- Joined
- Jan 14, 2011
Dude if you got the spare smack go 2X8GB (2x4) 1N/1T It improves responsiveness a tint bit!
From your sig you are running 4 sticks everywhere.
careful they might go out of spec and throw errors, dat 1t!
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Dude if you got the spare smack go 2X8GB (2x4) 1N/1T It improves responsiveness a tint bit!
From your sig you are running 4 sticks everywhere.
careful they might go out of spec and throw errors, dat 1t!
If I recall correctly the samsung LP LV sticks now aren't OCing as far as they did at launch. Could be wrong there.
My apologies for that extended spate of off-topic drivel; the problem has been rectified. We can get back to our regularly scheduled programming now.
I'm actually making a claim in favor of generally accepted knowledge, but it's the kind generally accepted by engineers but not necessarily by "enthusiasts". The data y'all presented haven't been vetted for accuracy or validity, either, and I've seen similar data for hard disks from various sources, and those data were all over the place in regards to ranking brands by reliability.You are making a claim against generally accepted knowledge without formal data. On the other hand, we are backing up generally accepted knowledge and do have formal data (the failure rates have been quoted earlier in this thread). That is the main issue here, and frankly I believe this ends the debate.
For #2, again, formal data to back up the claim, or this thread is pointless.
I don't know your definition of "high performance", but here's a Micron 2133 MHz CLK 14 unbuffered module, probably made with chips rated faster than almost all retail 2133 MHz and even 2400 MHz. Apparently no heatsinks:There're a ton of low performance RAM modules with heatsinks too. I don't see a heatsink conspiracy here personally.
You don't see high performance RAM without heatsinks often (ever?). Would you care to point some out?
But RMAs can take time, and memory defects aren't always obvious and cause less experienced users to pay for professional misdiagnosis.Well, those are some moderately big numbers but I still think you can rma if you are an end user, it's not the end of the world.
Now, for an enterprise that might be a bigger concern...
There's no way to tell without knowing the comparative costs of all those RMAs costs vs. cheaper chips hiding behindDo you really think that big corporations would sell so many memory chips barely stable giving them lifetime or 5 year+ warranty ? RMA costs are higher than module production costs.
Odd, but a few weeks ago, when I mentioned one module company claiming they didn't use chips faster than 1600 MHz because they couldn't buy them in the market, somebody said I was wrong and pointed out 1866 MHz chips from a dealer listed on Alibaba. I don't think you disagreed with him then, but now are you you saying 1866 MHz chips aren't in regular production (they are, but I couldn't find 2133 MHz)?Btw Micron chips linked by you are not in mass production. Check their database. Most 1866 and 2133 chips are only as engineering samples ( not only from Micron ). Thats why memory producers are binning lower series.
http://www.micron.com/parts/modules/ddr3-sdram/mt9jsf25672az-2g1
What does "effective" performance mean?I'm not really sure what you mean by faster. Higher clock, tighter timings ? Effective performance is somewhere in the middle.
No, of course not. Chips (or modules) are defective when they don't meet claimed specs.You don't even know how are these chips working on available desktop boards. Think that many boards won't even boot at CL13 and CL12 isn't even on compatibility list for most 2133 chips ... what would you say then, that all 2133 CL14 chips are defective ?
Judging by the amount of XMP "guaranteed" modules not working right, I don't trust that guarantee.Desktop memory = memory built within' JEDEC specification on IC compatible with these settings. XMP profile is guaranteed profile based on settings tested by producer who is giving warranty for that.
That's why the example modules I pointed to were all unbuffered, non-ECC.You could argue with all forum members if it was server memory but if we are talking about desktops.
It doesn't seem to be years for real-life errors but more like months, at most.Memory that generates single errors in memtest can be still stable for years in system as there is something like error correction on software level. Blue screens = in most cases system protection from data loss that is happening when you already have lot of errors and system can't correct them.
