EXTREME Overclocking Forums
Home | Reviews | Forums | Downloads | RealTime Pricing Engine | Folding Stats Contact Us


Go Back   EXTREME Overclocking Forums > AMD Specific > AMD Memory

Welcome Guest Visitor! Please Register, It's Free and Fun To Participate!
The EXTREME Overclocking Forums are a place for people to learn how to overclock and tweak their PC's components like the CPU, memory (RAM), or video card in order to gain the maximum performance out of their system. There are lots of discussions about new processors, graphics cards, cooling products, power supplies, cases, and so much more!

You are currently viewing our boards as a "guest" which gives you limited access to view most discussions. You need to register before you can post: click the register link to proceed. Before you register, please read the forum rules. By joining our free community you will have access to post topics, communicate privately with other members (PM), respond to polls, upload your own pictures, and access many other special features. Registration is fast, simple, and absolutely free! To start viewing messages, select the forum that you want to visit from the selection below.

After you have registered and read the forum rules, you can check out the FAQ for more information on using the forum. We hope you enjoy your stay here!

Note To Spammers: We do not allow unsolicited advertising! Spam is usually reported & deleted within minutes of it being posted, so don't waste your time (or ours)!


Please Register to Post a Reply
 
Thread Tools
Old 05-23-2012, 09:31 PM   #21
larrymoencurly
Learning To Overclock
larrymoencurly's Avatar
Senior Member
 
Posts: 170
Last Seen: 01-24-2014
iTrader: 0 / 0%
Quote:
Originally Posted by Josie Wales View Post
Your basic reasoning is not flawed. You want your memory specs to exactly match your expectations. There is nothing wrong with that. Beyond that I feel you are making a lot of assumptions and over-thinking a simple product purchasing decision.
My only assumption is that products built from properly rated quality parts are much more likely to work well than products built from junk or overrated parts, whether the user overclocks or not, so I buy only Samsung and no-heatsink Crucial. I don't know what could be more simple.
Offline
    Register to Reply to This Post
Old 05-23-2012, 11:25 PM   #22
kow_ciller
A billion megahurtz
Regular Member
 
Posts: 50
Last Seen: 12-15-2012
From: Texas
iTrader: 0 / 0%
Just so you know it is very easy to figure out which ram chips are on which sets of ram.

Most makers have a code built into the product number that allows you to identify which IC is used.

http://forum.corsair.com/forums/showthread.php?t=68811

Its also funny that you want all memory to be tested on memory stressing machines instead of PC's.
Last I checked the basic idea of memory is to run in a PC, not a memory stressing machine.
United States  Offline
    Register to Reply to This Post
Old 05-24-2012, 12:05 PM   #23
larrymoencurly
Learning To Overclock
larrymoencurly's Avatar
Senior Member
 
Posts: 170
Last Seen: 01-24-2014
iTrader: 0 / 0%
Quote:
Originally Posted by kow_ciller View Post
Its also funny that you want all memory to be tested on memory stressing machines instead of PC's.
Last I checked the basic idea of memory is to run in a PC, not a memory stressing machine.
Not everybody wants that. Apparently I'm the only one here who's fussing over it, and PCs are easier on memory than those testing machines are.

Still, do you really want memory from a company that says 2 errors per module is OK, even though they guarantee their products for life? This is one of the top 5 companies.
Offline
    Register to Reply to This Post
Old 05-24-2012, 03:33 PM   #24
kow_ciller
A billion megahurtz
Regular Member
 
Posts: 50
Last Seen: 12-15-2012
From: Texas
iTrader: 0 / 0%
Quote:
Originally Posted by larrymoencurly View Post
Not everybody wants that. Apparently I'm the only one here who's fussing over it, and PCs are easier on memory than those testing machines are.

