1866 + z68 + xmp

[Mir96TA]

I bought Corsair Vengeance LP
I try to load XMP profile 1866 system will not boot. It will fall back to safe Bios boot and then I have to clock down the memory.
System will boot to windows at 1800 Mhz but system will crash. Matter of fact it has trashed my OS. Luckily I was using SSD cache, so I kill the cache and boot from HDD and I got my butt saved.
My Question is XMP profile is not full a proof (Trail and Error ) or this memory is defective.
Computer specs
CPU i5 3570 K (No O.C. )
Mem old : Samsung Wonder memory @ 900 Mhz 9-9-927-1T Stock Volt
Mem Current : Corsair Vengeance LP
MB: AsRock Z68 Extreme3 Gen3
Here are Pics

 

[Deders]

Most likely defective, although there can be slight inconsistencies with boards so nothing is entirely foolproof.

 

[BonzaiDuck]

I can't speak with first-hand authority on the AsRock Z68 boards, only to say that a representative group of veterans here thought well of them when we were overclocking SB-K processors in late 2011. And now that I think of it, the Z68 Gen3 boards were "Ivy-Bridge-ready" like some ASUS boards. You would know more about your BIOS and its versions, so . . .none of that should be a problem.

You shouldn't need to use the "XMP" profile for the RAM in order to set it to the spec 1866 speed and spec timings. Then, you would want to run HCI Memtest from the bootable CD. On new memory I might recommend the 1000% coverage test, but hey -- 8GB of RAM might take as long as two days. 16GB took my system more than four days.

But for purposes of getting an RMA ticket, you'll actually be lucky if the RAM fails before 100%, and you can get the kit off in the mail soon for a replacement. Just make sure you test the replacement kit at spec settings and give it a reasonably thorough testing -- at least 500%, I'd say . . .

 

[Charlie98]

I didn't think Z68 supported XMP over 1600MHz? FWIW, I had my Samsung RAM OC'd using the basic settings in the BIOS (Gigabyte board, not ASRock) and it did well without using XMP.

I would agree with B-Duck... test it at stock clocks first before RMA'ing.

 

[BonzaiDuck]

I didn't think Z68 supported XMP over 1600MHz? FWIW, I had my Samsung RAM OC'd using the basic settings in the BIOS (Gigabyte board, not ASRock) and it did well without using XMP.

I would agree with B-Duck... test it at stock clocks first before RMA'ing.

Strange. I've got about five over-clock profiles and a "stock" profile, but the "stock" settings are not all "automatic" or default. I believe I had my XMP settings right: that is, for DDR3-1866 modules @ 9,9,9,24,cmd=2, the proper value of 933Mhz would show up in my monitoring programs for the RAM [DDr. . . 2x933 . . etc. etc.]

I am speculating that it is possible to get these settings "wrong" when choosing XMP -- such that the RAM is left running FASTER than its spec at boot-up. So people can have troubles and be totally mystified or confused.

Basically, it seems to me that you're fine not to select "XMP" if you manually configure to the specs. In my G.SKILL GZH DDR3-1866 kit's case, select "1866," set voltage to 1.50V and the essential latencies to 9,9,9,24 cmd=2 and leave the rest to "auto."

So -- yeah -- paying attention to your remark against my hot-air -- you shouldn't "need" to choose XMP, if you know what you're doing to manually configure the RAM for the appropriate speed, voltage and timings. . .

 

[Charlie98]

Strange. I've got about five over-clock profiles and a "stock" profile, but the "stock" settings are not all "automatic" or default. I believe I had my XMP settings right: that is, for DDR3-1866 modules @ 9,9,9,24,cmd=2, the proper value of 933Mhz would show up in my monitoring programs for the RAM [DDr. . . 2x933 . . etc. etc.]

I am speculating that it is possible to get these settings "wrong" when choosing XMP -- such that the RAM is left running FASTER than its spec at boot-up. So people can have troubles and be totally mystified or confused.

Basically, it seems to me that you're fine not to select "XMP" if you manually configure to the specs. In my G.SKILL GZH DDR3-1866 kit's case, select "1866," set voltage to 1.50V and the essential latencies to 9,9,9,24 cmd=2 and leave the rest to "auto."

So -- yeah -- paying attention to your remark against my hot-air -- you shouldn't "need" to choose XMP, if you know what you're doing to manually configure the RAM for the appropriate speed, voltage and timings. . .