Those 4GB Samsungs sold by NewEgg and Amazon sure seem to be in retail packaging, as were the Crucial DDR2s made by Samsung (whole module, not just chips).Memory without heatsinks are now mainly kits dedicated for OEM market and guess what, that Samsung 1600 CL11 memory is dedicated for OEM market.
You can't even buy most of them in retail stores or if you can then most were not planned to be there. Even producers like Kingston are selling now higher % of their blu series than value as price is almost the same.
Were they minis? I've never actually built a computer myself; the closest I came to that was wire-wrapping an 8-bit system, and the video never quite worked right.I'm working in IT for over 10 years and for about 6 years I was working for companies that were making 1000-1500 computers/month. Memory fail rate was at about 1% and guess what... almost all defective memory was actually that OEM/no heatsink series.
There is something like chips quality but also PCB quality, tests and shipping/package that takes part in overall memory quality.
I'm not very familiar with high performance graphics cards, but how often do those cards give bad pictures due to hardware defects? Also I've seen no effort to hide video RAM chips behind unnecessary heatsinks, and even the slowest RAM chips are brand name.Now please start next topic about all other memory as GDDR are also overclocked so most graphics cards can be defective...
I'm not reading all posts and I didn't say that there are no 1866/2133 chips at all but most are only engineering samples so availability of all other 1866/2133 can be limited. For example Micron has 1866 chips that are probably also used in Crucial Ballistix 1600 LP series mixed with 1600 chips.Odd, but a few weeks ago, when I mentioned one module company claiming they didn't use chips faster than 1600 MHz because they couldn't buy them in the market, somebody said I was wrong and pointed out 1866 MHz chips from a dealer listed on Alibaba. I don't think you disagreed with him then, but now are you you saying 1866 MHz chips aren't in regular production (they are, but I couldn't find 2133 MHz)?
It means that higher clock doesn't mean that memory is faster as there is lot of timings that can change real memory bandwidth. In most cases 1600 11-11-11 memory can be as fast as 2133 14-14-14.What does "effective" performance mean?
Then I see no point of your discussion in the first place as 99%+ memory is working within producers specification and you get warranty from memory module not chip producer.No, of course not. Chips (or modules) are defective when they don't meet claimed specs.
You have no official data so please stop with that. Most XMP is not working because board manufacturers can't make good BIOS. Look at ASUS and their MemOK buttons. For years they had lot of problems to stabilize memory that was working without issues on other boards. That's just an example.Judging by the amount of XMP "guaranteed" modules not working right, I don't trust that guarantee.
Noone cares about unbuffered RAM as much as you. As I said it's non mission critical hardware so you can live with single errors and costs of any failure are not high.That's why the example modules I pointed to were all unbuffered, non-ECC.
It doesn't seem to be years for real-life errors but more like months, at most.
These sticks are dedicated for OEM market and available in retail package only in the USA in some bigger online stores. Were probably made as upgrade option for Samsung computers.Those 4GB Samsungs sold by NewEgg and Amazon sure seem to be in retail packaging, as were the Crucial DDR2s made by Samsung (whole module, not just chips).
You know that covering chips has also reason ? Memory module producers are selling their product and don't have to show you what is inside. They are declaring that product is working without problems and nothing else should be important for regular consumer.I don't mind heatsinks in themselves, just heatsinks covering up the fact the chips are no-name (which are often also sold without heatsinks) or overclocked and insufficiently screened for quality.
I think that you know what I meant ... or I didn't understand you right and your knowledge base only on some partial theory ? ... then sorry but what are you talking about in 7 pages thread ?Were they minis? I've never actually built a computer myself; the closest I came to that was wire-wrapping an 8-bit system, and the video never quite worked right.
I'm not very familiar with high performance graphics cards, but how often do those cards give bad pictures due to hardware defects? Also I've seen no effort to hide video RAM chips behind unnecessary heatsinks, and even the slowest RAM chips are brand name.
I'm not reading all posts and I didn't say that there are no 1866/2133 chips at all but most are only engineering samples so availability of all other 1866/2133 can be limited. For example Micron has 1866 chips that are probably also used in Crucial Ballistix 1600 LP series mixed with 1600 chips.
