Should premium memories really work as advertised?

[Timothy Miller]

Not too long ago, I paid nearly $250 for a pair of 2GB DDR3 premium memory modules. They advertise that they'll meet timings 6-6-6 at 1333MHz at 1.75v.

The first pair of DIMMs didn't work anywhere near to spec and seemed to get worse with time. I called the vendor, and they were very quick to send me replacements. Their customer service seemed to have a habit of ignoring my questions, but they were very efficient with the exchange.

With the replacement pair, they seem to work NEARLY to spec, although I had to set the CAS latency to 7. The system is stable, but I ran the GIMPS Mersenne Primes stress test (http://www.mersenne.org/), and it reported a computation error. Since I was isolating the memories, all other system components are running at default speed and voltage, except the memories, which are configured to vendor spec (overclocked based on their binning process). Thus, I think it's likely that the computation error is a memory error.

Is this really a quality problem? Following their instructions and using more than adequate cooling, can I expect the memories to perform as advertised under all circumstances or not? How is it with other vendors?

Should I be upset about this and demand that the vendor send me something better? Or should I relax the timings further and just be mildly annoyed that I'm not getting _quite_ what I paid for, but move on with my life?

In other words, what is reasonable?

[maxfly]

if you paid for cas6 modules you should get cas6(at the rated speed).anything less isnt what you paid for.if your certain the mb isnt the culprit,i would say send them back.

[hafaphoto]

Give them hell, or get a refund from them and buy a different brand. I concur with maxfly.

[Timothy Miller]

My motherboard is rather high-end. It's an MSI X48 Platinum. It's designed for overclocking, and support 1600MHz memory, although I'm only running at 1333MHz.

Also, interestingly, I notice that one of you us using OCZ Reaper. Mine are OCZ Reaper as well. Is there any discussion somewhere of the various vendors and their relative ability to live up to their advertised specs?

[doz]

http://www.ocztechnology.com/products/memory/ocz_pc3_10666_reaper_hpc

Im assuming thats what you purchased?

I cannot speak for their DDR3, but OCZ has been top notch in my book. In my old AMD system, Ive been using PC3200 for the last 4 years without a problem and Im currently running PC8500 Platinum series with any issues. Granted, its a bit touchy trying to modify the timings, but it will run @ the rated speed no problem.

Have you checked the Compatibility List for your motherboard VS your DDR3 modules?

**UPdate**

I just checked the OCZ website and they arent showing compatibility with your motherboard. That could be the issue which would mean its technically not the ram thats a problem, but a compatibility problem. Although their compatiblity listing is limited. Doing some more searching for you :D

[JaY_III]

Can you reach the rated speed at 1.8-1.85V
Often times the BIOS reports a higher voltage than what is actually being used.
Or can you confirm that your are pumping in 1.75V with a multimeter.

What is the voltage/temp of your northbridge?
Remember you are also running that out of spec, so that has to be 100% good to go for smooth RAM operation

[Timothy Miller]

Yes, the model I have is OCZ3RPR13334GK.

Here's the sensor information I can get while the system is up, running the sort of load that showed the computation error:

f71882fg-isa-0a10
Adapter: ISA adapter
3.3V: +3.34 V
Vcore: +1.21 V (max = +2.04 V)
Vdimm: +1.17 V
Vchip: +1.92 V
+5V: +4.91 V
12V: +14.06 V
5VSB: +5.84 V
3VSB: +3.33 V
Battery: +3.22 V
CPU: 815 RPM
System: 1800 RPM
Power: 0 RPM ALARM
Aux: 0 RPM ALARM
CPU: +47 C (high = +85 C, hyst = +84 C)
(crit = +43 C, hyst = +42 C) sensor = transistor
System: +26 C (high = +85 C, hyst = +81 C)
(crit = +100 C, hyst = +96 C) sensor = transistor

I notice that the Vdimm is reading quite low. I had contacted MSI about that, and their response is that the sensor program I'm using could be producing incorrect numbers. There might be a missing offset. They suggested I install Windows and use some other tool. As if. One thing I can do is reboot into the BIOS and check the stats there. I can at least get a more trustworthy voltage reading.

