Jump to content
Corsair Community

250% testing isn't good enough!


theosib

Recommended Posts

I recently had a Corsair Value Select memory stick go bad, so I exchanged it through newegg for some XMS memory. TWINX1024-2700LL, to be specific, with the 2-3-2-5-1T certification, etc. I set up all of the memory parameters in the BIOS according to specs. Because of the options of the BIOS, I was able to set it to 2-3-2-6-1T timings. Four-bank interleave. Anything I understood, I set right. There were three parameters (tWTR, tWR, and DRAM Access) which I could not find any specs on for Corsair memories, so I set them to the most conservative settings. Then I ran memtest86 for 24 hours straight. No errors. Wonderful! I also ran SiSoft SANDRA burn-in for an extended period. No problems! And then I tried to install Gentoo Linux, and all heck broke loose. As you may know you install Gentoo by compiling everything. It's not for the weak of stomach. Cutting to the point, the compiler, GCC, would crash left and right. Most importantly, when GCC was used to build GCC and glibc (shared libraries used by all programs), the system became very unstable, because those core things had been compiled incorrectly, and that made all sorts of things work incorrectly. I was very frustrated by what I perceived as being glaring bugs in Gentoo. Other Gentoo users insisted that I had to be overclocking or running memory out of spec, but according to what meager information I could get from Corsair's web site, I was definately doing everything right. I was stubbornly insistent that the hardware was NOT the problem. I trusted Corsair's numbers! Still, being willing to persue any avenue to discover the cause of my problems, I began systematically testing each of my memory settings. By that process, I did discover the culprit: Command Rate. As soon as I set command rate to 2T, all of my problems went away. That's very unfortunate, because I know that Command Rate can have a 1%-2% effect on performance; I believe it's second only to bank interleave. Having a command rate of 1T was one of the reasons I paid extra for Corsair LL memories in the first place! This isn't the first time I've encountered problems with Command Rate with Corsair Memories. I've never had this problem with Crucial -- only Corsair. A colleague of mine got some 3200C2 XMS memories recently, and we discovered that using a command rate of 1T made his system extremely unstable. I realise that milage varies. I also realize that Corsair probably doesn't test their memories with ABIT motherboards (KD7, in my case, KT400 chipset). But Corsair is so [B]insistent[/B] on their extreme testing methods that one feels compelled to believe they have tested everything far beyond what any mere mortal would do. Corsair's marketing is all about their high-performing, thoroughly-tested memories. Forgive me if I sound bitter, but I've spent two weeks annoying other people because I believed Corsair's marketing. If it was just me, it would be one thing, but I'm guilty of blaming the wrong people, and I feel that Corsair should share in some of that guilt. I'm sure you can understand. Indeed, I would like to get some advise from others in this forum, but I haven't been successful in the past. I'm usually much nicer than this. It turns out Command Rate is a common problem. This I have learned from recent research into what others on the internet report as causes of instability and memory errors. Corsair publishes their memories as being reliable at a Command Rate of 1T, and this is misleading, because, of the specs that they do publish, Command Rate is the most likely to cause a problem. First of all, I think Corsair needs to improve their honesty and/or disclosure. If you're going to advertize that memories run with a Command Rate of 1T, fine, but clearly mention that it is the #1 cause of instability and that it may not be obvious immediately that this is causing your problems. If Corsair's own tests do not reveal the problem, admit that this is a common problem elsewhere. Corsair memories aren't as infallible as they imply, and Corsair needs to be more forthcoming about that. Second, Corsair needs to extend their testing methods. Doing heavy compiling seems to be very stressful on memory, demonstrating problems that most other burn-in tests completely miss. It would only be consistent with Corsair's claims for them to add a test this stressful to their testing repertoire. Third, there are numerous other memory settings in the BIOS that Corsair completely ignores. In my case, they are tWTR, tWR, and DRAM Access. My research suggests that they don't have much effect. But still, it would be nice to know what the best settings are for the RAMs that Corsair sells. And finally, there's Corsair's lack of spec sheets on the RAM chips they use. Whenever I buy memories from Crucial, I can easily determine which chips are on the stick, download the Micron PDF, and find out ALL of the memory timing information. Things like tWTR are right there for me to read. This is not the case for Crucial, and my experiences detailed here reinforce my distrust of companies which do not publish detailed specs on their memories. (My logic here isn't perfect, I admit -- I [I]think[/I] Command Rate isn't a memory timing but rather something affected by northbridge and motherboard characteristics.) I don't know if I would void the warranty for taking off the heat spreader, but even if I did take it off, it's unlikely that I would be able to get a spec sheet. The Value Select memories I had previously had SpecTek (never heard of them) RAMs on them, and it took me DAYS of calling around to find out what they were, so I could get a spec sheet. I had to call SpecTek directly, because people at Corsair didn't have a clue, and even SpecTek didn't have spec sheet -- they had to tell me the part number of the Micron silicon they'd packaged. *sigh* Could someone please help me find a spec sheet for TWINX1024-2700LL memory chips? Please, if my attitude towards Corsair is wrong, tell me why. Please, if I am doing something wrong, correct me. Please, help me to feel that the next time I want to buy or recommend RAMs, Corsair is what I'll choose. Again, I apologize for my tone, but I've just wasted two weeks, because I assumed that I could take Corsair's numbers at face value.
Link to comment
Share on other sites

