NOT All that Happy with RMA replacement DDR2-1000's

[BonzaiDuck]

Basic configuration: Q6600 @ [various speeds]; Crucial Ballistix DDR2-1000 2GB OR DDR2-800 Crucial Tracers [but not both]; BFG 8800 GTS 640MB, Striker Extreme with v.1301 BIOS.

History: Originally purchased DDR2-1000 2GB 2x1GB kit of Crucial Ballistix. Variously had them at different timings [3,3,3,8 or 3,4,4,9 ] for speeds between 3.0 Ghz and 3.2 Ghz with 1:1 ratio. Was able to run them at 1T command-rate with voltages between 2.15 and 2.2V.

December 24, 2007 -- A module went south, and I had a set of these brand-new Tracers -- anticipating family Xmas visit with people wanting to play games. I popped in the Tracer DDR2-800s. Could run and PRIME them 1:1 @ 3,3,3,8 up to 3.15 Ghz DDR2-700. Was able to make them run 4,4,4,10,1T tRC=14 at 4:5 ratio and VDIMM = 2.125V -- @ variously 835 Mhz, 870 Mhz and 875 Mhz. Last two settings required VDIMM = 2.150V.

January 30, 2008 -- received RMA replacement on DDR2-1000 modules. Popped 'em in; proceeded to recertify, test and revalidate.

RESULT: The RMA replacements won't run at 1T at 835 Mhz, even with voltage twisted up to 2.15V and timings loosened to 4,4,4,12. They seem to run at that speed with 4,4,4,10,2T, but getting them to 870 Mhz requires 4,4,4,12,2T.

And you can't run them at 1T . . . . period. Nada. No cigar. No dice -- no luck.

MEMTEST86+ reports between 26 and 120 errors -- first pass of test #5 -- for any attempt to use 1T CMD rate -- any voltages up to and including VDIMM = 2.150V (monitor in BIOS and Windows shows 2.19V)

What gives here? This is my question to fellow forum members.

EDIT AND UPDATE:

HYPOTHESIS ABOUT CRUCIAL RMA PROCESS:

1) People send back kits of double modules in which many are shown to have one module gone bad.

2) Crucial grabs the good modules after testing, and repackages in "kits."

3) Often, these good modules have been "weakened."

How else would the quality of an initial purchase vary from the an RMA replacement kit? First rule of QC is to refine the manufacturing process so that all products rolling off the assembly-line are of identical quality and tolerances.

 

[BonzaiDuck]

Bump.

What? Nobody else had this experience with RMA replacement? I'll check back again.

 

[nefariouscaine]

I've had hit or miss mr BonzaiDuck - I thought I'd hold off on my 2 cents so this didn't turn into another memory forum hosted by us

I've RMA'd single sticks and kits with Crucial - everything is a mixed result IMO. I had 1 kit of the Anniversary Ballistix that I loved so much (soon to have another replacement on a once thought to be totally out of stock product!!!) and that kit would manage to pull 1200mhz @ 2.25v - where the other kit (had 2 kits at a time...) I had had serious trouble over 1066mhz and capped at 1100mhz regardless of voltage.

1 stick went bad on the lower clocking kit and I just didn't bother testing high mhz with it (the RMA replacement) but was shooting for CAS3 T1 at 800mhz instead. After RMA'ing that kit from giving it way too much juice Crucial told me they didn't have anymore of my Anniversary in stock and I settled for a 667 kit of Ballistix instead that I'm currently pairing up with a 800mhz kit of Ballistix ($50 frys deal FTW) running 4 x 1 gigs ATM. I held on to the 1200mhz kit for a while but eventually they stopped wanting to run as a pair as 1 stick went defunct. The gold at the end of that Rainbow was they now told me they have Anniversary sticks in stock (don't know how that happened...maybe your theory on them reusing sticks is true...) But I just sent that 2nd or 3rd now...RMA in last week.