Well, also, the OP had Samsung RAM in prior... I had to manually set the voltage down to 1.35v when I installed mine. It detected the timings correctly, but not the voltage. It's not that big of a deal when you are already in the BIOS setting the timings, too, but you have to do it. :biggrin:

 

[Mir96TA]

I have return the memory

 

[BonzaiDuck]

Well, also, the OP had Samsung RAM in prior... I had to manually set the voltage down to 1.35v when I installed mine. It detected the timings correctly, but not the voltage. It's not that big of a deal when you are already in the BIOS setting the timings, too, but you have to do it. :biggrin:

It's a useful aspect of enthusiast over-clocking practice: One is wise to read articles on how double-data-rate RAM works; how the timings are linked to duration of certain steps of memory reads and writes in clock-cycles.

I'm not sure what setting those 1.35V Samsung modules to 1.5 would do. One would think it might still be within the factory specs and warranty tolerances.

All my G.SKILLs are spec'd to run at 1.50V with stock speed and timings. I thought I perceived second-hand that someone was able to run them at 1.35 to 1.45V. I just never took the trouble to find out, and it makes things easier to run them at spec. I have another two sets of G.SKILL GBRL 1600's, and you can overclock them to 1866 at ~1.5V and looser timings.

But as I understand it, you were assured the Samsungs would run at 1600 with 1.35V, and the results for running them faster looked pretty promising according to someone's reports of using voltages below 1.5.

 

[Charlie98]

But as I understand it, you were assured the Samsungs would run at 1600 with 1.35V,

That is the stock numbers for the Samsung RAM. As an aside, folks over at OCN and other places were reporting stupid OC numbers with this RAM. I never really went there, I was just happy with low-profile, low voltage RAM and called it a day.

 

[BonzaiDuck]

That is the stock numbers for the Samsung RAM. As an aside, folks over at OCN and other places were reporting stupid OC numbers with this RAM. I never really went there, I was just happy with low-profile, low voltage RAM and called it a day.

Just to footnote: Those of us who became star-struck with the over-clocking bug slowly learned several things, and I see that the newcomers to the practice show the same incomplete understanding. The performance improvements to OC'ing RAM were never all that great, but there was a time when OC'ing the CPU had implications for how you tweaked the RAM. Not so true anymore.

There's no doubt that the Samsung RAMs offered some advantages to include low voltage and an over-clocking range that was attractive. I myself feel lucky with the G.SKILL GBRL 1600 kits that would overclock to 1866 and looser timings at about the same (stock) voltage.

It will be interesting to see the initial offerings for DDR4. Every standard seems to mature over time, and so OC'ing the early RAM kits is a short-term solution to waiting for better. I just don't think I'll spend as much time trying to squeeze better benchies out of my RAM modules. Not in the future, anyway.

 

[Mir96TA]

Memory was defective.
I got Mushkin Stealth 16 Gig
It is 1600 Mhz with 9 9 9 24 1T

 

[BonzaiDuck]

Memory was defective.
I got Mushkin Stealth 16 Gig
It is 1600 Mhz with 9 9 9 24 1T

With every new machine I'd build, I would give the RAM modules a thorough testing. But as memory capacity jumped from 4 to 8 to 16 GB (and some folks seem to want even more) -- the time it takes for thorough testing doubled, and then doubled again. A 1,000%-coverage test with HCI Memtest on the 2x8GB modules took me more than four days.

So it becomes tempting, if one had been consistently impressed with RAM of a particular manufacture, to skip testing. And that's what I did with a 2x2GB set I put in an extra machine used for business. It doubled the kit already installed, which had been thoroughly tested. I don't know how I could have been so stupid; the errors revealed eventually would have occurred early, or the RAM was defective when I received it.

Luckily, I'd been watching the event-logs on that machine, and diagnosed the problem early. That didn't prevent me from reinstalling the OS and programs, though. Fact is, that particular system didn't really NEED more than 4GB or a single 2x2GB RAM kit.

Your Mushkins have good specs. Still, I would test them before proceeding to add more software or use the system for serious purposes.

 

[Mir96TA]

Well for testing........
System is 100% stable.
I am running few VM.
These are Redhat Machines for Cisco Voice
Cisco Call manager: Publisher, subscriber, Unitty,
While these are running I have using for Web browsing (Aka watch porn)
playing Diablo.
I am pegging up 12 Gig of Ram usage on my VM.
I am really missing 2 extra core I had with Phenom II.
For me that is pretty good test.
I think pretty soon I may to jump to 32 Gig.
Those silly VM really eats up the RAM.