So much for Corsair's "usually within 24 hours" response time. Is anyone going to give me some help here? I paid a lot of money for the memory I got, and I'm about ready to return it to Newegg and get something else.
Link to comment
Share on other sites

  • Corsair Employees
If you are able to pass Memtest86 for 2 to 3 passes it is not the memory. Once you received the errors did you re-test with Memtest86 ([url]www.memtest.org[/url])? Our memory modules are tested at there rated speed before they leave the factory. Since you are having an application specific problem, please test with a memory test program and not your application. Those memory modules should have no problem with Command Rate of 1T. That has been a standard on our XMS modules since our first 2400C2 modules came out. Please list the CPU that you are using, and since you are setting things up manually, what BIOS settings have you changed? Thanks!
Link to comment
Share on other sites

[quote name='RAM GUY']If you are able to pass Memtest86 for 2 to 3 passes it is not the memory. Once you received the errors did you re-test with Memtest86 ([url]www.memtest.org[/url])? Our memory modules are tested at there rated speed before they leave the factory. Since you are having an application specific problem, please test with a memory test program and not your application. Those memory modules should have no problem with Command Rate of 1T. That has been a standard on our XMS modules since our first 2400C2 modules came out. Please list the CPU that you are using, and since you are setting things up manually, what BIOS settings have you changed? Thanks![/QUOTE] Follow the logic here: When the command rate was set to 1T, I was having compiler crashes on a regular basis. When the command rate was set to 2T, the compiler crashes completely cleared up. Therefore, command rate is involved in the problem. Furthermore, since memtest86 ran for 24 hours straight with no errors with command rate set to 1T, we conclude that memtest86 is not aggressive enough of a test. You refer to this problem as being "application specific", which is misleading. An application cannot cause memory errors. It can only suffer from them. Memtest86 tests memory cells for their ability to store data properly, but it does nothing to test the sorts of access patterns that other applications would exhibit. Note: It is clear that this is not a memory cell problem. The memories are LOGICALLY fine. This is a SIGNALING issue. As chip designers say, it's an analog problem. When run at relaxed settings, the memories work fine. The problem is when they are run at the timings that Corsair says they can be run at. They are too aggressive. That's fine, except that I paid extra for low-latency memories. If I'm going to run "safe" settings, I might as well buy Crucials which I am familiar with AND CAN GET A SPEC FOR. As it turns out, switching to a command rate of 2T is probably not completely solving the problem. I was recently doing a compile which failed due to a syntax error. Looking at the source code, I could find no syntax error. Recompiling worked just fine. That means that the data that got into the compiler was corrupted. Given recent events I'm inclined to blame the memory first, especially since the only hardware I that has changed is the memory. I never had this happen when I was using the Value Select memories at their prescribed timings. The CPU is an Athlon XP 2800+. The motherboard is an ABIT KD7 which uses the KT400 chipset. I'm running the DDR2700 memories at 333mhz, which is their correct clock speed, and without any overclocking. The prescribed timings for the TWINX1024-2700LL are 2-3-2-5-1T, but I am running them at 2-3-2-6-2T. Bank interleave is set at 4, burst length is set to 4. There are some other timings, like tWTR, tWR, and DRAM Access, and those are all set to 3, which is the most conservative available. "Enhance memory performance" is set to Enable, because as far as I can determine, all that does is increase clock rate by 1mhz, which is trivial. I've built computers a few times before, and I've done plenty of reading on what the timing parameters mean. Furthermore, for the sake of the computer we are talking about, all settings which are not explicitly indicated by Corsair are set at the most relaxed value. Nevertheless, I am still having problems. This isn't the first time I've encountered Corsair memories not working at a command rate of 1T.
Link to comment
Share on other sites

