CMD12GX3M6A1600C8 and DX58SO2

This is a long post, but I like details! I've been building systems for years and usually go with that value ram stuff that is more or less plug and play. This is the first time I've had to fiddle with memory settings quite so much, and I'll tell you now I'm not extremely familiar with overclocking.

About 5 months ago now I built a system based on the Intel DX58SO2 motherboard and a 6GB kit (TR3X6G1600C8D). I had some success running the XMP profile, however occasionally experienced reboots with a nice red light flashing on the motherboard. When I entered BIOS memory would be reset to 1067mhz (This RAM is 1333MHz SPD). I also experienced problems with the system hanging when resuming from any form of sleep/hibernate when using XMP, but anytime I use a 1333Mhz frequency resumes from sleep work fine.

A month or so later, I decided to add a second kit of the same memory (TR3X6G1600C8D), and found I was unable to run at XMP settings without crashing during windows boot. I was able to run memtest and pass a full day of testing various combinations (single sticks, 3 sticks, 6 sticks). I used SPD settings of 1333MHz 9-9-9-24 1.5v and the system stopped crashing in windows. At this point I contacted tech support and they thought that using two separate kits together was perhaps a problem...even if they were the same model number. They offered to RMA the memory and exchange it for a 12GB kit (CMD12GX3M6A1600C8). Unfortunately, this did not change anything.

Now I've been back and forth with Intel on this issue several times, and I can never get past them telling me that 1.65v ram is not supported, or trying to get me to RMA the motherboard. Aside from those two nuggets, they offer nothing else. In all my browsing I did manage to find a PDF document somewhere for a X48 chipset and found in it a small note about populating the second channel, and it requiring a drop in the frequency. Unfortunately I can't locate that document again, nor can I find that information is relation to my board. I did find some references to that on these forums in a few places however. To be honest I'm not sure why its not printed in the manual for the board somewhere, or somehow stored in a second XMP profile, but that's not important.

There's the history and details, now I'm just trying to find the fastest configuration that won't fry my CPU and void my warranty like Intel claims 1.65v will, but is above all stable. So far I've run at 1333MHz 9-9-9-24 1.5v and made it 8 hours into Prime95 before threads started to get rounding errors. Next I tried 1333MHz 8-8-8-24 1.55v and made it 14 hours into Prime95 with 0 errors..yay! (but no). After running at that setting for 3 days, my PC rebooted and did the red flashy light thing and wiped my memory settings for me. Now I'm running at 1333MHz 8-8-8-24 1.6v and made it 17 hours into Prime95 with no errors. This is the XMP profile with the frequency lowered and voltages (memory and uncore) increased slightly. My only concern is that I passed a long Prime95 before only to have the system become unstable and freak out anyway, and I'm worried it might continue.

Is 1.65v safe for my CPU and motherboard or should I keep it lower if I want my hardware to live a long happy life?

If 1600MHz is overclocked for one full channel, is 1333Mhz overclocked for 2 channels? Should I be looking at 1067Mhz for real stability and hardware longevity?

Are there perhaps other settings I should be tweaking that might help? The following are my current settings, XMP profile with the following changes, mem multi changed from 12 to 10, mem voltage changed from 1.65v to 1.6v, Uncore voltage changed from 1.2v to 1.225v.

tCL = 8

tRCD = 8

tRP = 8

tRAS = 24

CMD = 2T

tRRD = 6

tWR = 12

tRTP = 6

tWTR = 6

tRFC = 128

tFAW = 30

tRC = 41

Memory Multiplier = 10x

Memory Voltage = 1.6v

Uncore Voltage = 1.225 (I increased this from the XMP setting of 1.2v to default)

Uncore Multiplier = 20x

Any suggestions are welcome! The system seems stable enough but having found my way here through a series scattered information and guesswork, and especially considering the evil red light looming around the corner, I'm still not confident in my PC's stability. I would like to get some sort of performance gain over the cheap ram I passed on for this, but at the same time I've sort of given up on the idea and I'm at a point now where I just want to see my system not crash or reboot anymore.

Is 1.65v safe for my CPU and motherboard or should I keep it lower if I want my hardware to live a long happy life?

This makes me laugh at Intelstech support every time. XMP is THEIR technology . It's THEIR technology that sets the RAM voltage to 1.65v. So NO it will not harm your CPU. I can tell you first hand that i have been running 1.65 v for almost two years now with no hiccups at all.

If 1600MHz is overclocked for one full channel, is 1333Mhz overclocked for 2 channels? Should I be looking at 1067Mhz for real stability and hardware longevity?

No 1600mhz would be your overclock period. Whether it be single or dual channel. Single or dual channel is a function of the MB and not the RAM. And 1333mhz would be stock max frequency in single or dual.

Are there perhaps other settings I should be tweaking that might help? The following are my current settings, XMP profile with the following changes, mem multi changed from 12 to 10, mem voltage changed from 1.65v to 1.6v, Uncore voltage changed from 1.2v to 1.225v.

