High Failure Rates for DDR2

[bgavin]

I'm starting to get memory failures in various DDR2 machines.
It is either DOA from the vendor, or dies 6~12 months down the road.

I spent a lot of time reading the Newegg comments for all brands of DDR2.
The complaints and ratings percentages are remarkably close across all product lines and brands.

All performance DDR2 is over-voltage from 1.8v stock to 2.1v per manufacturers recommendation.

Q: do you think 16.6% over voltage is killing DDR2 over time?

I am building three critical machines which cannot abide any memory problems.
To see if there is merit in my question, I will run all three machines at 1.8v and the slower 5-5-5-18 JDEC timings.

[willawake]

Re: High Failure Rates for DDR2

i run several ddr2 machines with zero problem
how is it possible chips from different manufacturers have some problem
are you suggesting the spec is wrong?

[bgavin]

Re: High Failure Rates for DDR2

I think they are dying from over voltage.

Memory, like processors, all start life on the same wafer.
The yield is tested, and each processor is assigned a speed rating.
Some are 3.0 GHz, some are 2.8, and so forth.

I suspect memory is the same.
It is all designed as DDR2, JDEC, 1.8v and 400 MHz.
Some chips test out faster and are purchased by Corsair, etc, and sold as high performance chips.

The failure rate for 2.1v DDR2 in the 3~12 month time frame is very similar between vendors and product lines. OCZ, GSkill, Corsair, Patriot and even Crucial (one of the largest actual RAM manufacturers).

When Crucial is failing at the same rate, that tells me it is the mode of operation that is killing these chips. I have servers running DDR and SDRAM at stock voltage/timings for 10 years on Crucial chips without a burp.
Granted, DDR2-800 is a wee bit faster than the older stuff.

I just finished a complete set of benchmarks comparing 2.1v/4-4-4-12 and 1.8v/5-5-5-18 in my workstation. The memory read is 2.7% faster with the hot timings. Memory write is the same in both cases. All the other CPU tests are virtually the same.

I conclude that over-volting DDR2-800 is not worth the 2.7% boost for Reads.
None of the failing machines run hot. The memory is about 37C on my IR meter.
All are equipped with Antec free breathing cases, multiple 120mm fans, and cool running processors. Heat is not the problem. It has to be voltage.

[Wizard]

Re: High Failure Rates for DDR2

GSkill and Patriot are lesser brands while Consair is average. Crucial is decent. Should have tried Mushkin and others. OCZ tend to use Samsung ICs and some other.

Are ram sticks covered with heat spreader plates? I tend to buy these for this if I wanted high reliablity.

What kind of chipsets VIA, SiS, Intel (?) that you usually see that had bad memory at 2.1V? And what was your initial reason you let them run at 2.1V? Is it OEM mainboard or Dell/Lenovo style, or Asus? Another reason I ask because some mainboard maker is too loose with the memory voltage filtering, keeping signal clean and least ringing/spiking as these kills memory if these exceed max spike voltages. Very acute so with DDR3 as Anandtech had documented while ago (last winter or so).

Some memory makers tested and rated their memory modules to do 2.1V, some 2.0V, Even 1.9V, all these above voltage is mainly for permitting overclocking and lower timing specs. But most stock DDR2 sticks are 1.8V. Even Nvidia told Asus maker to warn that chipset 6150 fails or problems if memory is above 1.8V.

Cheers, Wizard

[gg1978]

Re: High Failure Rates for DDR2

I've had one stick of regular DDR2 die, in a box at work.. I buy the corsair extra performance stuff, but don't overvolt and run at stock timings.. Works just fine that way, and comes out cheaper after rebate.. I like Corsair, Kingston, and Crucial for RAM..


If memory reliability is critical, you should be running ECC or Registered ECC, depending on what variety your board supports.. Yet another reason i like AMD, is that their support for ECC, on mainstream solutions, is MUCH better than intel.. For intel's, it seems you must buy either a server class S775, or Xeon class i7 CPU for ECC support...

[Topcat]

Re: High Failure Rates for DDR2

I have 2 dead DDR2 DIMM's here. Ironically enough, they're a matched pair of 512mb, and micron (maker of crucial). They were dead when I got them (came in someone's sx280 with bad caps). I have several other systems here running DDR2, including the one I'm typing this post from, no issues. On that note, I run corsair or transcend memory in all my boxes, and always reg/ECC if the board supports it. I don't over-volt or overclock, so no worried there.