Having fun on that soap box? There's some pretty bold statements coming from a guy who "on three parameters 'didn't know what they were' so I set them to the most conservitive settings." (BTW, on those 3 settings, YOU CHECK WITH YOUR MOTHERBOARD MANUFACTURER AND DO WHAT THEY RECOMMEND.) I agree with Ramguy...and seeing as you were "unclear" of some memory settings in your BIOS, my money is on user error. Corsair makes great RAM and I'm sure your speech would've been given by now if what you say had any foundation.
Link to comment
Share on other sites

[quote name='kec789']Having fun on that soap box? There's some pretty bold statements coming from a guy who "on three parameters 'didn't know what they were' so I set them to the most conservitive settings." (BTW, on those 3 settings, YOU CHECK WITH YOUR MOTHERBOARD MANUFACTURER AND DO WHAT THEY RECOMMEND.) I agree with Ramguy...and seeing as you were "unclear" of some memory settings in your BIOS, my money is on user error. Corsair makes great RAM and I'm sure your speech would've been given by now if what you say had any foundation.[/QUOTE] And you're making some bold assumptions yourself. I shouldn't say that I don't know what those three parameters are. I know what tWTR and tWR are. tWTR (write-to-read) is the time (specified in clock cycles) that the memory controller should wait when switching from writing to reading. tWR (write recovery) is the time (specified in nanoseconds) the memory controller should wait when switching from writing to doing a precharge. These things are explained in any typical DRAM spec... Micron, Samsung, Hynix, etc. Furthermore, these are properties of the RAMs, not of the motherboard, so the memory supplier should specify them, although Corsair does not. The motherboard manual lists these settings but does not explain them. It does, however, indicate which are the "safe" settings, which is what I set them to. The "DRAM Access" parameter is something that I have never found an explanation for, other than that a higher number is a "safer" setting. When in doubt, the best policy is to set parameters to "safe" settings. So, counter to your assumption, I did read the motherboard manual, and I did do research on top of that to find out what these are. Have you read a DRAM chip spec? I didn't think so. I have. Do you design chips for a living? I didn't think so. I do. Do I sound overly antagonistic? That is a possibility. I paid a premium for low-latency memories -- which do not seem to live up to the marketing hype. Also I can't seem to get basic information on the product I paid a premium for. In any event, how do you chalk up "doing what Corsair specifies" to user error? When I set Command Rate to 2T, it cleared up the errors I was having. Obviously, Command Rate has something to do with the problem. I am MORE THAN WILLING to be proven wrong. I'd rather not have to send the memories back to Newegg for a refund because I can't get support. PLEASE PROVE ME WRONG. If I am wrong, it'll make my life a lot easier, because I'll just fix the problem and move on. And I'll also wholeheartedly apologize. I am not interested in being "right" here. I am interested in a solution. I'm just angry that I'm not getting any help. Is that fair? Or do you think I should just go away and live with expensive memory errors?
Link to comment
Share on other sites

