Continuation Rant: G.SKILL DDR2-1000 2x2GB

[BonzaiDuck]

This footnotes my thread "What's the latest skinny on DDR2 . . . .. "

Per urgings of fellow members here, I went forward to RMA my set of 2x2GB G.SKILL DDR2-1000 modules. The customer support is very good, and I got my RMA number very quickly.

Newegg shows strong demand for these modules with 420 cumulative customer review posts. But I wish I had followed my own thinking about quality control and the sorts of problems reported by users when I bought the kit.

While the product rating is an average of 5 over those 420, one only needs to read through 3 or 4 pages of reviews to see that the failure rate is much too high. It is an "infant mortality" problem. And of course, with customer reviews, you cannot reliably tell whether the extremely positive review posts come from people who rigorously stress-tested those modules.

I told their RMA department that I'd give them good reviews here and at NewEgg if the RMA warranty replacements proved sound. The company needs to work on its quality control for these modules. And you wonder what they'd need to do to effect such changes. It would certainly push the cost of the modules higher in a market where supply and demand with DDR3 slowly gaining in usage and acceptance is pushing price lower.

 

[Dorkenstein]

Hmm, I was just about to order this set myself.

 

[BonzaiDuck]

Take your chances. The customer reviews at NewEgg show a lot of infant-mortality and RMAs, but the number of 5-star ratings seems to statistically swamp the less-than-stellar reviews.

As I said -- I promised the G.SKILL folks some good press if they would only send me a pair that runs DDR2-1000 at their own spec, with no MEMTEST86+, PRIME95-Blend, and BSOD errors.

I bought a set of Corsair DDR2-800 DHX's in the meantime. The Corsairs may still run at the clock speed I'm shooting for, but more likely at tighter latencies. Even so, the G.SKILLs show some customer reviews touting 4-4-4-12 latency settings at 800 Mhz.

the problem there is similar to the periodic"Post your results for the _____ CPU" threads. You can never be sure how well the "customer-reviewer" stress-tested the memory at the reported settings, just like you can't be sure if someone said they got their E8400 processor to run at 4.05 Ghz and 1.34V after running PRIME95 "small-FFTs" for only an hour.

Remember last year's fiasco over "dangerous Chinese toys" with toxic paint and other factors? I had also discovered that Sunbeam-Tech was selling an automated, USB/thermally-controlled fan-controller that didn't control!! I bought two of those units, and the little blue-and-white rheostat devices on the PCB were missing on both!!

You'd think that Taiwanese companies would've followed Japanese quality-control principles -- maybe not. But I'm convinced that it has not entered mainland Chinese industrial culture -- and Dr. Deming died back in the early 90s in Silver Spring, MD. [I worked with the guy for 2 years.]

If they had a rigorous testing process, and process-improvement-focused quality-control, there wouldn't be so many "DOA" and "had to RMA" reports. But also, if they had such a program, the cost of the RAMs would be noticeably higher, too.

 

[LOUISSSSS]

newegg reviews MUST be taken with a large grain of salt. after 6 x 1gb of crucial ballistix go bad on me back to back to back. RMA after RMA. i went ahead and bought these sticks and both tested 100% stable out of the box

 

[BonzaiDuck]

Originally posted by: LOUISSSSS
newegg reviews MUST be taken with a large grain of salt. after 6 x 1gb of crucial ballistix go bad on me back to back to back. RMA after RMA. i went ahead and bought these sticks and both tested 100% stable out of the box

Oh . . . . I agree to that, LOUISSSSS!! But "taken with a grain of salt" when users say they can run the hardware beyond spec, error free.

Yet, when it comes to failure, whether or not the user stress-tests adequately, that's got to be a sure thing. Why would anyone lie about MEMTEST86 errors, BSODs, or "DOA and had to RMA?"

Continuing on my attempt to get this sucker up and running with 4GB, I must say that I have always been over-cautious -- 25 years of excess caution with this technology. After my "bad BIOS flash" disaster last summer -- first time in half-a-hundred flashes since flashing BIOS's became possible -- I've made an effort to CLR CMOS or reset all BIOS parameters whenever I replace a piece of hardware.

But I forgot this time when replacing my temporary 2x1GB Crucial Tracers with a set of 2x2GB Corsair DHX. I thought about it, said to myself "Eh, what the heck!" and "No Problemo."

Nice to know that, if you do it with fixed latency settings that match the spec of the new modules.

Incidentally -- has anyone had experience with these TWIN2X4096-6400C4DHX 2x2GB kit ?