I'm not holding my breath expecting my new kit to do 1200mhz (well over the stock 667 ) But I always keep in mind that there are variances which I don't know about. not all CPU's clock the same or require the same amount of voltage - nor should I think ram would do the same either.

I can't say what kind of quality control gets done to decide which IC's go into certain sticks of ram but it sounds like we might have more room for debate on how it should be done - there has to be some reason that certain sticks are rated for certain latency values and speeds. You may have gotten "lucky" with the kit you had go bad or you might have gotten unlucky on the new kit.

I'll say for sure, just make sure you got a 60-80mm fan on these sticks at all times

I'm off to go Fatbody hunting again, just for fun and I'm eying that Lanfest kit Karaktu has up for sale cause I'm a junkie

 

[JustaGeek]

You know, I still don't completely understand overclocking the memory.

Let alone 1T on Intel platform.

Are you sure that the benefits outweigh the risks, Bonzai...?

I am running my G.Skill HZ's at stock 800MHz, 2.0V, perfectly stable for about 9 months, with the SANDRA bandwidth of ~6600MB/s.

How much more can you "milk" out of it...?

 

[BonzaiDuck]

Nefarious --

I think they pick the latencies for SPD on a 100%, guarantee-assured ability to run at the specified speeds. Probably, like Intel's chip-making, they make them to run at a higher spec, and then rate them at a lower spec.

So . . . . I am sort of sticking to my suspicion about recycled sticks from RMA's pairs. Especially if they didn't make the modules anymore, which is certainly true with the DDR2-1000s -- the supply on anything but singles seems to have dried up.

JustaGeek -- I wouldn't run them at CMD=1T if I hadn't stressed them under PRIME for a long, long time. Previously, I'd been reticent about this. On-line guides say "I won't tell you what to do about the command rate -- you be the judge." Other sources, like "Adrian Rojak's 'Rojakpot'" BIOS authority note that "data-corruption can occur," but it leads you to conclude that the modules have to be right on the edge. I ran the DDR2-1000's practically all year at 1T, and never had any corruption nor anything go wrong. But in December, as I said above, when I shut down the system, and then tried to start it up "cold-boot" again by flipping on the power, the VGA didn't go from "amber" to "blue" -- nothing posted. And by swapping components, I determined it was the RAM. I'm not sure that running modules at a 1T command rate would inherently damage the modules, and the closest thing I did that could increase probability of failure was to run them at voltages between 2.15 and 2.2V, with the BIOS monitor showing +0.035V above the set value.

But the kicker here is that I've got four or five pairs of Ballistix and Tracer modules, and all of them will do "1T" except this RMA replacement.

And that -- after I'd sent them a nice letter with lavish praise, they e-mailed a request that they be allowed to use the letter in their promotions, to which I answered "sure. Go ahead. Glad to do it."

Still -- they came back with a set that "worked," even if it didn't run at 1T. But the fact that all the new and spare kits I have do not show any problems with a 1T command-rate -- and these don't cut it?

Small "sample size," but think about the odds here. what's your Bayesian estimate that the RMA replacements would work fine at 1T -- IF they were never used before, and IF some four different sets were "problem free?" You see what I mean?

 

[nefariouscaine]

Originally posted by: JustaGeek
You know, I still don't completely understand overclocking the memory.

Let alone 1T on Intel platform.

Are you sure that the benefits outweigh the risks, Bonzai...?

I am running my G.Skill HZ's at stock 800MHz, 2.0V, perfectly stable for about 9 months, with the SANDRA bandwidth of ~6600MB/s.