  • Corsair Employees
By application error I mean that it works fine in Memtest86, a Unix based memory test program, in SiSoft Sandra burn-in, a windows based stress test program, but not in compiling your specific Linux distro. That is an application specific error. The app is not causing the error, but it’s the only one having the error. I have no problem replacing your memory, we have a lifetime warranty. You have one system that you are doing this all on, I would highly suggest testing on another system. Feel free to get an RMA on your memory. To get an RMA send us an email with a copy of the [URL=http://www.corsairmemory.com/main/rma_request.xls]FORM[/URL] or you can send all of your info(full name, shipping address, phone # with area code, the module part #, quantity to be replaced, and a copy of the link to this post). Email it to [email]rma@corsairmemory.com[/email]. If after 1 day or 24 hours excluding weekends you do not get the rma please email the same to [email]warranty@corsairmemory.com[/email] and we will help to resolve it. Thank You!
Link to comment
Share on other sites

Let's see....you design chips for a living, LOVE public speaking, wrote the book on ethics, and also know marketing and advertising from front to back. What a busy little bee you are! All taunting aside, all you had to do was simply ask for help. Hopping up on a soapbox and telling a company how to do things, and that they should do this and should do that, and they should "improve their honesty and disclosure"...blah blah blah, only antagonizes people and makes them not want to help you. Especially when such critisism is aimed at a very customer oriented, high quality driven company. Besides, NO company is perfect. Perhaps you simply got a bad set of sticks? Have you even thought of that yet? You're on your first set of the XMS, right? Read through ANY company forum...shit happens...it sucks sometimes, but it happens. But thankfully Corsair stands behind their ram. First make sure it's not software related, as Ramguy suggested. Then make sure it's not a bad pair of sticks. THEN if they aren't doing as they claim, hop on the soapbox! Good luck in finding a solution and I apologize for egging you on a little....but just remember, if you're not happy with the result I'm sure Newegg and/or Corsair will take care of you.
Link to comment
Share on other sites

You're right, kec789, I have been rather unpleasant. And I apologize. However, from what I can gather, the memories that Corsair sells aren't actually designed to run with the parameters they are advertized with. I speculate that, for instance, Corsair will take a chips designed to run at 2.5-3-3-7-2T and test them to find out how reliably they run at more aggressive timings. Those that pass at 2-2-2-5-1T will be sold that way. This is a yield test, and it's not all too uncommon a practice. As reasonable as this may sound, it kinds scares me when doing this with something as critical as memory. What I don't know is how far beyond the advertized specs the chips are actually tested. For instance, consider a chip advertized to run at 333 MHz, 2-2-2-5-1T. Was that chip tested with those settings at only 333 MHz? Or was it tested at, say, 400 MHz? If it's the former, then the chips could be marginal. If it's the latter, then there is some built-in margin for error, and the chips are more likely to be reliable when subjected to access patterns that are not tested for. I suspect that somewhere in there is the reason why Corsair is so evasive when I ask for a spec sheet on the RAM chips. They know I'll find out that the memories are not designed to run at the timings they advertize. Of course, this isn't really a problem if the memories are tested with sufficient margin for error. Marginal memory is what bit me here. The memories I got from Corsair passed every "memory test" I threw at them. But when I started doing something "real world", like compiling, I encountered access patterns which those memory tests did not account for. Through experimentation, I was able to determine that, despite what the memory tests indicated, the memories were still not reliable at their prescribed timings. I have a suspicion that Corsair's memory tests may not be a whole lot more demanding than the memory tests I used, which is why the memories passed for both of us in those circumstances. Memory tests tend to have rather regular access patterns; it's when you hit it with something highly irregular, like compiling which is very memory-intensive and with poor locality of access that we get a surprise. With all due respect for Corsair, I have decided to return the memory sticks to Newegg for a refund, and I have ordered some Kingston PC3000's (370mhz) spec'd at 2-2-2-5, although I will be running them at 333mhz. This way, I can be sure that I get that all-important margin for error. Also, the Kingston PC3000's were cheaper than the Corsair PC2700's, although that may be primarily due to market fluctuations. (I'm not suggesting that Corsair overcharges.) I should note that this is the second set of XMS memories that I have gotten from Corsair. The other was a single 512M stick of PC3200, it was the 3200C2 model. Those memories also would not work properly when Command Rate was set to 1T. I guess that makes Corsair 0 for 2 in my limited experience with them. They may want to see if there is some issue with ABIT motherboards with VIA chipsets, because those may be the cause of the problem. If the MB/Chipset are indeed the cause of the problems, then I should have problems with the Kingston memories as well. If I do, I'll eat my hat and report the results right here, letting everyone know that Corsair can't be to blame... as well as a full report on what my hat tastes like and how my digestive system dealt with it. :)
Link to comment
Share on other sites