Still, do you really want memory from a company that says 2 errors per module is OK, even though they guarantee their products for life? This is one of the top 5 companies.
If the memory runs rock-solid for everything I do with my computer, I wouldn't care if it gives 5000000 errors
United States  Offline
    Register to Reply to This Post
Old 05-24-2012, 04:04 PM   #25
Josie Wales
Extreme Overclocker
Josie Wales's Avatar
Senior Member
 
Posts: 2,388
Last Seen: 03-25-2017
From: St Paul
iTrader: 4 / 100%
Quote:
Originally Posted by larrymoencurly View Post
Still, do you really want memory from a company that says 2 errors per module is OK, even though they guarantee their products for life?
"2 errors" by itself is not a standard. Please provide the full specified standard and a supporting link. How does this manufacturer/brand standard compare to the recognized industry standard? How does this memory error rate compare to the allowable error rate for your CPU and GPU?

Last edited by Josie Wales; 05-24-2012 at 04:38 PM.
United States  Offline
    Register to Reply to This Post
Old 05-25-2012, 01:24 AM   #26
larrymoencurly
Learning To Overclock
larrymoencurly's Avatar
Senior Member
 
Posts: 170
Last Seen: 01-24-2014
iTrader: 0 / 0%
Quote:
Originally Posted by Josie Wales View Post
"2 errors" by itself is not a standard. Please provide the full specified standard and a supporting link. How does this manufacturer/brand standard compare to the recognized industry standard? How does this memory error rate compare to the allowable error rate for your CPU and GPU?
I don't have the full specified standard because the company wouldn't tell me much at all, and their representative didn't seem to understand. Actually I was surprised he even admitted they allowed 2 errors per module (0 for 2133 MHz products). I'll PM you the name.

What are the error rates for CPUs and GPUs, and do they trap them, or do errors just cause undefined states? All I know is that some CPU L2 caches have ECC.

Last edited by larrymoencurly; 05-25-2012 at 01:24 AM. Reason: .
Offline
    Register to Reply to This Post
Old 05-25-2012, 12:08 PM   #27
Josie Wales
Extreme Overclocker
Josie Wales's Avatar
Senior Member
 
Posts: 2,388
Last Seen: 03-25-2017
From: St Paul
iTrader: 4 / 100%
Thanks for the PM. So the test standard in this instance was a Memtest "burning test" and the duration was "overnight". QVL was also mentioned. Now would 2 reported errors in Memtest indicate that the memory was faulty? Of course the answer is no. Let me explain.

Burn-in is designed to reduce the number of early physical failures in the field. QVL is an interoperability testing process. Neither are a DDR validation test per se. By design, Memtest results are not an indicator of a physical memory fault. Memtest is fine as a burn-in tool and can be helpful for QVL, but it does not have the ability to capture functional data at the physical level. As long as the error rate is minimal the results are not significant. Keep in mind that each individual memory module and PCB was previously subjected to signal integrity analysis (at a high sigma) as part of the physical layer validation process. Assembled memory sticks are also subjected to system level behavioral testing as part of the final functional validation process. So during QVL or burn-in you can have a JDEC compliant memory stick that is functionally perfect, but may show an occasional Memtest error as a result of interoperability issues such as transmission errors. It doesn't mean that the BER/SER are an issue.

I mentioned CPU and GPU errors to get you thinking about permissible error rates at a system level. All of the silicon in consumer grade computer products are permitted to have functional errors. That includes your VRM's, sound chip, south bridge, networking chip, CPU, GPU, memory, etc. It is more an issue of tolerance than of perfection.

Last edited by Josie Wales; 05-26-2012 at 03:40 PM.
United States  Offline
    Register to Reply to This Post
Old 05-26-2012, 09:37 PM   #28
larrymoencurly
Learning To Overclock
larrymoencurly's Avatar
Senior Member
 
Posts: 170
Last Seen: 01-24-2014
iTrader: 0 / 0%
Quote:
Originally Posted by Josie Wales View Post
Thanks for the PM. So the test standard in this instance was a Memtest "burning test" and the duration was "overnight". QVL was also mentioned. Now would 2 reported errors in Memtest indicate that the memory was faulty? Of course the answer is no. Let me explain.