How much more can you "milk" out of it...?

thats another instance where FSB triumphs over memory speed - my SANDRA bandwidth is 7400~ while running ram at 667 but my FSB is 375mhz vs your 325mhz - running 3-3-3-3 T1

I give up T1 and run it at T2 - my score drops to 7000~

the actual memory speed is more dependent on the FSB rating - I'm sure if I was running at 266fsb my scores would be far lower than yours. T1 isn't always easy to get and as we see its only worth 400-500 MB/s

 

[BonzaiDuck]

Well, between 400 or 500 MB/sec improvement with 1T command rate, and another several hundred MB/sec by tightening tRC, I think I made out by about 1,000 MB/sec.

But I'm going to lose some of that, because I just don't want to run these too close to 2.2V, even if the warranty's still good at the recommended max. I mean -- look what happened with a good warranty, a swift (thank you, Crucial) response, and a pretty fast turnaround?

But "Deez ain't da ones I was runnin' all las' year. Dass a fack, too!!"

 

[JustaGeek]

Exactly - it is a function of the FSB frequency, and the memory should ONLY be able to accomodate the increased bandwidth generated by the FSB.

And then there is an overall speed of the system - the gain of what...? 1 - 2% overall...?

It's only my personal opinion, but "stressing" the memory with higher voltage, frequency, "overtight" timings and command rate will always lead to premature failure.

 

[nefariouscaine]

I wish I knew the answers for such things - like why my kit of 800mhz won't do much over 900mhz at CAS 5 T2 at 2.178v but runs right fine at 667mhz CAS3 T1 on only 1.9v (might do it on 1.8v but never tried...) while the kit of 667 I'm using only runs those same timings at 2.178v. Both kits do well up to 750mhz at CAS3 but I need to give up T1 at that point.

 

[JustaGeek]

It always comes down to a compromise...

Finding that sweet spot...

See, I am happy at stock RAM speed of 800MHz, because not too many people can even run the 4 sticks at 800 on our MB.

But my VDimm is only 2.0..V.

 

[nefariouscaine]

Originally posted by: JustaGeek
Exactly - it is a function of the FSB frequency, and the memory should ONLY be able to accomodate the increased bandwidth generated by the FSB.

And then there is an overall speed of the system - the gain of what...? 1 - 2% overall...?

It's only my personal opinion, but "stressing" the memory with higher voltage, frequency, "overtight" timings and command rate will always lead to premature failure.

I could be wrong (and I'd like to know if I am, very much so actually) but I thought only voltage & maybe frequency played a role in premature failure on ram. More voltage (temps) than any other factor. I am aware of the frequency adding some stress but thought it was lower than voltage. And I didn't think it was physically possible to damage my memory by running all "stock" but making stable latency adjustments - regardless of how low

 

[chizow]

I think its all luck of the draw honestly, and is why the real hardcore memory enthusiasts will often purchase multiple kits and keep the best. Reality is memory manufacturing along with motherboards/chipsets are the red-headed stepchildren of the semi-conductor world. Poor QC and regulations result in widely varying results.

 

[JustaGeek]

Originally posted by: nefariouscaine

Originally posted by: JustaGeek
Exactly - it is a function of the FSB frequency, and the memory should ONLY be able to accomodate the increased bandwidth generated by the FSB.

And then there is an overall speed of the system - the gain of what...? 1 - 2% overall...?

It's only my personal opinion, but "stressing" the memory with higher voltage, frequency, "overtight" timings and command rate will always lead to premature failure.

I could be wrong (and I'd like to know if I am, very much so actually) but I thought only voltage & maybe frequency played a role in premature failure on ram. More voltage (temps) than any other factor. I am aware of the frequency adding some stress but thought it was lower than voltage. And I didn't think it was physically possible to damage my memory by running all "stock" but making stable latency adjustments - regardless of how low

Lowering the timings and command rate is overclocking, too...

It definitely puts more stress on the modules... same as increased frequency with relaxed timings...

 

[BonzaiDuck]

I'm inclined differently, without wanting to get into a ____g contest about it.

Here's this company, with all sorts of a geek [whoops I mean ] enthusiast following for their fat-bodies and D9 parts. They have to worry about their bottom line. They offer a limited lifetime warranty, and the only spec they give for honoring that warranty is the voltage recommended maximum.