  • Corsair Employees
Well let's try and get some things straight, first off you have made some strong accusations and with out merit I might add. Let me put it another way, if what you say were true, even if it was half true, I don't think we would be in business. When in fact we are growing and have been growing steadily for the last 3 years. In fact we have been profitable as a company for more than 8 consecutive quarters. Name another company that has done that consistently through that time period. More importantly, the accusations you have made here tend to suggest that we are personally out to take advantage of our customers, and Sir, I tell you this, I have never worked in a company where everyone and I do mean everyone; from our president to the staff in our shipping Dept. doesn't do everything they can to produce the best product they can and provide the best service possible! What you have said really is a blow at me personally and everyone in the company I represent, and I don't appreciate it. However, this is a free country and you are entitled to your opinion. No matter how ridicules I personally feel someone might be, not necessarily you, but in a general sense. Our memory is not for everyone and you are welcome to send the modules back to your reseller I am sure they will be glad to help you, if their return policy permits it. The information on the type and make of IC is available with a little research, for every module we make, or an easier way would be for you to post that question in this forum along with the specific module part # and revision and I would be happy to tell exactly what IC is used on any module we make. In addition, I would have possibly suggested another module, based on the make and model of MB and the intended use. Had you asked me and gave us some more details prior to your just making some assumptions based on how you have dealt with IC makers modules. That's not to point the blame back at you; however I would have personally had a different approach. After all the forum is for Post and [b]Pre[/b]-Sales support! To answer some of your direct questions: You have stated "for instance, Corsair will take a chips designed to run at 2.5-3-3-7-2T and test them to find out how reliably they run at more aggressive timings." I would suggest that you take a look at [url=http://www.houseofhelp.com/forums/showthread.php?t=8048]XMS Qualification and Testing[/url]. All of our XMS modules are listed along with the MB we use and the bios settings that are used to validate every module. And, in some cases XMS modules are tested 2 or 3 times. As far as I know we are the only module maker that actually uses a MB to test our modules at the rated speed before they leave our factory. Within this post you will also find a link to the data sheet for that respective module. Now here is where I think maybe you have made some miss-conceptions. We are a module maker not an IC maker thus you will not find an IC data sheet, like you have suggested that you can find with IC makers modules, i.e., Micron, Samsung, SpecTek, and so on! In addition you stated "I suspect that somewhere in there is the reason why Corsair is so evasive when I ask for a spec sheet on the RAM chips. They know I'll find out that the memories are not designed to run at the timings they advertise. Of course, this isn't really a problem if the memories are tested with sufficient margin for error." Our modules are tested to the spec that's published under the link I gave you. And as you know if you are indeed a chip or IC designer there is always margin built in to virtually any electronic device. And what you may not know, with memory specifically the IC's all come with a data sheet with the CAS settings that are supported and a speed rating, however a -6 and -5 IC usually come from the same FAB with the deciding factor usually being where they are physically located on the DIE. Honestly what you are doing is very memory intensive and the memory would need to be 110%, however as you also mentioned the MB would also need to be 110%. You had made some remark about all of the settings not being available, and there is a reason for that along with what I have already said. The timings are published and we do suggest that any settings not listed be set to bios DEFAULT or AUTO, as the SPD on every module is set to either the IC makers suggested settings and/or our predefined settings and are tested at these respective settings. I think we do everything that we can to make sure our modules perform at the rated speed, sometimes there will be one that will fail. For whatever reason it gets through the process and just will not run in a given platform, whether there is an apparent reason for it not. Our return rate is less than .02% on average for all of our modules and these modules specifically XMS2700LL and XMS3200C2 have been very successful for us. We do sell a large volume of both. However, the fact that the modules would run [url=www.memtest.org]memtest86[/url] and memory tests within Windows with out problems, but would have problems with a version of Linux (that is untested by the way) suggests some issue with either the configuration or some other un-identified issue. Lastly you stated "If the MB/Chipset are indeed the cause of the problems, then I should have problems with the Kingston memories as well. If I do, I'll eat my hat and report the results right here, letting everyone know that Corsair can't be to blame... as well as a full report on what my hat tastes like and how my digestive system dealt with it. :)" I would have to say I do understand your frustration; but this is not permitted under the forum rules. However, I do wish you the best of luck and honestly hope things work out for you!
Link to comment
Share on other sites