It means that higher clock doesn't mean that memory is faster as there is lot of timings that can change real memory bandwidth. In most cases 1600 11-11-11 memory can be as fast as 2133 14-14-14.
Then I see no point of your discussion in the first place as 99%+ memory is working within producers specification and you get warranty from memory module not chip producer.
You have no official data so please stop with that. Most XMP is not working because board manufacturers can't make good BIOS. Look at ASUS and their MemOK buttons. For years they had lot of problems to stabilize memory that was working without issues on other boards. That's just an example.
Latest boards have less problems with memory.
There are single series made with bad XMP. Even these are usually corrected in next revision. When I got Geil 2133 memory that couldn't work at rated 1.50V and it was new product then producer fixed SPD/XMP in about 2 weeks. Friend bought the same memory with PN 20 numbers higher and it was already rated as 1.65V.
If memory isn't working within' declared specification then you make RMA or just return product and that's all. For that you get warranty after all.
Other thing is that some chipsets are not working good with some IC. Noone will guarantee you that every memory will work with XMP settings on every board but it should on standard JEDEC profiles and it's almost always working ( at least I haven't seen DDR3 that couldn't boot at 1066 or 1333 and standard timings ).
Noone cares about unbuffered RAM as much as you. As I said it's non mission critical hardware so you can live with single errors and costs of any failure are not high.
On the other hand if you care about your data then you make backup. It's normal in these days. Every hardware can fail, it's just electronics.
Also you have no idea how high % of data loss is caused by memory and since there is 1% of RMA for unbuffered RAM then count that maybe 90% of that 1% never caused any data loss.
So you confuse all these forum members for ~0.1% cases ?
Much higher data loss % is caused by HDD/SSD.
You need really damaged memory to actually lose data on current PCs. As I already said, system is protecting most of the data and will freeze or show blue screen before anything serious happens. It's not always working but there is such feature.
If memory has serious problems then it won't even enter system.
These sticks are dedicated for OEM market and available in retail package only in the USA in some bigger online stores. Were probably made as upgrade option for Samsung computers.
You know that covering chips has also reason ? Memory module producers are selling their product and don't have to show you what is inside. They are declaring that product is working without problems and nothing else should be important for regular consumer.
Also their tests are usually better especially now when high end memory is additionally "hand tested" on actual motherboards, not only memory tester and you get compatibility list for each memory or can ask support for it.
I think you don't remember when some years ago OEM memory fail rate was much higher than now and all were trying to buy memory with heatsinks just because of much higher quality.
Probably the only brand popular for years without heatsinks is Kingston with their Value series but even they are rebranding chips as lot of them are only with Kingston sign and PN. The same is making Corsair in their Value series or most other producers.
It's almost impossible to get DDR3 memory from actual chip producer that won't be dedicated for OEM market just because these big corporations don't care to sell retail products. They have agreements to sell chips to memory module manufacturers.
I think that you know what I meant ... or I didn't understand you right and your knowledge base only on some partial theory ? ... then sorry but what are you talking about in 7 pages thread ?
Point is that memory fail rate in graphics is as high as in DDR memory modules and in the same way gfx cards producers are "overclocking" these chips. And guess what ... noone cares.
As I already said 99% users don't care what brand is IC as long as it's working ... and it's working in 99% as it was already said couple of times in this thread. That 1% of other users are overclockers and computer enthusiasts who in big part don't care if memory producer overclocked these chips or not. What counts is what can you do on these modules. Thats all point of high end memory now as even cheapest sticks are fast enough for almost everything.
If you want 2133+ memory then you can get couple of random, standard rated sticks and count that you make it work at higher clock or you can buy 1 kit already tested by memory producer who is giving you warranty that it will work at rated settings. If not then you can return this product or exchange it.