Burn-in is designed to reduce the number of early physical failures in the field. QVL is an interoperability testing process. Neither are a DDR validation test per se. By design, Memtest results are not an indicator of a physical memory fault. Memtest is fine as a burn-in tool and can be helpful for QVL, but it does not have the ability to capture functional data at the physical level. As long as the error rate is minimal the results are not significant. Keep in mind that each individual memory module and PCB was previously subjected to signal integrity analysis (at a high sigma) as part of the physical layer validation process. Assembled memory sticks are also subjected to system level behavioral testing as part of the final functional validation process. So during QVL or burn-in you can have a JDEC compliant memory stick that is functionally perfect, but may show an occasional Memtest error as a result of interoperability issues such as transmission errors. It doesn't mean that the BER/SER are an issue.

I mentioned CPU and GPU errors to get you thinking about permissible error rates at a system level. All of the silicon in consumer grade computer products are permitted to have functional errors. That includes your VRM's, sound chip, south bridge, networking chip, CPU, GPU, memory, etc. It is more an issue of tolerance than of perfection.
I thought that for all practical purposes, at least with digital devices, perfection meant the variations didn't exceed design tolerances, provided those tolerances were correct. I know perfection isn't possible, but final testing done with just PCs seems so lax that I wouldn't necessarily trust memory even if it passed with zero errors.

Doesn't interoperability testing require running a memory module in several different motherboard designs? But many companies don't seem to do that, as indicated by their own factory tour videos and roughly 10% of my heatsinked or no-brand chip modules failing my own testing.

Don't most module companies either copy an industry-wide reference design circuit board rather faithfully or get their boards from BrainPower, meaning signal integrity issues should be about the same when all the recommended components are installed? (I haven't noticed any left out.) But I get a lot fewer errors with branded and non-overclocked chips, so that's why I think almost all memory errors are due to the chips, not something outside them.
Offline
    Register to Reply to This Post
Old 05-28-2012, 01:57 AM   #29
Josie Wales
Extreme Overclocker
Josie Wales's Avatar
Senior Member
 
Posts: 2,388
Last Seen: 03-25-2017
From: St Paul
iTrader: 4 / 100%
This is a complex subject so it is critical that you remain laser focused on one or two points if you want a satisfactory answer. I have read your comments surrounding this topic from around the internet and know you are primarily concerned with testing, so let’s start with the basics. In recent years, memory chip manufacturers that sell DDR3 desktop memory sticks tend to have an annualized product failure rate in the range of 1.2% (Samsung, Micron, etc.). Popular brand name memory companies that sell DDR3 desktop memory sticks using OEM parts tend to have an annualized product failure rate in the range of 3.4% (Corsair, GSkill, etc.). How much of this ~2.2% difference can be attributed to testing? Not much.

Testing is done at each step of the memory production process. A short list would include: design testing, materials testing, prototype testing, environmental and longevity testing, manufacturing process testing, wafer testing, manufacturing process continuous improvement testing, die testing, die continuous improvement testing, assembly process testing, assembly process continuous improvement testing, assembled stick functional testing, functional testing and algorithmic design continuous improvement testing, burn-in, burn-in process continuous improvement testing, validation testing, forensic testing, etc.

Looking at that list you will notice that the majority of production testing is done upstream by the chip manufacturers themselves, but the chip manufacturers and the test equipment manufacturers also provide downstream assistance by guiding the assemblers with their functional quality programs. All of the major memory companies also send their products to independent test labs or directly to major motherboard companies for final QVL validation (Intel, Asus, etc.).

With this avalanche of testing you would expect there to be even less of a difference in failure rates between brands of memory. With the broad availability of automated testing equipment name brand memory companies have the ability to test every single stick of memory that comes off the assembly line. Most actually do (even small memory companies like Geil). Surprisingly the failure rates have not declined much since comprehensive automated memory testing has become pervasive. The one area we see a high degree of variation is with burn-in. Burn-in systems which induce environmental, electrical, and functional stress in an effort to “age” memory are becoming more widespread. Some firms still do a short 20 minute burn-in on a motherboard, while others go as long as 24 hours in an automated burn-in station. Automated burn-in stations and long duration burn-in cycles are expensive, but they measurably reduce early life failure. Unfortunately ELF only accounts for a small percentage of total failures on average.