I feel that it's important that I continue on this issue for the sake of your customers. Both in terms of where I have been wrong and where I feel you may be mistaken. Furthermore, I believe there may be room for me to be in error. [QUOTE]Well let's try and get some things straight, first off you have made some strong accusations and with out merit I might add. Let me put it another way, if what you say were true, even if it was half true, I don't think we would be in business....[/QUOTE] You're rather non-specific in this statement here. On your Qualification and Testing page, you admit that you overclock memory IC's. You seem to think that I disagree with this practice. This is not true. There's nothing wrong with speed-grading chips. What I take issue with is that, you don't test memories with a margin for error. In my line of work, we too do some limited amount of speed-grading. In one of my early ASICs, we'd designed it to run at 100mhz. When I say that, it's important to realize that that is 100mhz with [B]worst case[/B] silicon. Well, it turned out that all of the chips we ever got were actually somewhere between "typical" and "best case", so we could reliably overclock them. But here's the thing: We decided that the chip could run at 133mhz, but we didn't do it by cranking it up to 133mhz in an open-air test station, run some tests and say we were done. NO. When you're overclocking a part, you want to run it in worst-case circumstances. There are three variables to consider which are clock speed, temperature, and voltage. We eliminated voltage since we could guarantee it and tested only for temperature and clock speed. In the case of temperature, we put the device in a temperature chamber and heated the ambient temperature so that the transistor junction temperature would be at maximum. We heated to 55C. With a temperature gradient from transistor to package to heat sync to air of 30-35C, we were on or about the temperature limit of 85C. We selected the parts that passed at 133Mhz, and the ones that failed got put onto our 100Mhz boards. Now, the difference in silicon performance between typical and worst case temperature is about 15%, which means that running the chip at a 15% higher clock rate is about the same as running it maximum temperature. We also did this test also because we figured that, being overclocked, the chip would run at higher than typical temperature (even more margin). That's 153Mhz. I think we ended up using something around 151Mhz, because that was in stock, but close enough. Anyhow, we selected the parts that ran at 151Mhz and sold them to work at 133Mhz. Now, THAT is how you speed-grade an IC. Now, we do stuff used in "mission critical" systems, so we have to have significant margin, probably much more than you do, but my point is that when you say that you test your PC2700 at "only" 333mhz when you have also cut down all the other timing numbers as well, what that tells me is that your parts are marginal at best. Unless you have some additional margin for error, you can only prove that your parts work within the bounds of your memory tests. As soon as someone throws something different at it, there's a significant chance that something will break. In the two cases where I have used Corsair XMS memories, they did not live up to your marketing. Let me put it another way: Chips are designed with margin for error. You can buy plenty of chips that are spec'd at 333Mhz, and they have plenty of margin. But that's because the timings are like 3-3-3-7. Now you want to run those same chips at 333Mhz, but you want to run them with timings of 2-2-2-5. Fine. But when you do that, you have to make up for the margin you just threw away by tightening something else like increasing clock rate, increasing temperature, or lowering voltage. Corsair doesn't do that, which is a problem. [QUOTE]What you have said really is a blow at me personally and everyone in the company I represent, and I don't appreciate it.[/QUOTE] Well, I apologize if I've hurt your feelings, but I really don't think it's appropriate to play the emotional angle here. This is business, and in business people do what is profitable. People in business don't typically set out to cheat others, but they seldom see themselves as others do. People at Microsoft don't see the company as a predatory monopolist. I'm sure many are insulted by the accusation. But they are, nevertheless, a predatory monopolist. Besides, I never meant to imply that Corsair was being directly dishonest. I was simply suggesting that they could improve their testing a bit. [QUOTE]The information on the type and make of IC is available with a little research, for every module we make, or an easier way would be for you to post that question in this forum along with the specific module part # and revision and I would be happy to tell exactly what IC is used on any module we make. In addition, I would have possibly suggested another module, based on the make and model of MB and the intended use. Had you asked me and gave us some more details prior to your just making some assumptions based on how you have dealt with IC makers modules. That's not to point the blame back at you; however I would have personally had a different approach.