Now the frequency and the voltages both stress these parts. The frequency requires more power, and the voltages are also part of that equation -- and the voltages and power create heat and at excessive levels, electron migration.

But the timings? The memory is spec'd for some loose timings, and if you run them too tight, they either work, or they don't work. I need to hear an argument that the timings inherently increase power consumption, but I don't believe they do. Maybe I'm short on some facts and I'm willing to listen. But they only thing I see going south with timings beyond the modules' capability is the data being pushed through those things, and the stability of the system.

Early in 2006, Anandtech published a whole article on the Crucial product line, comparing the Ballistix 667's, the 800s and the 1000's. I think they were able to show that in a range above 667, the 800's were great with timings 3,3,3,8. But those weren't the "spec" timings -- which were looser than that. And since we're running the 800's and the 1000's at lower FSB than they're spec'd at -- at least I am doing so -- there's less stress.

But to make the tight timings work, we go up the voltage ladder. The timings either work or don't work; to make them work, we provide more voltage. So? I can see -- and I've seen too many posts in addition to Nefarious' and his PMs I've received about his RMAs -- suggesting that the modules are not quite up to snuff on their voltage spec.

I THINK they miscalculated. And I THINK they're attempting to honor their warranty by recycling the better halves of their RMA returns. If the modules actually ran at 2.2V, why, they'd even manufacture them to provide a margin of tolerance, since the motherboard "set" voltages aren't always in line with "monitor values" or physically measured values. They just chose the wrong warranty limit. It probably should've been closer to 2.15V. So now, all these people are saying "Hey!! Can you replace these?" And they've moved on to produce DDR2-1066 and higher, and shifted production into DDR3 flavors. So . . . . where are they gonna get replacement modules, to avoid the cost of setting you up with a nice new pair of DDR2-1066-ers?

Anyway, even if my friend here Mr. JustaGeek thinks -- asserts, even successfully argues -- that tight timings shorten the product's life, I think in the meantime -- for now -- that the tighter timings we're ambitious to achieve compel us to move up that voltage ladder, and we're hitting the range of the company's miscalculation in creating its warranty agreement.

I'm trying to be nice to Crucial, because for other than the DDR2-1000's and my slight disappointment with the replacements -- they've been nice to me. But there's too much skinny on the street -- too much howling in the night air. I've had other modules of different manufacture which would fritz out on you for going over the warranty voltage, but keep the setting within that limit and "no problemo."

On the bandwidth thing, if you can squeeze 600 to 1000 MB/sec more from tweaking the timings, I, for one, think it makes a noticeable difference. I don't just think it's 1 or 2%. But as JustaGeek says, it's a balancing act; you have to consider how fast the L2 cache is running; it's about finding a sweet spot, and like Ponce de Leon and the fountain-of-youth, you can look and look and . . . . you ain't gonna find any better. If you do, it may come with a voltage curse, and your RAM isn't gonna have "eternal life and sunshine," either.

 

[Idontcare]

Originally posted by: BonzaiDuck
Basic configuration: Q6600 @ [various speeds]; Crucial Ballistix DDR2-1000 2GB OR DDR2-800 Crucial Tracers [but not both]; BFG 8800 GTS 640MB, Striker Extreme with v.1301 BIOS.

Bonzai man our fates continue to parallel to the extreme!

I killed not 1, not 2, not 3, but 4 sticks of Mushkin Redline DDR2-1000 on my Striker Extreme board.

The sticks are being RMA's now. Here's an excerpt of the "advice" I received:

Many with the 680i based boards users are finding that using over 2.275 on the ram is where the problems start happening. It is best to keep 2.25 or below on your board due to this.

In my case I never went above 2.2V in the BIOS...but the point is pretty clear that i680 boards, and for whatever reason Striker Extreme, seem to really like killing DDR2 when the voltage is >2V area.