"With a grain of salt," one or more Newegg reviewer speaks of running them up to DDR=900 and even beyond at the voltage spec, possibly with looser timings. Again -- you don't know how well they tested. I thought someone had said in those reviews that they'd stress-tested for a day or more.

I was able to get my DDR2-800 Tracers to run rock-solid at 875, 2.125V, 4-4-4-10, 2T when the latency spec at 800 is 4-4-4-12.

The problem with the Tracers and Ballistix -- maybe not the way it happened to you, but as described by others and by my own experience -- you could push the voltage to between 2.175 and -- say -- 2.225V, and they would "burn out." In my case, though, I was running them at 2.175V with a command-rate of 1T. Crucial tech-support went "ballistic" when I told them that AFTER I'd received my RMA replacements.

 

[LOUISSSSS]

Yet, when it comes to failure, whether or not the user stress-tests adequately, that's got to be a sure thing. Why would anyone lie about MEMTEST86 errors, BSODs, or "DOA and had to RMA?"

because many newegg users are morons and don't know how to properly set vram and proper dividers and timings.

and a few duds out of 1000+ buyers (keep in mind most people that are happy with their purchase don't post reviews and only people that get duds do).

i wouldn't hesitate getting another one of the Gskill kits again.
Ballistix on the other hand.. lol

 

[BonzaiDuck]

I see your point, and I recognize its validity.

But if a mobo at stock settings will "auto" configure RAM to SPD or EPP timings [looser than necessary], and run the memory in DDR2-800 "Native" mode, you'd think there would be something to the "DOA" observations.

Between my scenario and yours, however, I'm even looking at the user-reviews on these Corsairs I'm testing at the moment, and they have their . . . . DOA/RMA observations as well.

But user-reviews, taken with a grain of salt, offer some insights to use in qualified judgment. Nothing takes the place, however, of benchtest reviews like those made by Anandtech and other sites.

Example -- you'll like this -- I've got these two Striker Extreme motherboards. From "the git-go," customer-reviews were split, because, as you say, there was the "moron and newbie component." The reviewers either raved about it, or claimed that it was an expensive rip-off piece-a-s***. Meanwhile, I was watching forum posts in Australia, here and elsewhere, where people were rigorously putting the board through its paces, and the "expert" hardware reviews were very positive. I watched the forum posts by people who obviously knew what they were doing, and tracked the improvements with BIOS releases.

That's when I finally decided to ignore the moron-newbie group and shell out my hard-earned cash.

And . . . . I'm still havin' fun . . . . . despite the fact that even some here will deride me for stickin' with Striker. The 680i chipset is on the sunset-side of its life-cycle, and I'm not sure what I plan to do next. I'll bide my time . . . . patiently . . . .

 

[BonzaiDuck]

UPDATE::

Well, now I've gone and done it! I RMA'd G.SKILL a perfectly good set of modules, thinking they failed!!

It is confirmed now: This system with VISTA-64 seems to BSOD with PRIME95 Blend Test with ANY 2x2GB 4GB memory kit. I threw in the set of Corsair DHX's, and got the same result.

Does anyone know of some bug in the latest PRIME95 version? If a 2GB kit would run perpetually under "Blend" without fail -- what would cause a 2x2GB kit with PRIME95 to BSOD?

My apologies to all G.SKILL users and G.SKILL itself. Looks like I'll be paying postage upon the return.

As to the cause of my problem, this is a burning question. I think I need to find another stress-test program and go from there.

 

[nevbie]

Good to hear that my memory is good.

How about using HCI memtest for RAM, prime95 small FFTs for CPU?

 

[n7]

Originally posted by: BonzaiDuck
UPDATE::

Well, now I've gone and done it! I RMA'd G.SKILL a perfectly good set of modules, thinking they failed!!

It is confirmed now: This system with VISTA-64 seems to BSOD with PRIME95 Blend Test with ANY 2x2GB 4GB memory kit. I threw in the set of Corsair DHX's, and got the same result.

Does anyone know of some bug in the latest PRIME95 version? If a 2GB kit would run perpetually under "Blend" without fail -- what would cause a 2x2GB kit with PRIME95 to BSOD?

My apologies to all G.SKILL users and G.SKILL itself. Looks like I'll be paying postage upon the return.

As to the cause of my problem, this is a burning question. I think I need to find another stress-test program and go from there.

No & no.

What version of P95 are you running?