It is fair to estimate that no more than a third of the difference in failure rates can be attributed to variations in final product testing and burn-in. This is why I suggest not getting too hung up on testing. Other factors dominate. Chip selection is certainly important. The difference in product failure rates between memory SKU's can be shocking. But other factors, such as system hardware, also play a significant roll. Of course you should already know this as you cited the Toronto study on HardForum last year.

Last edited by Josie Wales; 06-06-2012 at 07:49 PM.
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 03:12 AM   #30
larrymoencurly
Learning To Overclock
larrymoencurly's Avatar
Senior Member
 
Posts: 170
Last Seen: 01-24-2014
iTrader: 0 / 0%
In my experience, Samsung/Micron/etc. modules (or any modules with brand name chips) have had a much lower failure rate than 1.2%, probably quite a bit under 0.5%, while Corsair/Gskill/etc. with unidentified chips has been way higher than 3.4%, more like 10%. Is it just a coincidence that the average DRAM wafer's yield is around 90%, for products that go out as packaged chips marked with standard part numbers and are guaranteed to meet data sheet specs? I have a feeling companies that buy whole wafers get higher yields due their standards being more lax.
Offline
    Register to Reply to This Post
Old 05-28-2012, 07:23 AM   #31
Mr.Scott
Chief Engineer
Mr.Scott's Avatar
Senior Member
 
Posts: 12,551
Last Seen: Yesterday
Age: 55
From: Spencerport N.Y.
iTrader: 67 / 100%
Quote:
Originally Posted by larrymoencurly View Post
In my experience, Samsung/Micron/etc. modules (or any modules with brand name chips) have had a much lower failure rate than 1.2%, probably quite a bit under 0.5%, while Corsair/Gskill/etc. with unidentified chips has been way higher than 3.4%, more like 10%. Is it just a coincidence that the average DRAM wafer's yield is around 90%, for products that go out as packaged chips marked with standard part numbers and are guaranteed to meet data sheet specs? I have a feeling companies that buy whole wafers get higher yields due their standards being more lax.
Define 'unidentified' chips. Corsair, G.Skill, Etc. all use completely identifiable OEM chips made by Samsung, Micron, Hynix, Elpida, etc. The only difference is each manufacturers particular binning process. No manufacturer is going to accept sub standard OEM chips and put their name on them. That would be business suicide.
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 01:30 PM   #32
Josie Wales
Extreme Overclocker
Josie Wales's Avatar
Senior Member
 
Posts: 2,388
Last Seen: 03-25-2017
From: St Paul
iTrader: 4 / 100%
And I have had a 0% failure rate with Corsair/Gskill/etc. Our personal experience is not specifically relevant to a discussion of industry testing practices and product failure rates. Having said that, I am very concerned about your excessively high failure rates as they may be indicative of poor hardware selection, incorrect system settings, and improper handling on your part. When your personal experience doesn't track with published results or the experience of the enthusiast community as a whole you know you are doing something wrong. As you saw in the Toronto study the majority of memory errors are system specific and continued after installing replacement sticks. What were you doing wrong?
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 01:39 PM   #33
jbmcmillan
OLD FART
Senior Member
 
Posts: 3,877
Last Seen: Today
Age: 63
From: Langley,B.C.
iTrader: 9 / 100%
Quote:
Originally Posted by Josie Wales View Post
And I have had a 0% failure rate with Corsair/Gskill/etc. Our personal experience is not specifically relevant to a discussion of industry testing practices and product failure rates. Having said that, I am very concerned about your excessively high failure rates as they may be indicative of poor hardware selection, incorrect system settings, and improper handling on your part. When your personal experience doesn't track with published results or the experience of the enthusiast community as a whole you know you are doing something wrong. As you saw in the Toronto study the majority of memory errors are system specific and continued after installing replacement sticks. What were you doing wrong?
That's kind of what I was saying but you did it much more eloquently.The fact that you never read about a huge or abnormal failure rate(with exceptions) in the enthusiast community and if it was going to show up it certainly would there.
Canada  Online
    Register to Reply to This Post
Old 05-28-2012, 02:09 PM   #34
Josie Wales
Extreme Overclocker
Josie Wales's Avatar
Senior Member
 