My redlines would not even pass memtest+1.70 at DDR2-533 speeds. And they had active cooling too!

I have no idea what Mushkin will replace the Redlines with, my expectations are not too high as I personally don't think Mushkin has to replace the sticks. The sticks were not the problem, i680 and the Striker are what killed the sticks, Mushkin is just being too nice about replacing them. Talk about excellent customer service!

 

[chizow]

I've read a lot of what you posted and not sure if you directly address the issue, but are you having trouble running the sticks at the rated timings and voltages? 1T vs. 2T is a sticking point I suppose, as I don't think RAM kits specifically denote that spec, but again, I think its all a crapshoot beyond what the RAM is tested and rated for.

I know everyone holds Crucial to a higher standard but there's some "rumors" and "speculation" out there working against them as well. Its been widely speculated that memory makers go through extensive efforts to hand-pick their ICs by identifying a high performing batch or part. They pay Crucial a premium for this right, and in return are able to mark the resulting kits up at a premium for those willing to pay for them. What's unclear is if Crucial tests and bins any for their own modules before selling off to the highest bidder first.

In the end, Crucial will guarantee you an excellent D9-based chip that *should* run at its rated timing. This will end up costing you about the same as other premium brands/speeds and may not offer you any performance increase over lower-spec'd modules that slipped through to slower rated parts. We've all taken a look at the memory IC site that lists the memory ICs used on various brands and parts, and seen many slower rated/standard parts have the same ICs as the premium parts. Many have also had their premium parts die or fail to operate at their rated speeds/voltages and require RMA. Again, this points to a general crapshoot and low QC in this segment.

For me its changed my buying strategy. I won't pay more for premium RAM anymore until it starts yielding significant benefits over slightly lower rated standard kits. As it is now, RAM speed doesn't do much for performance and doesn't impact my overclocking at all, so paying more for slightly faster rated modules makes no sense for me when I can get 2x as much RAM for the same price.

 

[Idontcare]

Originally posted by: chizow
For me its changed my buying strategy. I won't pay more for premium RAM anymore until it starts yielding significant benefits over slightly lower rated standard kits. As it is now, RAM speed doesn't do much for performance and doesn't impact my overclocking at all, so paying more for slightly faster rated modules makes no sense for me when I can get 2x as much RAM for the same price.

I went from DDR2-1000 with Mushkin Redlines (4x1GB sticks for $1k back in Dec 06) to DDR2-533 with A-Data (2x2GB sticks for $81 in Dec 07) and I'll be damned if I can't tell the difference in system response or performance.

There was a graysky thread around here highlighting the near-zero improvements in real-world applications for increasing the ram multiplier, I added to it by showing near-zero improvements in my apps of interest from increasing both FSB and ram speeds.

I don't regret wasting the money on Mushkin redlines, it was an experience I needed to live thru firsthand to cure me of my perception at the time. Now I know better and have zero doubt about it.

 

[BonzaiDuck]

Chizow: I pretty much agree on the change of strategy. Right now, since DDR2 is older technology, the prices are pretty near bottom. I know, for instance that DDR2-667 is not "top-end," but for Crucial to offer 667 Ballistix so that the after rebate price is $25 per 2GB kit, it says something about current RAM prices.

I was even tempted momentarily to pick up some Tracer DDR2-1066 and put the RMA replacements in the "spare parts" bin, because the 2GB kit of those works out to an after-rebate price of just below $70.

But I have enough. I pulled my Ballistix 1000's last night and reinstalled the Tracer 800's.

IdontCare:

I don't think it's just the STriker board, but it MAY WELL BE the discrepancy between "monitored" and "set" voltage values -- if "monitored" is anywhere close to the real thing. The two boards I have, and I think those of some other folks, showed a consistent 0.035V bias of "monitored" versus "set." So setting the voltage at 2.2V is going to give a reading around 2.24. Whether this matters or not, I'm using the larger of the two voltages to govern my choice of "setting."