[/QUOTE] You may note that my original post listed exactly what part I purchased. I have "TWINX1024-2700LL". I also mentioned my motherboard (ABIT KD7, KT400 chipset). Since I have harped on this question, and you tell me that all I have to do is ask, which I did, why have you not taken this opportunity to tell me which IC's are on that module? When I ordered the Kingstons, NewEgg sent me the wrong parts. You may still have an opportunity to convince me to get Corsair parts instead. [QUOTE]All of our XMS modules are listed along with the MB we use and the bios settings that are used to validate every module. And, in some cases XMS modules are tested 2 or 3 times.[/QUOTE] When you say this, are you referring to samples from given production runs, or are you referring to individual modules? Unless you're making a HUGE profit and can therefore afford to put every individual module into a PC and test for an extended period, I have difficulty believing you. How do you meet demand? [QUOTE]...to test our modules at the rated speed before they leave our factory.[/QUOTE] This is part of the problem. You say "[B]at the rated speed[/B]". I can make all sorts of guesses about what your testing lab must be like, but I can assure you that testing parts at exactly their rated speed is ABSOLUTELY NO WAY to properly speed-grade parts. [QUOTE]Now here is where I think maybe you have made some miss-conceptions. We are a module maker not an IC maker thus you will not find an IC data sheet, like you have suggested that you can find with IC makers modules, i.e., Micron, Samsung, SpecTek, and so on![/QUOTE] I was never under this impression. When I referred to "Crucial" memory modules and said that I could always look up the "Micron" datasheet, you should have inferred that I understood the difference between memory modules and IC's. However, given that your company USES memory IC's, it stands to reason that your engineers have been diligent in choosing vendors wisely. Part of what an engineer will look at when evaluating a model of IC is the datasheet on that IC. Furthermore, before you apply what appears to be a heat spreader to your memory modules, one can look at the ICs and read off the part number. I would appreciate it if you could tell me what IC is found on "TWINX1024-2700LL" memory modules. [QUOTE]Our modules are tested to the spec that's published under the link I gave you. And as you know if you are indeed a chip or IC designer there is always margin built in to virtually any electronic device. And what you may not know, with memory specifically the IC's all come with a data sheet with the CAS settings that are supported and a speed rating, however a -6 and -5 IC usually come from the same FAB with the deciding factor usually being where they are physically located on the DIE.[/QUOTE] Yes, I am very familiar with all of this. The ICs from the manufacturer are designed with significant margin, and the published specs are conservative, even for the -5 parts. So, when a part publishes, say, a 40ns tRAS, then it probably really only takes 30-35ns, depending on temperature. But your company admits to pushing and even eliminating that margin. So, your Qualification page says that if a module fails at 300mhz, it's tested again at 266. I assume that this is the case for all modules. That makes sense on the surface, but demonstrating that the part works at a given speed using YOUR TESTS does not prove that the part is not just barely on the edge of failing. On the other hand, if you were to test PC2400 at 333mhz and it passed, then I could see you as being justified in selling it as being reliable at 300mhz. I should note here that it might be slightly misleading talk about just "Mhz" here, because a number of the memory timing numbers are measured in microseconds, where we have to program in the number of clock periods that is greater than or equal to the required time. However, it does apply to a reasonable extent; tRAS of 5T at 370Mhz is shorter than tRAS of 5T at 333Mhz, so if the part passes at 370Mhz, it even more likely to be reliable at 333Mhz. In my particular case, the result of the overclocking is not something internal to the RAM ICs. Instead, it's a signal integrity issue stemming from Command Rate being too aggressive. For the most part, your speed-grading has been successful, but had you tested the memories to run at a higher clock rate, you would have been able to better account for signaling issues between the ICs and the Northbridge on the motherboard. My final comment is that any kind of memory error is completely unacceptable. Memory errors cause data corruption and crashes. Even in the best of cases, errors still occur, which is why we have Registered and ECC memories for servers. But in the case of non-registered, non-ecc memories, it's important that nothing be done to make the system more error-prone than necessary. Limited testing at exactly the desired clock rate breaks this rule. When you advertize that your memory modules run at a given speed, you have a responsibility to increase your guarantee by making sure that the parts actually run FASTER than what you advertize.
Link to comment
Share on other sites