Posts: 2,388
Last Seen: 03-25-2017
From: St Paul
iTrader: 4 / 100%
larrymoencurly has stated at HardForum that his personal failure rate is based on "my own experience with branded vs. unbranded RAM chips, based well over 300 samples bought from probably at least 10 sources"(link) so it isn't merely a couple of bad sticks. He has even reported multiple problems with Kingston, which is the most reliable brand of memory sold. Very suspicious.
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 02:38 PM   #35
Mr.Scott
Chief Engineer
Mr.Scott's Avatar
Senior Member
 
Posts: 12,551
Last Seen: Yesterday
Age: 55
From: Spencerport N.Y.
iTrader: 67 / 100%
I'd say so. The last major brand failure that I can remember was the Ballistix DDR line. Pretty much everybody that bought it was affected. Very short lifespan, if you added voltage even shorter. By short I mean 6 months or less before failure. Other than that, failures have been onesy twosy with all manufacturers lately.

Last edited by Mr.Scott; 05-28-2012 at 02:43 PM.
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 05:01 PM   #36
jbmcmillan
OLD FART
Senior Member
 
Posts: 3,877
Last Seen: Today
Age: 63
From: Langley,B.C.
iTrader: 9 / 100%
The Ballistix must mean Crucial sticks it's crappier ram under a heat spreader.
Canada  Online
    Register to Reply to This Post
Old 05-28-2012, 05:03 PM   #37
Mr.Scott
Chief Engineer
Mr.Scott's Avatar
Senior Member
 
Posts: 12,551
Last Seen: Yesterday
Age: 55
From: Spencerport N.Y.
iTrader: 67 / 100%
Quote:
Originally Posted by jbmcmillan View Post
The Ballistix must mean Crucial sticks it's crappier ram under a heat spreader.
I see what you did there.
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 05:11 PM   #38
Josie Wales
Extreme Overclocker
Josie Wales's Avatar
Senior Member
 
Posts: 2,388
Last Seen: 03-25-2017
From: St Paul
iTrader: 4 / 100%
The Ballistix was so fun to OC that you guys bought it knowing it had a short service life - then RMA'd the hell out of it. LOL. Don't forget the Corsair Donimators which had double digit failure rates with both DDR2 and DDR3 kits. The DDR2 was recalled if I remember correctly. Corsair XMS has had some models in the 5% range. OCZ had a few DDR2 Gold and DDR2 Value kits with elevated failure rates as well. GSkill has had a few models at around double the average failure rate. In all of our examples only a few specific models from each brand had problems and the sticks used clearly labeled JEDEC compliant IC's from top memory manufacturers. With a little research all of them were easy to avoid.
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 05:16 PM   #39
Mr.Scott
Chief Engineer
Mr.Scott's Avatar
Senior Member
 
Posts: 12,551
Last Seen: Yesterday
Age: 55
From: Spencerport N.Y.
iTrader: 67 / 100%
Quote:
The Ballistix was so fun to OC that you guys bought it knowing it had a short service life - then RMA'd the hell out of it. LOL.
Truth.
United States  Offline
    Register to Reply to This Post
Old 05-28-2012, 05:50 PM   #40
jbmcmillan
OLD FART
Senior Member
 
Posts: 3,877
Last Seen: Today
Age: 63
From: Langley,B.C.
iTrader: 9 / 100%
Just wondering what caused the few product lines that had problems and please don't tell me it was lax testing.
Canada  Online
    Register to Reply to This Post
Sponsored Links:
Please Register to Post a Reply

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT -6. The time now is 01:04 PM.

Copyright ©2000 - 2016, Jelsoft Enterprises Ltd.
Powered by vBulletin
Copyright ©2000 - 2016, EXTREME Overclocking