I have a digital multimeter. Unfortunately, I really have no idea where to measure. I'm wary of setting the voltage too much higher (it's at 1.79 right now) because that will shorten the life of the device. JEDEC specifies that the parts are supposed to survive up to 1.95 volts (IIRC), but they're not expected to perform properly or last incredibly long. Of course, I can run a short term test (as soon as my genetic algorithms experiment is done).

The northbridge voltage is default right now. 1333MHz isn't out of spec for this particular chipset.

There are two sorts of reasons for the data corruption. One is that the problem is internal to the memory, caused by timing numbers that are too tight. The other is a communication issue, where the NB got bad data over the bus. The bus is rated at 1600MHz, but that doesn't mean that the signals are at the right voltage level (or swing anyhow). CAS latency, on the other hand, is an issue purely internal to the RAM chips.

OCZ has started ignoring comments I add to my trouble ticket.

It's certainly a possibility that the problem is on my end. I don't know how scientifically rigorous OCZ's testing is, though. I've written some really SSE-heavy code that really stresses the memory system. If all they do is try the timing numbers on a few mobos with a weak test like memtest86 at ideal temperatures, then that's not a really sound strategy. But they don't reveal their testing methods, do they?

I really appreciate the help you all are giving me!

[JaY_III]

Memtest86+ is pretty much the standard for stability,
If you have something better by all means, share with us, you might have the new best thing....

You really need to share what your BIOS says your Voltages are. But even then i dont fully trust them.

What Power supply do you have?
As not only does your VDimm (like you said) look low, but your +12V is VERY low too.
That can cause general system instability.

I would grab your MulitMeter and confirm your 12V, and its safe to do when the system is up and running IMHO.
Not 100% sure what your level of knowledge is so sorry if this is redundant to you.

Yellow = +12V
Red = +5V
Orange = +3.3V
Black = Ground
Blue = -12V
White = -5V

Reason i would say they should all be checked is your +12V, too low for my liking and makes me worry about the PSU

[doz]

You may want to contact MSI and have them give you a memory compatibility list. I search near and far and couldnt find one (which makes me wonder) for this board. Even downloaded the x48 manual.

I did some searching and it seems theres a few others who are having issues running certain ram modules w/ this board. Like I said, OCZ isnt showing compatibility. I checked some other ram sites and they do show compatibility so it could be that.

Is there a page in your BIOS where it shows the current running voltages (not just the settings)? If so, see what the VDimm is running at in the bios.

and +2 for the 12v rail. Either your monitoring program is junk, or maybe the board sensors, but the 12v is showing 14v. Seems pretty high.

[Timothy Miller]

Memtest86+ is pretty much the standard for stability,
If you have something better by all means, share with us, you might have the new best thing....

Maybe memtest86 has improved, but I have only ever found it to be reliable at finding hard memory errors, like if a RAM has gone bad. Otherwise, you have to do things with more complex and demanding memory access patterns, like compiling Firefox or running mprime. memtest86 does a lot of linear memory tests, but I don't think it does any random address/random data tests, or if it does, it's still not aggressive enough.

You really need to share what your BIOS says your Voltages are. But even then i dont fully trust them.

Indeed. I'll get these numbers within the next day or so. I still have something running that I can't stop without having to start it from scratch.

What Power supply do you have?
As not only does your VDimm (like you said) look low, but your +12V is VERY low too.
That can cause general system instability.


Interesting point. This is my PSU:
http://www.newegg.com/Product/Product.aspx?Item=N82E16817256008

I'll provide the BIOS's report on all voltages it measures.


