NF4 Ultra-D> Corsair CMX512-3200XL - The Final Chapter

[Minsc]

Let's get this working. These chips are $200 and rated at 2-2-2-5.

 

It's unstable in several games (BF2, Bloodrayne 2, Silent Hunter III) within 5 minutes. Disabling Bank Interleave seems to be the key, making the system stable regardless of any other settings.

 

I have used RGone's guide to put the most conservative settings I know of, with the exception of selecting the rated 2-2-2-5 settings and Bank Interleave ON, which is still unstable. Even 3-3-3-8 with B.I. ON is unstable. They are here for your viewing:

 

http://knepfler.com/DFI.htm

http://knepfler.com/dfi.xls (use this if you want to send it back to me with suggestions - dfi at knepfler dot com)

 

I have also tried 3-3-3-8, as well as 2-2-2-5 2T. That one seems to work with Bank Interleave On, but the performance penalty is very great.

 

Ultimately I want to find the magic combo that gets me 2-2-2-5 1T Bank Interleave ON, so my Everest write times are above 2000 MB/s.

 

Optimized defaults w/ B.I. OFF = 1554 MB/s

Best I've ever seen with these chips on my system: 2325 MB/s

 

Because I've seen 2325 MB/s I don't consider 1554 MB/s acceptable for RAM that costs $200 for 2x512MB.

[prodigy_]

Hmm, do you really think it has anything to do with price? No offence, but your complains sound like: "WTF, why WD Raptor is $180 if it's only 74GB?" It's just pointless. Buying right HW that suits your needs requires some research. Do not buy it just b/c "Corsair XL is kewl!" or smth like that. You will always get into trouble...

 

---

 

On-topic: Just for comparison - my memory was $80 per stick and it does 2-2-2-5-1T with Bank Interleave.

 

But mine is based on BH-5 chips and yours is on TCCD. And TCCD is no good for tight timings. It's for high freqs, like 300+. Yeah, it IS technically capable of 200 2-2-2-5, but it will NEVER perform as Winbond with these settings.

 

Don't misunderstand me - you've got a great RAM indeed. Max freqs are just insane for TCCD, especially with good PCB, like Corsair's. But if you want to run @200 - exchange your sticks for some UTT-based pair. Twice as cheap (or even more if you'll manage to find right TwinMOS/M.Tec sticks) and ten times less problems with tight timings.

[RGone]

you need to read this very closely and recognize what Revision I say the memory is and that this is a very very poor data base as it is only 2 sticks.

 

Dram Config Down the Page 200/1:1> CPC=AUTO> 2.0> 03> 06> 02> 09> 19> 04> 03> 02> 04> 2560> DBI=Enable>>>

 

 

Auto> 0> Level 7> Level 2> 9.0ns> Normal> 6.0ns> 256cycles> 16x> 4x> Disable 4 Burst.

 

2.8Volts and was 1T in Orange slots. Bios 704-2BT.bin at 200FSB speed and 10 multiplier. Memory is REV 1.2. IF any other revision all bets are off as that is not the 2x512 set I tested. I tested Revision 1.2.

 

One final note is the memory ran much better at 290FSB when treated as if it were truly TCCx memory.

 

RGone...

[WoL-Shiver]

minsc, i promise, to get you a 24 hour occt and super pi stable once i get you my settings out of ym board. psu should arrive at the rma center tomorow. then itll be back within 6 days.

 

that is a promise to my fellow corsair/xl supporter.

 

i cant help you any further, untill i get ym psu back. but see if you can get that revision number for me! please! i can talk with ramguy and see if he can get you a 1.3 revision as a cross ship rma. you receive the new ram then you send him the other, so your not without it.

 

in regards to what rgone said, ive come to understand that minsc has no plans to oc at the moment. if you have a 1.2 revision, consider it the same as 1.3, as its just an upgraded brainbread pcb, which handles and transfers cleaner power.

 

misnc, send me your email or msn by pm.

 

Regards, WoL-Shiver

[Minsc]

Haven't checked the revision yet, but I'm trying your settings. I like living on the bleeding edge. Booting now. CPC is 1T autodetected. Windows seems to be loading OK... I'm on a laptop right now.

 

Booted. Everest Read: 5820. Write: 2026. Not bad, definitely better than any stable BI:Off settings I've used.

 

Starting the BF2 test... while that's running, let's talk. You mention in another thread that you got some XL's just like mine from Corsair to test, and that you were going to post it into the Overclocking forum. You posted a link from another thread but it didn't work, and I don't see it there. All I have to go on is this post.

 

My theory all this time has been that I can't or shouldn't overclock if I can't get the memory to run at the rated timings at 200 FSB, no overclocking whatsoever. But what you say is interesting - I know you have to loosen the timings to overclock, and of course the slight loss in performance from that is gained many times over in the overclocking.

 

So perhaps I should try that next. BF2 is holding up so far. My god - I accidentally left 16x Anistropy and 8xAA ... and it's running perfectly. My god that looks unbelievable.