QUOTE from theosib: "I feel that it's important that I continue on this issue for the sake of your customers. Both in terms of where I have been wrong and where I feel you may be mistaken. Furthermore, I believe there may be room for me to be in error." "Both in terms of where I have been wrong.....Furthermore, I believe there may be room for me to be in error"... Where exactly did your paper...errr...post relate to what you said above? Seems like you still just want to make Corsair look bad make yourself look right. I can't believe I just read through that entire post. You know what, all I know is I have XMS 2700C2 ram. It runs like a charm at 7-2-2-2 at 400mhz. That's tighter than SPD and faster than advertised. Runs any stress test I throw at it. Seems to me that's kick ass ram. God help the folks at Kingston. I'm sure they will soon be read the riot act as well....
Link to comment
Share on other sites

I have seen some people complain about some pretty retarded problems, but this is just ridiculous. Errors while compiling an OS that is basically always in beta form. Syntax errors that occur in 1t or 2t. Hmmm. That would defiantly make it your problem. As for your other question, about which IC's are on the module, try doing a search, it would save you a lot of typing. You can find the answer [URL=http://www.houseofhelp.com/forums/showthread.php?t=22611&highlight=2700LL+chips]here[/URL]. As for your other problems, you may wish to give a quick read through [URL=http://www.gentoo.org/doc/en/handbook/handbook.xml?part=1]here[/URL]. Otherwise, have fun ranting and raving about a problem no body but you gives a **** about.
Link to comment
Share on other sites

  • Corsair Employees
XMS2700LL would be as follows XMS2700LL Rev 1.1 Winbond -6 Rev B XMS2700LL Rev 1.2 Winbond -6 Rev C Moreover, we have changed the suggested timings with XMS2700LL to Cass 2-3-2-5 simply because some of the MB's just will not run with the Rass to Cass set to "2". In addition, as I remember you had a MB that was not dual channel so it may not be able to run at Cass 2-2-2-5. However, as I have stated many times; if one of our modules will not run at the advertised settings in an approved platform we will be happy to replace them. What is the make and model of MB you are using as well as the speed of the CPU and it's FSB? For the record, some or our modules are tested above the rated spec. all I can officially tell you with in the forum, would be what they are guaranteed to run at.
Link to comment
Share on other sites

I was running my memories with timings 2-3-2-6-1T, because for tRAS, my only options were 6 and 7. When I had the compiler crashes, I found that changing to 2-3-2-6-2T made the problems go away. I [B]think[/B] that Command Rate has to do with signaling, that is to say, it doesn't appear to correspond to an IC timing number, so I can see why having two modules on the same bus would increase capacitive loading and slow down certain signals. As I say, I have full confidence that the IC's themselves are in perfect working condition INTERNALLY. Motherboard: ABIT KD7 (Which has the KT400 chipset) CPU: Athlon 2800+ FSB: 333mhz I'm also running the memories at 333mhz. Thank you for the information on the IC's. Sorry for all the fuss.
Link to comment
Share on other sites

  • Corsair Employees
Well thanks for the apology, but we could have averted all of this if you would have simply asked the question with out all of the hoopla! I would suggest that we test the modules one at a time with [url]www.memtest86.com[/url] Version 3.0, as 1.X will not run with this chipset. I think we will find the modules pass and if so I would try them again, but please make sure they are in slots one and three and set the Dim Voltage to 2.7 Volts. In addition I would set the timings manually to Cass 2-3-2-6 with the Rass to Cass set to "3" and see if that does not solve the problem. If not or if you find one of the modules is indeed failing. Please follow the link in my signature and we will be happy to replace them!
Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...