I would grab your MulitMeter and confirm your 12V, and its safe to do when the system is up and running IMHO.
Not 100% sure what your level of knowledge is so sorry if this is redundant to you.

Yellow = +12V
Red = +5V
Orange = +3.3V
Black = Ground
Blue = -12V
White = -5V


I can't find my digital meter, unfortunately, so I'm having to use an analog one. I think I know where my digital one is, and it's a 40-minute round trip, so I'll do that a bit later. I pulled out one of the unconnected power connectors with four conductors. The 5v is dead on. The 12v might possibly be a tad low. It's hard to be sure.


Reason i would say they should all be checked is your +12V, too low for my liking and makes me worry about the PSU

I hadn't thought of this. When my GA stuff is finished running, I'll conduct more detailed measurements. If it's the PSU, I'll complain to its manufacturer and/or newegg and see if I can get something better.

[Timothy Miller]

Just thought I should point out that in addition to the help I've gotten here, OCZ has finally responded to my queries. I was worried they were ignoring me, but I think they took a while because my case had to be escalated to someone higher up. Their suggestions are good. I should have a little more patience. :)

[JaY_III]

What did OCZ say?


Oh and as i am sure you figured out on your own, i ment to say 14V is too high for my liking on the 12V rail.


Any word on what your multimeter or analog (they work just as well) is reading?

[Timothy Miller]

Ok, answering some of the questions here.

The BIOS provides me with very limited voltage info. Here's what it says:

Vcore = 1.240
3.3V = 3.344
5V = 4.961
12V = 12.056
5V SB = 5.016

Here's what I get using my multimeter:

Orange: 3.35
Red: 4.97
Yellow: 12.09
Blue: -11.80
White: There's a hole for it, but no wire!

Those all look good to me, although the system is not under load right now.

As for what OCZ told me:

"okay, well this kit should work so lets try and fine tune it. just to make everything stable id like you to try the following settings: cas 8 trcd 8 trp 8 tras 28 command rate (cmd) 2t trc 36(or higher) trfc 80 (or higher) memory voltage of 1.75 .... on a few x48 boards ive heard that 1.66v or 1.67v actually helped a lot with stability. Northbridge voltage increase of 0.15v (so if its 1.2, then it would be changed to 1.35) if we can get these settings working then we can try more aggressive timings. you are only using 2 sticks total correct? thanks <name of tech>"

This BIOS stupidly won't tell me what is the "auto" voltage for anything. I've taken a guess and set the NB voltage to 1.33. I don't know how much higher, but it's higher than auto, because it improved stability when I was trying to overclock earlier.

I'm going to run mprime now with their recommended settings and see what happens.

Any other thoughts? Thanks again for the help!

[doz]

Again, what are you running the ram at voltage wise? Its kind of important. It shows 1.17vdimm above in your post, but that cannot be right.

[Timothy Miller]

Again, what are you running the ram at voltage wise? Its kind of important. It shows 1.17vdimm above in your post, but that cannot be right.

Good question. They suggest 1.75. I've tried 1.75. I've also tried 1.79. Right now, it's at 1.67 because OCZ suggested that that's more stable with some X48's.

Can anyone suggest how I might measure the voltage? The chips have a heat spreader on them. Not a lot is exposed, so I'm not sure how I could get at anything to measure it.

Oh, one thing I had noticed some time back was that what the sensors program under Linux reports changes depending on load. When the system is unloaded, it'll show REALLY low voltages for the RAMs. Impossibly low, like they would fail. On the other hand, if the system was under load, the voltage it reports is almost spot-on what I set it to.

Yeah, right now, it's reporting Vdimm at 0.08v. The system is completely unloaded.

And now, something's going wrong again. I was able to run mprime for probably 24 hours before it reports a compute error. Now, it takes about 10 seconds. This is driving me nuts. I'm going to shut down, set the clock rate to 1066, and try again. I need to double check to make sure that the CPU isn't being a problem by running the memories at a safe speed.