 

Will let you know if it crashes.

[Minsc]

Oops just crashed. Back to BI:Off, 3-3-3-8 2T. Will check the RAM revision.

 

Edit: Correction: My CMOS Reloaded "stable" profile uses 1T, 2-3-2-5 (detected by Auto) with Bank Interleave Off and is perfectly fine.

[Minsc]

Revision is 1.2. Wol you have a PM and thanks very much for your offer. I'll keep trying the Corsair just to help get it working stable at the rated speeds with this board. I don't mind suffering to help other noobs who don't check this site before buying Corsair and this DFI board. I only did it because I equated low timings with high speed, Corsair had the best timings I could find on the site I was buying from within my price range, and I've never, ever had to do this much work to make memory work on a hundred systems that I have built.

 

Not that I mind - the more work, the more reward.

 

One of the reasons I'm trying to pull all of this off is to learn about memory for work. We have a custom memory benchmark that we use around the office - we compete for bragging rights of who has the fastest WRITE speed on their memory. Our company's software is very dependent on this, hence the benchmark. I want to dominate it.

 

It seems like overclocking will get me a much higher MB/s write throughput than screwing around trying to get these tight timings.

 

RGone, one more thing... I didn't have a refresh period option of 2560, it was some other weird number like 2596. I picked the closest one.

[WoL-Shiver]

minsc, using the 2-2-2-5 1t timings, and bi on. what times do you get errors at in memtest? if any?

 

also since you have 1.2, they are the same as mine. an the practically the same as the 1.3's. thats one hitch off the list.

[RGone]

Oops just crashed. Back to BI:Off, 3-3-3-8 2T. Will check the RAM revision.

Forget the games at first and 'know' a couple of things. If you BSOD windows more than twice the operating system is likely truly history for all practical purposes when trying to PUSH a game. That is why I have images of all my partitions so that I can slide a fresh o/s back on and not be wondering if windoors is "coming apart internally" and causing a crash and I wind up clueless. I start with memtest86 to give me a 'clue' to windoors operation and then when I get to windoors I use OCCT, Prime, 3Dmark01 and SuperPi to guide my stability. After that a game might stand a chance of running.

 

To run right back to some lesser setting and NOT test is useless also. I 'again' use my images to get back to fresh so I up memory voltages. I try different timings and I experiment within a VERY TIGHT CLOSE area using almost the same timings I showed you. FOR any memory I run > I only change the 4 major settings listed mem makers and the Dram Response time in newer bioses and Dimm and Dram strengths.

 

Also many games are crash right now due driver problems and loosening the memory JUST MASKS that failure of driver.

 

Now you want to know how much faith I put in Memtest86 for testing memory for windows? NONE. WHY? I have two (2) sticks of memory that will pass memtest at 310x9 for hours and hours and hours. B utt will not boot windows Prime Stable at over 250FSB. So much for memtest86 and the real world of windoors. SUCH only happens probably one in 500 times or so that I know of b utt when it happens you have to be able to know that memtest86 has been lying to you as far as windoors goes.

 

Yeah, you send me copy of that mem speed tester you guys use in house and I will show you some big number on that puppy for sure. Hehehehehe.

 

RGone...

 

Oops just crashed. Back to BI:Off, 3-3-3-8 2T. Will check the RAM revision.

 

Edit: Correction: My CMOS Reloaded "stable" profile uses 1T, 2-3-2-5 (detected by Auto) with Bank Interleave Off and is perfectly fine.

[PoppyMcShotgun]

RGone, one more thing... I didn't have a refresh period option of 2560, it was some other weird number like 2596. I picked the closest one.

 

Yeah you do, scroll down... These numbers are not in numberical order... It tripped me out the first time I tried to change this value as well... There are TONS of options for this setting.

[Minsc]

Forget the games at first and 'know' a couple of things. If you BSOD windows more than twice the operating system is likely truly history for all practical purposes when trying to PUSH a game.

 

Fortunately this is not the case for me. I've BSOD'd dozens of times testing the memory and it hasn't caused any problems, though once it did corrupt a critical file, but I know enough about Windows to get it back up and running without imaging. I can't Ghost the drive due to the RAID setup, but I do have backups of everything on another drive that is never accessed via Windows other than showing up under My Computer, so everything is safe there.

 

That is why I have images of all my partitions so that I can slide a fresh o/s back on and not be wondering if windoors is "coming apart internally" and causing a crash and I wind up clueless. I start with memtest86 to give me a 'clue' to windoors operation and then when I get to windoors I use OCCT, Prime, 3Dmark01 and SuperPi to guide my stability. After that a game might stand a chance of running.

 

I know my Windows isn't coming apart when I revert to the slower settings, and I'm able to play BF2 for four solid hours, and the aggressive settings crash it, and several other games, within the first 5 minutes.

