually then at 2600 read transfer is dropping without this option.

Yesterday I was checking if higher voltage let me run this memory @2800 but bumping vdimm to 1.93V didn't help to keep even basic stability for AIDA64. The only progress is that I made it enter to Windows.
Higher voltage didn't help to set 2600/2666 CL9 too while 10-12-12 1.7V is pretty stable.
Also tightening timings didn't really help to achieve higher performance and HyperPi results were +/- 2 seconds so I assume that 2600 CL10 result in the 2nd post is optimal for this memory.

 

[Woomack]

It took some time but I added couple of photos. Not much to see there

 

[fromthewatt]

thnx for the pics woomack, i noticed these are made in phillipines,
i think there was rumours that the ones made in korea were best for ocing....
i have seen of some of these made in china aswell,...

 

[Woomack]

These sticks are ECC and not many users are actually overclocking ECC modules so it's also hard to gather more results to compare.
I can't complain on overclocking as I saw many high end memory series based on similar chips that couldn't OC above 2400 and here there is pretty nice 2600 CL10.
Anyway I doubt that you can find new sticks like that as all new 1600 C11 that I saw in DELL/IBM/HP were on new IC and mainly it was Hynix.

 

[unclocker]

Who is checking and logging ECC errors on Haswell / X99 platform? Is it done by memory controller inside CPU or memory controller on a DIMM module? How to see logs of these errors? And if it is done in CPU does it mean that overclocking ECC unbuffered memory is perfectly safe and easier? First you'll see some corrected errors grows and only then uncorrectable errors