I have recently run into the same problem as you. I picked up an Asus Maximus Extreme Z87 motherboard (got a killer deal), and I also bought a Trident X 2400 kit 4x8 GB.

I have always run Mem Test86 on my new ram sticks, and never had any errors. I mostly use G.Skill too.

When I run the Mem Test86 I get one error on each pass, on the Hammer test (test #13 if I recall correctly).

I also searched around and did not find any definitive information either. The best I could surmise is that it will be ok for every day use. However, I have not put these sticks into actual use yet.

This hammer test was added on the last release of Mem Test86. I'm certainly no memory expert, so I hope someone can enlighten us here.

 

I ran HCI Memtest (windows based, 4x copies to utilize all 16GB) overnight and woke up to a rebooted computer this morning. Blue Screen code 124 - hardware.

Now, the build as a whole is over clocked (see thread http://www.overclock.net/t/1552447/i5-4690k-oc-results-first-post-rog-realbench-virus for full details).

So I think I need to drop the CPU and Cache back to stock clocks (will likely leave voltages fixed) and try again over night. Need to work out if the issue is with the ram sticks, or the IMC. Anyone got any advice or a plan of attack.

System agent has already been bumped, as have IO-D and A ... probably higher than they need to be.

Add to the mix, I have occasionally had Exception Breakpoint Errors thrown from IE 11 - but only when visiting this site. There is definitely something happening on the memory front which I'd like to find an answer to.

 

Quote:

Originally Posted by Jakescode

I ran HCI Memtest (windows based, 4x copies to utilize all 16GB) overnight and woke up to a rebooted computer this morning. Blue Screen code 124 - hardware.

Now, the build as a whole is over clocked (see thread http://www.overclock.net/t/1552447/i5-4690k-oc-results-first-post-rog-realbench-virus for full details).

So I think I need to drop the CPU and Cache back to stock clocks (will likely leave voltages fixed) and try again over night. Need to work out if the issue is with the ram sticks, or the IMC. Anyone got any advice or a plan of attack.

System agent has already been bumped, as have IO-D and A ... probably higher than they need to be.

Add to the mix, I have occasionally had Exception Breakpoint Errors thrown from IE 11 - but only when visiting this site. There is definitely something happening on the memory front which I'd like to find an answer to.

Yes for sure put the CPU and Cache back to stock. That will eliminate the CPU overclock as a problem.

The Blue screen 124 code usually needs more v-core to fix.

 

I ran into it trying to overclock my 4790k. Not overclocked Kingston fury 1866 16GB i get around 20 faults on one test loop. On my 2133 on same timings overclock I got 54 but then I gradually upped the power from 1.5v to 1.63 - 1,575 and 1.58v was the sweet point with only 12 faults. after that it rose to around 20. Reducing Command rate from 2 to 1 made a big difference. Not surprising that voltage needed to be higher as 2133 timing is 10-11-10-35 cr 1 that is about the same as 1866 stock! The funny thing is my overclock for 4.6 now a behaving much better: test on OCCT 2 hour stable on 1.60, 7.5 hour on 1.62 and now testing 1.64

. So apparently it have helped to pinpoint max memory stability witch help CPU stability too?

Anyway don't think it is too important to me as it is only a gaming computer and it is a very extreme test.

But i have been thinking there will come better ram in the coming years for DDR4's higher bandwidth, maybe using more layers for bigger cell's like Samsung 850 SSD's 32 layers. But they first need to sell old tech first

EDIT: Before i could't not get past 40 min on 1.275 v

 

RMAing often won't help because the bulk of DDR3 seems to be affected and you might need miracle ICs to be immune at common speed/timing bins.

Very few real world apps will produce loads that can hammer rows, but memory vulnerable to row hammer is a security risk.

If you are concerned about row hammer on DDR3, you want to increase the DRAM refresh interval. Reducing the tREF cycle number to something between 1/2 and 1/4 normal should dramatically reduce row hammer vulnerability with only a small hit to performance.

DDR4 seems much more resistant to row hammer, at least on Haswell-E (no idea if it's DDR4 itself or the new IMC that is responsible for the bulk of this, however). I can pass the row hammer test with half stock refresh interval on OCed memory.

 

Thanks everyone. It appears most people are just living with the ram the way it is then. I will look at dropping the tREF to reduce/remove the vulnerability. I bumped the CPU Input Voltage up 2 notches from 1.79 to 1.808 (which while using high LLC fluctuates between 1.808 - 1.824 when clocks change) and passed 11.5hrs of 100% CPU 100% Memory using HCI Memtest over night, so it appears that was the issue. Must dipped a little too much at times.

 

I tried 4 of my Samsung 4GB green modules from mid-2012, and they all passed the hammer test of MemTest86 ver. 6, but my 8GB Kingston HyperX Fury 1866 MHz made this year showed over 50 failures. I don't know what chips are in the Kingston, but late last year these modules were built from 1.35 volt Hynix H5TC4G43AFR. It didn't matter whether I set the voltage to 1.35V or 1.50V.

At Reddit, a DRAM engineer who goes by the user name memthrowaway215 said this:

"Row hammer is the easiest way to force channel hot carrier stress on dram. This will slow access to that row primarily, but will have lesser effects to rows that share the same circuitry. If your dram is on the higher voltage side and cold (cold = 0c and hot = 85c) it will accelerate the stress. Do note this is really only an issue if you are running a row hammer overnight. This will not make the row hammer worse."

 

Some of my Samsung Green also passed the row hammer test, but my last DDR3 board that was new enough to support UEFI is dead, so I can't even test the rest.

 

These sticks use the Samsung modules - don't overclock well at all and offer very limited ability to tighten timings. Seems same ram for 2133/2400 - just different SPD for XMP; increased clocks and voltage.