[PCBONEZ]

Re: High Failure Rates for DDR2

I just found a bad Patriot module in a customer's box.
-
I think part of the problem with this one [and possibly many others] is the stupid aluminum heatsinks. - They are so thick that if the modules are right next to each other in the slots there is like no room left for and airflow between the RAM sticks.
.

[bgavin]

Re: High Failure Rates for DDR2

My dead ones are in multiple Gigabyte boards. Some P35, some P45.
All Corsair memory comes with clear printed instructions on running at 2.1v and 4-4-4-12 timings. The DDR2-1066 is also 2.1v, but different timings for the faster bus.

I've lost: Dominator, XMS2, XMS2-DHX, and ValueRam all in Corsair DDR2-800 speeds, all at 2.1v per manufacturer.

After running the benchmarks and finding that 2.1v only yields a 2.7% boost on memory READ, it ain't worth the risk. None of the other memory tests showed any significant boost from the higher voltage memory.

I correlated user happiness ratings from Newegg for all their brands. OCZ, Mushkin, etc, have the same percentages of unhappiness. This wide spectrum of uniform complaint ratio tells me it is not vendor-specific.

Kingston has a 91% happiness rating on a product running at 1.8v and slower timings.

I will do the three new builds with Corsair Dominator, but at 1.8v JEDEC timings. I think the 16% over voltage is killing them over time.

[gdement]

Re: High Failure Rates for DDR2

I suspect memory is the same.
It is all designed as DDR2, JDEC, 1.8v and 400 MHz.
Some chips test out faster and are purchased by Corsair, etc, and sold as high performance chips.

The failure rate for 2.1v DDR2 in the 3~12 month time frame is very similar between vendors and product lines. OCZ, GSkill, Corsair, Patriot and even Crucial (one of the largest actual RAM manufacturers).

If you look up the datasheets for the RAM chips, I bet you'll find none of them are actually rated for 2.1v. Some manufacturers put their own logos on the chips so you can't look them up, but the Crucial probably has the original Micron part number on it.

I haven't looked up many DDR2 chips, but from SDRAM and DDR1 I've never found a single chip that supported anything outside of JEDEC standard voltages. Yet there were lots of higher voltage DDR1 gamer modules out there, so I assume they were simply overvolted and operating out of spec.

Same is probably happening with DDR2, where the gamer crowd has seemed to adopt a nonchalant attitude about running 2.1v.

I wouldn't use overvolted memory on a critical computer, just game machines. Buying it for price and running it at standard voltage with SPD timings should be fine though.

[370forlife]

Re: High Failure Rates for DDR2

I see no need in overclocking my OCZ DDR2-800. It runs very fast at stock timings and voltages. Vista business 64bit, I'm not even joking, boots faster than XP pro 64bit on my machine (5000+ dual core, 4gb OCZ Vista Upgrade ddr2-800, nvidia 750a chipset)

Though, I'm sick of vista crashing when about 5 mins into a dvd using K-lite Codec pack.

[kc8adu]

Re: High Failure Rates for DDR2

i have recently rma'd ddr2 from kingston,corsair,and buffalo.
seems as it will die in 30-60 days if it is going to.
have 2 dead sticks of nanya that came in a gx280.dead as in no post in any machine.board had the dell 280 bad caps.all the caps on vtt were open.that must have been what killed it.no fault of the ram.

[kc8adu]

Re: High Failure Rates for DDR2

btw you have to watch that factory overvolted ddr2.one of our regulars brought in some ocz "special ops" ram with camo heatsinks and rated to run at 2.1v.
he did not understand how to set it up so i showed him.
it ran fine about 15 minutes on the bench and then killed the vdimm reg.
easy fix but that board could not take the load from this ram.this ram is fine in one of my asus boards so it was just the marginal design of the msi board.

[MixMasta]

Re: High Failure Rates for DDR2

I've had memory go dead after a year or so too... it was also DDR2 "performance ram". I'm not too surprised, considering how many sticks of ram show up bad straight out of the box compared to years past, but remember many of these manufacturers offer lifetime warranties... Now it's a matter of whether your time is worth more to simply buy a replacement from Newegg or try the RMA process...

[bgavin]

Re: High Failure Rates for DDR2

My problem isn't so much replacing the failed memory, but the fact that it failed in the first place. My clients are all business types, no gamers. BSOD 0x50 is not acceptable.

The Corsair RMA process isn't particularly fast.
I was so exasperated with 3 consecutive DOA right out of the box, that I asked them to send me a matched, tested pair. Whatever they sent worked correctly.