Yes, you may need to add voltage to the memory controller. It should be labeled QPI/VTT or something similar in your BIOS. It should be set to a value of 1.35v to 1.45v.

If increasing memory controller voltage does not work you may have to back the frequency down to 1333mhz and see if it is stable then.

There is always the possibility that your CPU's memory controller just wont handle the overclock. Out of all the I-7's the 950 has the worst memory controller and does not always handle the overclock real well.

This makes me laugh at Intels tech support every time. XMP is THEIR technology.

I'm not gonna lie to you the tech support was aweful. I did try to point out that XMP was an Intel term as far as I could tell, based on all the documentation. At one point I asked Karla too many questions and she scolded me about taking risks, voiding warranties, and all the information is right there in the documentation. Then she literally hung up on me (via live chat online). None of that courtesy stuff at the end.

No 1600mhz would be your overclock period. Whether it be single or dual channel.

It's a triple channel motherboard, x58 chipset. What I meant to describe was going from 3 dimms to 6 dimms.

Yes, you may need to add voltage to the memory controller. It should be labeled QPI/VTT or something similar in your BIOS. It should be set to a value of 1.35v to 1.45v.

Ok I increased the Uncore voltage (QPI) from the XMP setting of 1.2v, past the "default" setting of 1.2250, all the way up to 1.25v. I figured I should start small and test. As I'm writing this reply, I'm running 6 sticks of ram at 1600MHz again and running the memory stress test in Intel Extreme Tuning Utility just to see if it would bluescreen on me. I also ran a few minutes of Prime95. Obviously not long enough to know much but long enough to pass the "crash in the first 2 minutes" test. I wonder why the XMP setting would reduce the voltage below its default....anyway I'm still here and its seems stable so far.

Thank you for the quick reply and the info. I really appreciate it. I'm not counting my chickens yet but this is further than I've gotten in a while. I'll post again when I know more.

Update:

I tested sleep and resume just now, and for the first time ever while running at 1600MHz, I was able to resume from sleep! The only problem I see is it takes a few seconds longer than it should, then an icon appears in the notification area showing "Display driver has stopped responding and has recovered". At 1333MHz resume has always worked well and doesn't display the video driver error. At 1600MHz it hasn't worked at all, ever, until today. This is definitely a step in the right direction, but I suspect the system still isn't 100% happy. I know this isn't the nvidia forums, but it seems directly related to ram settings. I thought it was worth mentioning.

I would add a little more to the QPI voltage. It shoul be around 1.35v-1.45v.Just dont go over 1.45v That I can tell you is intels spec. That should smooth out the rest of your problems there.

As far as the XMP setting not being right. I'm not sure. Some MB get them right , others miss it. So it's just a good idea to check and adjust if necessary.

Well I ran pretty event free with 1.25v uncore/qpi for about a week. I came home today and clicked a mouse button to wake my PC and it didn't successfully resume. It was however a bit different than the way it used to not quite wake up when i ran the ram at 1600MHz with a 1.2 uncore. By that I mean the post code displayed on the boards LED read "30" which is is consistent with a successful resume, whereas before it displayed a different post code (that I don't recall offhand). Still, the video didn't seem to want to wake up.

Oh I should also mention the issue noted in my last post. The problem with the video driver telling me, "Display driver has stopped responding and has recovered", appears to have been resolved by tweaking the power management features related to PCI express "Link state power management". Some forums suggest using the high performance setting that would turn off this feature, but it seems setting it to moderate power saving actually resolved the issue with the driver. With the exception of today's issue with the video waking up of course.

Power management problems could be any number of things, but since ram settings have had a huge impact on the success of resuming from sleep in the past, it's worth mentioning as a possible ram setting issue still. I've gone ahead and stepped the qpi/uncore to 1.30v now. Still below your suggested 1.35v - 1.45v. I'll give it another week of sleeping/resuming and see if it has any noticeable effect. Since I only had one unsuccessful resume out of perhaps a dozen, it might be hard to recreate the problem with any reliability.

Overall I'm still quite happy with the overall stability so far with only slight changes to the voltage settings. It's been pretty depressing having to run my $225 memory like its $60 memory. Sure it looked pretty with the fins and the fan on top, but that was just a reminder that my RAM wasn't running warm like it should.

Again many many thanks to Peanutz: D

ower management problems could be any number of things, but since ram settings have had a huge impact on the success of resuming from sleep in the past, it's worth mentioning as a possible ram setting issue still. I've gone ahead and stepped the qpi/uncore to 1.30v now. Still below your suggested 1.35v - 1.45v. I'll give it another week of sleeping/resuming and see if it has any noticeable effect. Since I only had one unsuccessful resume out of perhaps a dozen, it might be hard to recreate the problem with any reliability.

This just reaffirms that it was memory controller voltage. It may take a little more to completly get the rest of the bugs worked out. If 1.3v does the job , thats great! Generally the lowest voltage to maintain stability is the best. If it doe not stay stable or you continue to have the sleep/wake issue just add a little more QPI .