I have run the old one, 24.14 64-bit, & the current one, 25.6 64-bit without issues with 2x2 GB & 4x2 GB.

Only time it BSODs is if something is unstable.

And in your case, that something is likely to do with your board/chipset or timings not set right for 2 GB DIMMs.


HCI Memtest is indeed good though.
Best would be IntelBurnTest: http://www.xtremesystems.org/f...howthread.php?t=197835

 

[nevbie]

Originally posted by: n7
Best would be IntelBurnTest: http://www.xtremesystems.org/f...howthread.php?t=197835

Hey thanks for mentioning such an interesting tool. Is it restricted for CPU&memory stress testing only? Or would lots of runs with tiny amount of memory mimic a CPU only stress test? Too bad 1mb doesn't fit my conroe-L cache, haha.

EDIT: Well, I did some testing and answered my own question. Based on temperatures, small memory usage tests do not stress the CPU much, even if I put a lot of them. So this program is strictly CPU&memory.

 

[LOUISSSSS]

you're probably BSODing in your OS with any ram because your OS is already corrupt. put in some TESTED (memtest86) ram and reinstall your OS.

i've been strictly using memtest86+ (for bad memory outside of the os) and OCCT to test ram and cpu within the os. and a single 24 hour run of p95 for long term (24 hours @ ___ voltage) stability when i've finished with everything else

 

[Tweakin]

I'm going to step out on a limb and say that maybe you should verify your "tweaking" and "testing" abilities before you put blame on any software that is currently being used by a good portion of the enthusiast community and consider it to be a stable, tried-n-true benchmarking utility.

I noticed that you have a "couple" of threads regarding this situation you are currently in. Maybe you are just lonely and need a place to rant and vent. I have no problem with that...but how about just one thread then.

 

[BonzaiDuck]

REPLY FOR n7, LOUISSSSS, and Tweakin :

Tweakin is both polite and correct here. The frustration was "gettin' to me" and with other stuff on my plate, my intuition should've led me to calmly resolve this [which I did . . . . Check the "Software->Operating Systems" forum.]

I think we all need to be more careful how we run these stress-testing programs under controlled conditions.

In this case, it was an OS feature conflicting with the use of RAM (exclusive to 4GB kits) while running Blend-Test. I'm still getting familiar with VISTA-64, but the trail-of-tears-and-sweat in installation from the pre-SP1 version and upgrade to that service pack is well documented.

For some reason -- it could still be a set of reasons which includes SLI implementation -- the two, perfect 4GB kit installations at stock settings caused the BSOD crash after exactly 20 minutes and with the 20-minute screensaver default activation. I installed the latest video driver and turned off the screen-saver. It's currently "Blend-Testing" rock-solid now for almost an hour-and-a-half and counting.

But this doesn't happen with a 2GB kit -- only the 4GB kit. There is something about VISTA-64 memory addressing that dovetails with the issue about Win XP's inability to "show" a full 4GB anyway -- that is, I think from the XP viewpoint, it "uses" the memory but it doesn't make it available for programs or show it in system info, and VISTA-64 has a different addressing scheme. Pardon my lack of confidence and specificity here with these latter observations.

But . . . . "It's the screensaver!!" Ordinarily, I turn off AV programs and other "stuff" -- and I disable the screensaver when running PRIME95. But they reorganized the VISTA OS and you have to "drill down" through some menus to find the dialog for disabling the screensaver to "None." Also, I think there are two screensaver features: an ACPI feature in Power Options that turns off the screensaver [and I've seen it as a BIOS [enable/disable] feature as well] -- and the "display properties" feature which I have now set to "None."

THAT's WHAT IT WAS, FOLKS. JUST A LITTLE MORE TESTING, and my tentative conclusions will now become "conclusive."

I'll take refuge in my "over-60" cerebral deterioration for now, but I had always had a good intuition for sorting these things out.

* * * RETRACTION OF MY FLAME-COMMENTS ABOUT G.SKILL * * *

G.SKILL RMA-dept is sending me back the DDR2-1000 kit, per an URGENT e-mail I sent them yesterday. While their rules about "who pays for postage" states that RAM found to be error-free will require the customer to pay for return postage, they assured me that they'll pay for it. [Two thumbs up.]

I'll want to test these DDR2-1000's and make a decision which of the kits (G.SKILL DDR2-1000 or Corsair DDR2-800) to keep in the system.

But -- at least for now, they both get a thumbs up, and G.SKILL gets it for "Price" as well as Quality.

[I'm gettin' freakin' old . . . ]