You probably also didn't think about some other things. Most motherboard producers won't guarantee that their boards will work at higher clock than 1866 and actual chance for higher clocks in big part depends from memory controller, not only memory. Memory errors on higher clocks are caused more often by IMC than memory.
Also many boards can't set right voltages from even best programmed SPD/XMP so settings are correct but sometimes you have to change it manually in BIOS.
You don't have to add your comments in every thread that memory producers are cheating and higher clocked memory will be probably faulty. No one really needs that theory.
It would be nice to see something more constructive from you than all the time the same conspiracy statements without any proof.
If I recall correctly the samsung LP LV sticks....blah blah blah advice
As I said, it doesn't seem to be 99%+ .Then I see no point of your discussion in the first place as 99%+ memory is working within producers specification and you get warranty from memory module not chip producer.
You have no official data, either, but I'm not telling you to stop posting your unsubstantiated claims (unsubstantiated by the standards you expect from me), either. I'll bet when XMP doesn't work, the prime reason is because the memory wasn't tested very thoroughly, but the only way we'll ever know is if somebody sells 2133 MHz XMP modules with chip factory certified 2133 MHz chips.You have no official data so please stop with that. Most XMP is not working because board manufacturers can't make good BIOS. Look at ASUS and their MemOK buttons. For years they had lot of problems to stabilize memory that was working without issues on other boards. That's just an example.
But I thought module companies tested just as strictly as chip companies did, or the difference between $2,000 and $4,000,000 testing machines wasn't enough to matter.There are single series made with bad XMP. Even these are usually corrected in next revision. When I got Geil 2133 memory that couldn't work at rated 1.50V and it was new product then producer fixed SPD/XMP in about 2 weeks. Friend bought the same memory with PN 20 numbers higher and it was already rated as 1.65V.
I'd rather have a very good product than a so-so product with a great warranty.If memory isn't working within' declared specification then you make RMA or just return product and that's all. For that you get warranty after all.
Again, that says something about factory overclocking.Other thing is that some chipsets are not working good with some IC. No one will guarantee you that every memory will work with XMP settings on every board but it should on standard JEDEC profiles and it's almost always working ( at least I haven't seen DDR3 that couldn't boot at 1066 or 1333 and standard timings ).
Woomack said:You could argue with all forum members if it was server memory but if we are talking about desktops.
larrymoencurly said:That's why the example modules I pointed to were all unbuffered, non-ECC.
I don't like being annoyed by defective products, even if I can replace them free, especially when much more reliable substitutes are available for the same price. Again, I cited unbuffered modules because that's by far the most common and is what almost all memory discussions here have been about.No one cares about unbuffered RAM as much as you. As I said it's non mission critical hardware so you can live with single errors and costs of any failure are not high.
Which is why I run RAID and make frequent backups. OTOH the 1% RMA rate you cite doesn't seem to be trustworthy -- just like all the reliability rates mentioned by everybody in this thread.Also you have no idea how high % of data loss is caused by memory and since there is 1% of RMA for unbuffered RAM then count that maybe 90% of that 1% never caused any data loss.
So you confuse all these forum members for ~0.1% cases ?
Much higher data loss % is caused by HDD/SSD.
So I shouldn't worry if the system boots and rarely crashes?You need really damaged memory to actually lose data on current PCs. As I already said, system is protecting most of the data and will freeze or show blue screen before anything serious happens. It's not always working but there is such feature.
If memory has serious problems then it won't even enter system.
Woomack said:Memory without heatsinks are now mainly kits dedicated for OEM market and guess what, that Samsung 1600 CL11 memory is dedicated for OEM market.
You can't even buy most of them in retail stores or if you can then most were not planned to be there. Even producers like Kingston are selling now higher % of their blu series than value as price is almost the same.
larrymoencurly said:Those 4GB Samsungs sold by NewEgg and Amazon sure seem to be in retail packaging, as were the Crucial DDR2s made by Samsung (whole module, not just chips).