[Timothy Miller]

Now these memories seem to be malfunctioning. As instructed, I set them to 8-8-8-28-80, 1.67v, 1333MHz.

I started mprime at Fri Oct 10 21:15:29 2008. It ran successfully until Sat Oct 11 06:07:20 2008, when it quit, reporting a compute error. That's about 8.9 hours. After that, I ran it again, and it quit immediately with compute errors. I waited a bit and ran it yet again. It lasted about 10 seconds before reporting a compute error.

Now, I'm running them at 1066MHz. They're working fine and have been running mprime for about an hour. No errors so far.

I have followed OCZ's instructions precisely. After making settings as instructed, I stress-test them to be sure the whole system works reliably. After running the stress test, I discover that memories appear to have become permanently damaged.

I should also note that the airflow across the memory heat sinks is very good. There's no chance that they're overheating due to poor cooling.

What is going on here?

[Niku-Sama]

i think your caught up in new tech glitches, it seems when a new style of system memory comes it out takes the manufacturers a while (on both ends) to get things working happily together but it eventualy makes it there, i remember many issues when it went from SIMM to DIMM its just kinda how it works.

odds are if your getting errors i would hit up OCZ again and see what they say and mabe see if they will send out another replacment, memory has a lifetime warranty for a reason and i think this is evidence why.

odds are they are going to ask you to run Memtest so just get things back to what the stock settings should be at and give memtest a shot and when it errors call em up for replacments.

[Timothy Miller]

Quick update: mprime has been running for almost 4 days now with zero errors. The only variable I changed was that I underclocked the RAMs to 1066MHz. I think I've adequately isolated the problem to the memory system. Now, if OCZ will respond to my trouble ticket, that would be very nice.

[Timothy Miller]

In case anyone is interested, I have given up on OCZ. They don't answer their phones, they don't respond to my support tickets, nothing. I'm wondering if they're going out of business or something.

I had to argue a bit with them, but Newegg decided to give me a refund on them. When they tried to refer me to OCZ for tech support, explaining how contacting OCZ is impossible seemed to help with that. Of course Newegg is charging me a restocking fee, but I can kinda understand. It's not Newegg's fault. I'm just glad I can get most of my money back.

I'm going to get some Kingston memories. They sell some fully JEDEC compliant (non-overclock) 9-9-9 memories for about $145. Compare that to the $245 I paid for the 6-6-6 (that wouldn't perform at 6-6-6) memories from OCZ. I'm going to get 8 gigs of that from Newegg and go. I'll overclock the CPU probably, and I'll TRY fiddling with the timing numbers on the RAMs, but probably not the voltage.

And speaking of voltage, I think that was an issue with the OCZ RAMs. 1.75v violates JEDEC spec, although it makes physical sense that if you up the voltage you can up the clock speed and reduce timing delays. The odd thing was that while the OCZ memories would work nearly to their advertised spec initially, after 9 hours of aggressive memory testing, soft memory errors would start to occur, and the degradation was permanent. It was as if the memories had become physically damaged by a combination of overvolt and very heavy load. (And mind you, the voltage was set as OCZ specified.) I'm guessing that OCZ never bothered to put any of their RAMs through such an aggressive test, at least not the ones they were selling.

Imagining how one might go about setting up a testing facility for bin-testing RAMs, I can see how it would not be cost-effective to test every RAM stick for 9 hours. That being said, they could test SOME of them for that period, and they should realize quickly that their claims don't generalize. It's false advertising due to insufficient testing and wishful thinking.

The fact that I could not contact OCZ by phone (either sales or tech support) is also disturbing. There's no answer, not even an automated system. Where did they go? Doesn't matter anymore, though.

Thanks again to everyone who offered me advice!

[erricrice]

Spend ten bucks more and get some gskill that will do the timings you expected the OCZ stuff to do. http://www.newegg.com/Product/Product.aspx?Item=N82E16820231200