I haven't had a chance to hammer them since bumping the cpu input voltage a touch. I will give it a crack tonight perhaps.

 

Quote:

Originally Posted by Jakescode

These sticks use the Samsung modules - don't overclock well at all and offer very limited ability to tighten timings.

Samsung actually makes the whole modules??? I haven't seen that since my 2 x 1GB DDR2 Crucial with "SAMSUNG" etched into the copper. OTOH if Samsung makes just the chips on the modules, I don't see how they can't be overclocked a lot, especially because Samsung Green 1600 MHz modules are famous for their overclockability, thanks to being made from real 1600 MHz chips and needing just 1.35V to meet specs, meaning loads of overvolting can be done.

 

Quote:

Originally Posted by larymoencurly

Samsung actually makes the whole modules???

Samsung makes and sells modules. One of the few DRAM manufacturers that does under their own brand.

The Samsung Green "wonder RAM" you mention; that was just Samsung.

Regardless, Samsung makes a lot of different ICs and modules that use them, not all of them OC well.

 

Quote:

Originally Posted by Blameless

Samsung makes and sells modules. One of the few DRAM manufacturers that does under their own brand.

The Samsung Green "wonder RAM" you mention; that was just Samsung.

Regardless, Samsung makes a lot of different ICs and modules that use them, not all of them OC well.

I'm surprised G.Skill would splurge for quality because this is a company that's used 1333 MHz chips in their 2000+ MHz modules, and some of their 2666 MHz stuff is made from unmarked junk chips:



Does G.Skill change the SPDs of the Samsung modules by adding XMP profiles? Because I don't think Samsung has ever put such profiles in their modules.

OTOH where can we still buy Samsung branded modules? I'd like to find some 8GB.

 

Quote:

Originally Posted by larymoencurly

I'm surprised G.Skill would splurge for quality because this is a company that's used 1333 MHz chips in their 2000+ MHz modules, and some of their 2666 MHz stuff is made from unmarked junk chips

Nothing unusual about this. ICs are highly variable and often minimally binned/tested, so most DIMM manufactures selling performance memory do it themselves.

Those chips weren't originally unmarked. G.Skill probably sanded them down and stenciled their own logo all over them to conceal their original type.
Quote:

Originally Posted by larymoencurly

Does G.Skill change the SPDs of the Samsung modules by adding XMP profiles?

G.Skill doesn't buy whole modules, just ICs. The ICs have no storage for any profiles. SPD and XMP are stored in a ROM chip on the PCB.

G.Skill programs their own SPD and XMP profiles on to whatever modules they build.
Quote:

Originally Posted by larymoencurly

Because I don't think Samsung has ever put such profiles in their modules

Samsung doesn't make "enthusiast" memory, so their modules generally don't contain XMP profiles.
Quote:

Originally Posted by larymoencurly

OTOH where can we still buy Samsung branded modules? I'd like to find some 8GB.

Samsung themselves sell some directly, I believe. Many etailers, including Newegg, Superbiiz, Amazon, and other also sell Samsung Original.

Few Samsung 8GiB DDR3 DIMMS OC especially well. They do not use the ICs that made the MV-3V4G3D/US famous.

 

Thanks for clearing up all the terminology.

Appears the particular batch consists of single sided Samsung chips.

Here is a thread, matches my experience with this Trident X to a tee! http://www.overclock.net/t/1385526/gskill-trident-x-2400mhz-wont-overclock

And a review where they experience the same as well. http://www.thinkcomputers.org/g-skill-tridentx-ddr3-2400-8gb-dual-channel-memory-kit-review/

 

Quote:

Originally Posted by Jakescode

Appears the particular batch consists of single sided Samsung chips.

What are single-sided chips? Non BGA? But every DIMM I've seen probably since DDR2 has used BGA packaged chips. Could you point out at least one single-sided BGA packaged chip from the the Samsung catalog? LINK

 

Quote:

Originally Posted by larymoencurly

What are single-sided chips? Non BGA? But every DIMM I've seen probably since DDR2 has used BGA packaged chips. Could you point out at least one single-sided BGA packaged chip from the the Samsung catalog? LINK

He must mean single sided DIMM.

 

Quote:

Originally Posted by Blameless

He must mean single sided DIMM.

I must indeed!

I'm no memory guru.. I don't know the terminology or componentry of a stick of ram as well as I could (mostly due to not having that much interest). I'm pulling parts of information reported by others from memory to a large extent.

Not all Trident X is created equally.

 

From the previous thread I linked to

"Which exact GSkill model do you have? By that I mean what are the 5th - 8th digits of the serial numbers of your specific kits. These particular kits were made with several different ICs, some way better than others. The double-sided Hynix based kits were good, and there were also some double-sided Samsung HCH9 based kits that also performed well.

Unfortunately, there are also some single-sided 2x4GB DDR3-2400 C10 kits with "1500" in the 5th thru 8th spot of the Serial Number which used Samsung chips that simply do not clock beyond 2500-2500, not matter lose the timings, or the amount of voltage (see first linked review). There were also some with "0340" (second linked review) which are also Samsung. Near the end of Page 3 in the second review they specifically mention the modules being single-sided.

If yours are these single-sided ones then you've already gotten all your going to get out of them."

Mine are 2 x 8GB 2133 CL9 with the 1500 in the serial. My overclocking results were the same. They are easily stable running 2400 CL10 timings, but very little room to push even if relaxing the timings further. I have not removed the heat spreader, and its entirely possible that due to being 8GB dimms that they may be double sided, but I still believe they are using the Samsung chips mentioned above. Further to not performing well in terms of OC potential (doesn't worry me personally), they do fail the hammer test which is a little disappointing.

 

Double post