I saw them at a British website, with about the same retail packaging. Britain is not in the USA.These sticks are dedicated for OEM market and available in retail package only in the USA in some bigger online stores. Were probably made as upgrade option for Samsung computers.
larrymoencurly said:I don't mind heatsinks in themselves, just heatsinks covering up the fact the chips are no-name (which are often also sold without heatsinks) or overclocked and insufficiently screened for quality.
And RAM heatsinks are primarily for cooling purposes.You know that covering chips has also reason ? Memory module producers are selling their product and don't have to show you what is inside. They are declaring that product is working without problems and nothing else should be important for regular consumer.
I don't want memory that's screened like that with PCs because so much of it has failed. I want professional quality testing instead.Also their tests are usually better especially now when high end memory is additionally "hand tested" on actual motherboards, not only memory tester and you get compatibility list for each memory or can ask support for it.
I think you don't remember when some years ago OEM memory fail rate was much higher than now and all were trying to buy memory with heatsinks just because of much higher quality.
Probably the only brand popular for years without heatsinks is Kingston with their Value series but even they are rebranding chips as lot of them are only with Kingston sign and PN. The same is making Corsair in their Value series or most other producers.
Samsung (even if sales are restricted to just that tiny nation known as the USA). Crucial/Micron.It's almost impossible to get DDR3 memory from actual chip producer that won't be dedicated for OEM market just because these big corporations don't care to sell retail products.
Woomack said:I'm working in IT for over 10 years and for about 6 years I was working for companies that were making 1000-1500 computers/month. Memory fail rate was at about 1% and guess what... almost all defective memory was actually that OEM/no heatsink series.
There is something like chips quality but also PCB quality, tests and shipping/package that takes part in overall memory quality.
larrymoencurly said:Were they minis? I've never actually built a computer myself; the closest I came to that was wire-wrapping an 8-bit system, and the video never quite worked right.
Sorry, but I don't know what you mean. I assumed you were involved with minicomputers because that's real computer manufacturing, not merely assembling PCs from finished components.I think that you know what I meant ... or I didn't understand you right and your knowledge base only on some partial theory ? ... then sorry but what are you talking about in 7 pages thread ?
larrymoencurly said:I'm not very familiar with high performance graphics cards, but how often do those cards give bad pictures due to hardware defects? Also I've seen no effort to hide video RAM chips behind unnecessary heatsinks, and even the slowest RAM chips are brand name.
I don't now if the first sentence is true because I've never seen failure rates for such RAM. Also aren't higher defect rates acceptable, unless the graphics card is being used by China for simulating nuclear weapons? Micron once said that's what they did with some of their defective DRAM chips.Point is that memory fail rate in graphics is as high as in DDR memory modules and in the same way gfx cards producers are "overclocking" these chips. And guess what ... no one cares.
And that's been my contention -- it doesn't work 99% of the time but more like 90%, even without overclocking.As I already said 99% users don't care what brand is IC as long as it's working ... and it's working in 99% as it was already said couple of times in this thread. That 1% of other users are overclockers and computer enthusiasts who in big part don't care if memory producer overclocked these chips or not.
I don't want to, because I'm not building a super computer or trying to overclock a PC just for the best benchmark results, and if I wanted maximum performance I'd concentrate on the CPU and graphics instead.If you want 2133+ memory then you can get couple of random, standard rated sticks and count that you make it work at higher clock or you can buy 1 kit already tested by memory producer who is giving you warranty that it will work at rated settings. If not then you can return this product or exchange it.
But with mobos, virtually everything available is crap quality.You probably also didn't think about some other things. Most motherboard producers won't guarantee that their boards will work at higher clock than 1866 and actual chance for higher clocks in big part depends from memory controller, not only memory. Memory errors on higher clocks are caused more often by IMC than memory.
Also many boards can't set right voltages from even best programmed SPD/XMP so settings are correct but sometimes you have to change it manually in BIOS.
I don't. OTOH it wouldn't hurt if people quit buying so much junk instead of quality.You don't have to add your comments in every thread that memory producers are cheating and higher clocked memory will be probably faulty. No one really needs that theory.