I burn in each machine for 24 hours with Memtest86+ before it goes out.
Any memory failures at all are not acceptable.

My problem stems from trusting the manufacturers stated operating environment of 2.1v and the faster timings. In light of my testing, I now refuse to push any faster than the JEDEC standards. These are readily available for viewing with CPUID.

The memory in use in all cases is on the approved memory list for the Gigagbyte board in place.

[MixMasta]

Re: High Failure Rates for DDR2

I couldn't even imagine selling a business client performance memory in the first place... They would never know the difference from DDR2-800 and DDR2-1066... Forget cas latency.

My problem isn't so much replacing the failed memory, but the fact that it failed in the first place. My clients are all business types, no gamers. BSOD 0x50 is not acceptable.

The Corsair RMA process isn't particularly fast.
I was so exasperated with 3 consecutive DOA right out of the box, that I asked them to send me a matched, tested pair. Whatever they sent worked correctly.

I burn in each machine for 24 hours with Memtest86+ before it goes out.
Any memory failures at all are not acceptable.

My problem stems from trusting the manufacturers stated operating environment of 2.1v and the faster timings. In light of my testing, I now refuse to push any faster than the JEDEC standards. These are readily available for viewing with CPUID.

The memory in use in all cases is on the approved memory list for the Gigagbyte board in place.

[acstech]

Re: High Failure Rates for DDR2

I have a little rule that keeps me out of trouble. Nothing above 1.95v for DDR-2. No exceptions. If I can get good timings at 1.9v or 1.95v then fine, but I'm not putting more voltage than that into it. Often times I can set the voltage at 1.8v and run the quicker timings anyway.

My usual stuff is this:

http://www.newegg.com/Product/Produc...82E16820231219

I always figured 2.1v was asking for trouble...

Edit: Oh, and I have yet to see a dead DDR-2 module.

[NxB]

Re: High Failure Rates for DDR2

I've only seen dead laptop DDR2 from overheating. Not much success bringing it back with the heat gun either. It runs for a day or 2 then goes back to getting errors or being dead. No dead desktop sticks yet.

[Wizard]

Re: High Failure Rates for DDR2

NxB, do you know what brand maker of laptop DDR2 that is failing?

Heatspreaders on memory sticks, I mean thin plates. Not THICK.

Oh, thanks for the comments bgavin. Is your main recommandation of gigabyte boards? For business, I rather use very dumb boards like Intel, Foxconn due to bios designed towards business sector and permits no overclockings. Oh, I use Asus with all settings to defaults with few tweaks for certain things but not for performance and 1.8V ram sticks only. No issues. Only ones that died were generic quality even low quality brand names like GB, Azza etc died and they were not my purchases.

DDR 1066 does have 1.8V in some models but not recommanded in my opinion. Also DDR2 1200 are all above 1.8V. I notice that most boards even intel specify DDR2 800 as max. There is a reason. But Asus does support higher DDR2 above 800 but I don't chase these memories above 800. That's main reason I haven't had any issues at all.

Cheers, Wizard

[bgavin]

Re: High Failure Rates for DDR2

I use Gigabyte boards because they are long term reliable.
Yes, the Intel and Foxconn business boards are tightly locked down, have wet caps, and suffer higher dissatisfaction rates. AFAIK, this Gigabyte board does not support ECC.

I like to use the same board for both servers and workstations. They become obsolete so quickly... if a server loses a board I can swap an identical board from a workstation. Rebuilding an Exchange server with a new board technology is a painful experience. Especially when the business is flat down.

I'm partial to the E8400 and 1333 FSB, so I look for stable boards that deal with that bus speed. I'm perfectly happy with DDR2-800 performance... I got bit by following the manufacturers recommendations. Even Crucial (Micron) is pushing the 2.1v thing. I think they are so desperate for sales, they will push anything and calculate the 10% failure rate and RMA as the cost of getting sales.

I have a new board on the bench as of this writing. GA-EP45-UD3R with Dominator DDR2-1066 DHX installed. It's running at 1.8v, 5-5-5-15 by JEDEC SPD. Memtest86+ has finished 7 passes without error. I will let it run 24 hours to be sure. RAM temps are 35.5C, open board on the bench.

[PCBONEZ]

Re: High Failure Rates for DDR2

Until recently DDR2-1066 had no official JEDEC spec or approval.
That the main reason no one except companies like Assus listed it in their mobo specs.
.