Certain advice posted at ASUS for the Striker notes that "it [the board] doesn't like high-volted memory."

But mainly, as I said . . . . whether or not the modules are 'spec'd" at a 1T command-rate, if you stick your Deming paddle into the urn and come out with all red balls -- even just four or five of them (knowing they represent kits of two), and you pick one ball out of another urn (the RMA return bin) and pick just one ball and it's white, it may not be conclusive, but it leads you toward a tentative conclusion -- or a guided speculation, anyway.

 

[Idontcare]

Originally posted by: BonzaiDuck
But mainly, as I said . . . . whether or not the modules are 'spec'd" at a 1T command-rate, if you stick your Deming paddle into the urn and come out with all red balls -- even just four or five of them (knowing they represent kits of two), and you pick one ball out of another urn (the RMA return bin) and pick just one ball and it's white, it may not be conclusive, but it leads you toward a tentative conclusion -- or a guided speculation, anyway.

LOL dude, I wonder how few people on this forum have the background to grasp both the history of the statistics reference you give let alone the humor with which you leverage it and deliver. As always I appreciate the good read.

 

[BonzaiDuck]

Well, it's always a pleasure. And I need to say that in a few years' time, I've learned a heck of a lot from my fellow forum members here, including Idontcare.

 

[nefariouscaine]

Hey man - you can always RMA again and spin the wheel for a red ball

 

[BonzaiDuck]

That is so, so true . . . . I guess. I have to say that they were really nice to me. They replied personally to my e-mail "thank you note," asking to use it in a promotion. So you see, they not only have a good product that probably needs a little more rigorous QC and less eager legal-eagles on the warranty department formulating "recommended warranty maximum" language, but they're TRYING to be good.

TRYING TO BE GOOD is worth a WHOLE LOT.

But I don't think I can RMA a set of DDR2-1000's because they won't run at CMD=1T. I'm pretty sure of it.

EDIT:

Of course, I could lie and say they didn't work anymore, but there's only some unknown probability that it would get me better modules. Now -- there's a different slant on moral relativism . . . .

 

[lopri]

Originally posted by: chizow
I think its all luck of the draw honestly, and is why the real hardcore memory enthusiasts will often purchase multiple kits and keep the best. Reality is memory manufacturing along with motherboards/chipsets are the red-headed stepchildren of the semi-conductor world. Poor QC and regulations result in widely varying results.

Yup. I'd add SPP (NB) to the 'luck' factor. You need a real good chemistry between SPP and DIMMs to get 1T to work on 680i. Not necessarily the case is more expensive sticks are better at 1T. But if modules doesn't do advertised specs on 680i (I'm guessing DDR2-1000/CL5/2.2V ere?) when configured with 2 sticks, then RMA is in order, for sure. In any case, don't be in a rush when you evaluate memory. Unlike CPUs, there can be other factors than just the sticks. One parameter set wrong in the BIOS can make wonderful set of modules look like a ass.

Also, I highly suggest you guys start using 'Unbuffered' bandwidth test in Sandra. The bandwidth test defaults to 'Buffered' mode, and from my experiences it has little to no bearing on actual performance, other than obvious occasions.

 

[lopri]

The biggest reason on premature death of memory on 680i is overvolting. Many 680i boards give extra ~0.1V than what you set in the BIOS. All EVGA boards I owned did this and I confirmed it by a DMM. But then again, if you're not too ambitious, 680i is also very generous in a sense. Often times you will achieve better frequency than Intel chipsets with same voltages. (Performance is a different issue here) The Stryker board, however, has been known to be a very different beast and I have no first-hand knowledge of it.

Memory is overrated when it comes to general performance, that much is for sure. But half the time the real fun in OC'ing comes with memory tweaking and checking those minuscule improvements. Otherwise OC'ing would be just all about cooling.