Again many many thanks to Peanutz: D

Glad i could help! Just post back if you still have problems.:)

I didn't make it very long at 1.3v. Within just a few hours later I had a crash with the flashing red light red light. I was in the middle of installing an Nvidia beta driver using the "clean" install method and it had just removed all my video drivers. I wasn't paying very close attention, but it was right about the time the installer might be rebooting the system, or the system may have crashed. Usually a crash precedes the red light.

A hardware watchdog timer is included as part of the board circuitry and is enabled by default in the BIOS. This timer will automatically enable the board to startup if unstable system operation or a failure to boot is detected. If the watchdog timer detects an issue, the screen shown in Figure 3.1 will be displayed. In addition, on Intel Desktop Board DX58SO2/DX58OG, at the Diagnostic LED bank, the red WD Fire/B2B LED will light when the Watchdog Timer is activated. The user can then go into the BIOS screen and manually reset performance parameters or use the F9 key to reset all BIOS defaults to restore system stability.

I reverted back to a 1.25v qpi/uncore and haven't had a problem. At this point I can continue to run at 1.25v and wait to see if I ever have a problem, or I can go back to 1.3v and see if I have a problem again, or I can go up to 1.35v. I'm just slightly concerned that 1.25v made it a week with no real problems and only 3-4 hours at 1.3v.

Since my last post, I've been running the QPI at 1.25v again. Today I encountered another watchdog event. This time it happened while running a boot time defrag of the pagefile with perfectdisk. The last two events occurred just before or just after an intended reboot. This could be a coincidence, but I thought I should mention it.

So it appears that 1.25v isn't any more stable than 1.3v. Having eliminated voltages in that range from being stable, I'm going to finally try 1.35v and see how it goes. I will keep you posted.

Can someone explain XMP a little bit for me? I've read Intel's "brochure" on the subject, but it doesn't really go into great detail. Is it a profile that is stored somewhere on the memory itself, or are the settings stored by the board? If they are stored on the DIMMs, are there multiple XMP profiles stored for various boards? The docs mention that the settings are a result of a collaboration of Intel and various memory manufacturers. In the end, who is really responsible for determining the settings? The reason I ask is that it seems odd that the QPI would not only be below default, but way below the settings recommended by you fine people. Is this perhaps the result of Intel being overly cautious? XMP seems like a great idea, but in my case it's not able to do what it claims it's there to do. I'm just curious as to why this might be.

Thanks Mr. Ram Guy for your input on this. I appreciate the advice. A second opinion is always reassuring.

I don't have a screenshot to show it, but I ran HWInfo64. The way it organized the information in regards to XMP profile and voltages was interesting. It sort of looked like voltages such as QPI are not dictated by the XMP profile at all but by the motherboard. This might explain why it was necessary for me to make adjustments to the memory settings after enabling the XMP profile. On the other hand, the uncore/QPI didn't remain at the default of 1.2250..it was reduced to 1.2 when loading the XMP profile.

I found this wikipedia page that describes the information stored in the DIMM, and the only voltage i see mentioned is VCC:

http://en.wikipedia.org/wiki/Serial_presence_detect

This might be a moot point now that the corsair forums have helped me resolve the issue for the most part. I guess I'm just curious why this XMP feature that is intended to make configuration easy, fails to achieve stable results without knowledgeable user intervention. Learning is fun and all, but I know more about RAM settings now than I ever intended to know. I still wish that I hadn't had to wrestle with my PC for 5 months to get it to run right.

As an update to the QPI voltage conversation, I've been running at 1.35v for several days now and all is well so far. I'll keep an eye on it and post back if anythign happens.

I would start the stress testing again, except I've found that my system has managed to pass very long memtest and Prime95 sessions with no errors, only to crash while simply using windows under normal conditions. I was under the impression that "Prime stable" was harder to achieve than "sitting there doing nothing stable", but the last few events that occurred seem to show otherwise.

Well, I've been running several weeks with pretty much no problem...until about 30 minutes ago. My system blue screened and I saw the words "Memory Management" in the text before the system attempted to reboot. I say attempted, because I was unable to post or enter BIOS. The red watchdog light was flickering away to let me know something bad happened and I should check BIOS. The post error code was 1c and the speaker emitted 3 long beeps, then a pause, the repeated (memory error).

I removed 3 of the 6 sticks of ram and powered up the system and it's now working. I have to assume for now that one of those 3 sticks just up and died on me. I guess I'll be testing ram sticks again this weekend.

This may or may not be related to all the other troubles I've had, although I've swapped memory at one point early on in the months of past issues. If it is related in any way I would not blame faulty ram for my past troubles. I might however blame my the other issues on the now faulty ram.

Sometimes I wish I just bought cheap ram specced for the system instead of trying to get a performance gain out of more expensive memory. If it turns out I need to RMA again I might just do that.