 

To run right back to some lesser setting and NOT test is useless also. I 'again' use my images to get back to fresh so I up memory voltages. I try different timings and I experiment within a VERY TIGHT CLOSE area using almost the same timings I showed you. FOR any memory I run > I only change the 4 major settings listed mem makers and the Dram Response time in newer bioses and Dimm and Dram strengths.

 

Initially I also tried many individual settings. That's how I've narrowed it down so closely - it is clear that the system is clearly highly unstable, in Windows games, if Bank Interleave and 1T is on. Disable either of these and it's stable for hours, though I haven't extensively tested BI:On w/ 2T because the performance was so bad I didn't care to know. To get what I paid for in this RAM requires 1T and BI:On. If I can't get it, there's no point in trying to find the next closest stable setting, because they're all crap in the performance department.

 

However, I should narrow this down so that we can determine if it's a single setting that is causing the instability. I'll bet my bottom dollar that 2T with BI:On will still be unstable, albeit stable a little longer than usual. I'll bet everything that it's the BI *alone* that my memory just won't handle.

 

So far, all of the settings I've tried have involved scaling back many, many other settings, all to get BI stable. They haven't worked, so unless we find some simple, single change that does the trick, it's likely that getting BI stable will involve so much of a penalty to the other settings that I'd might as well just turn BI off and go with that. When I've tried the highly conservative settings you suggested in another thread, combined with other people's suggestions, my MB/s write speed drops down to the same rate as BI Off.

 

I do run memtest, my system has racked up hundreds of hours doing it. I leave it out now because I have a good idea of what will happen in it now - with BI:On and 1T, it will crash within 4-20 hours. BI:Off runs for over 24 hours. There are so many combinations to try that it's unrealistic to run memtest for 24 hours after each tweak - especially when I've already got a reliable test - three games that all crash within 5 minutes, versus running for hours stable.

 

Also many games are crash right now due driver problems and loosening the memory JUST MASKS that failure of driver.

 

Maybe so, but if that's the case, it's still a mobo or RAM issue.

 

Now you want to know how much faith I put in Memtest86 for testing memory for windows? NONE. WHY? I have two (2) sticks of memory that will pass memtest at 310x9 for hours and hours and hours. B utt will not boot windows Prime Stable at over 250FSB. So much for memtest86 and the real world of windoors. SUCH only happens probably one in 500 times or so that I know of b utt when it happens you have to be able to know that memtest86 has been lying to you as far as windoors goes.

 

I'm getting conflicting messages here. On one hand, I should memtest after every setting to get an idea, yet it is not representative of Windows performance. So what's the point of it? Even so, I've already done enough memtest settings to be able to narrow it down like I have. So the issue I'm trying to solve is how to enable 1T and BI without Windows crashing. That's the issue. If memtest runs for a week with those two settings on, but the games still crash, and then go right back to working for hours on end with the settings turned off, what am I supposed to do? Start changing drivers? I'm willing to do that if I had an idea of which drivers to start looking at. As of now I'm running the latest, manufacturer certified WHQL certified Windows drivers for all components.

 

So I'm at a loss at this point. A large part of my job is repeatable benchmark - I designed and created, and operate benchmark dot zeno dot com, as well as the procedures that are the basis of it. I designed a defect tracking system and build management system from scratch, so I believe I know a repeatable test/benchmark when I see one.

 

I'm not trying to come off like an butt. Really! I just feel that this getting to be a lost cause.

 

I honestly think my best strategy is to use the TCCD like it was intended - abandon the tight timings and go for overclocking with loose timings. I believe this will get me higher MB/s throughput, and will avoid the instability problems (at least those involved with the tight timings - the overclocking will introduce all sorts more I would imagine but that's expected!). Higher MB/s write throughput is what I'm going for. If I have to overclock to get it, so be it. I wanted to avoid it because I thought the system would be more stable, I felt it was fast enough already, and that I would be more likely to get "what I paid for" on my RAM. I know it's capable of 2300MB/s write with the rated timings, but it's only stable at around 1800MB/s, caused by the looser timings. But overclocking to get the higher speed is acceptable as well, since it won't shorted the life of my components enough to matter (like someone already said, it'll get replaced in a year anyway.)

 

Yeah, you send me copy of that mem speed tester you guys use in house and I will show you some big number on that puppy for sure. Hehehehehe.

 

Sent a PM. We freely provide it on our site without restriction so it's free to be shared if you find it useful, but it is restricted from redistribution for profit of course.

[Minsc]

minsc, using the 2-2-2-5 1t timings, and bi on. what times do you get errors at in memtest? if any?

 

also since you have 1.2, they are the same as mine. an the practically the same as the 1.3's. thats one hitch off the list.

 

It's been a little while, but if I recall, it's stable in memtest for three to four hours, but sometimes as much as 10-20 hours before ultimately failing. For some reason or another, BF2 and a handful of other games crash reliably within the first 5 minutes, but are stable for many hours with BI:Off 1T, so I consider them a better test.