Memory problems w/ M6 Extreme - XMP setting

[LaserPete]

I'm getting freezes, lockups, BSOD's and corruptions of Windows 7 OS when using XMP settings with default BIOS settings.

Memory is G-Skill F3-2133C11Q-32GZL. 32-GB kit - 4X8GB. XMP frequency is 2133 Mhz.
BIOS is 1302.
Memory was tested w/ Memtest 86+ ver 5. Each stick for 3 full cycles at 2133 Mhz passed. Using slot A2.
Memory was also tested with OCCT for 5 minutes with no errors. All four sticks together.
The bad things will start happening around 30 minutes after using the XMP profile or when its warmed up.
Same bad things if memory Volts, Freq and Timings are set manually using the XMP setting information.
BIOS is at the "Default Optimized" settings except for XMP and the graphic card PCIe settings.
Window 7 is in Performance mode (%100-%100) with no power saving features enabled.
CPU is running at a steady 3900 Mhz.
No problems running with Auto at 1333 Mhz but thats kind of slow.


Also,
BIOS locks up at Q-code "15" after the power supply is first turned on. After warming up it will get past Q-code "15".
No problems getting past "15", if the memory is set manually for Volts, Freq and Timings.

After weeks of repairing the OS over and over I have now become good friends with ChkDsk /f.

The reasons why I bought this memory:
Its on the QVL list.
32-GB for possible future uses, like a RAM disk or video editting.
Low 1.5 volt to help out with heat on the processors Uncore.
Frequency should be enough for my target OC of 4.2 Ghz.

I have screen shots of the BIOS settings if needed. What size, BMP or JPG and do I use "attachments"?

Any help would be appreciated.

Thanks,

Pete

ASUS Maximus VI Extreme
Intel i7-4770K
G-Skill 4X8 GB 2133 MHz C11
Samsung 840 Pro 256GB - X2
WD 1TB Black
EVGA GTX-780 Ti SC w/ACX
EVGA SuperNOVA 1000 P2 PS
Corsair H110 cooler
Corsair 650D case
Windows 7 Pro 64

[COLDSTATIC]

LaserPete wrote:
Thanks for joining in Coldstatic,

With answers come more questions.

To start with I'm running 3900 MHz in Turbo mode at 100% through Windows 7, fixed. BIOS is 1302.
I'm using 32 GB's of RAM at 2133 MHz XMP.
I've left Hyper Threading on for now, might like to use it later.
I added .302v to System Agent, 0.848v to 1.15v. The OS doesn't freeze or give BSOD's during logins now, so far that is. Thanks Nate!

Some of the question I have,
Do any of these voltages scale with load or frequency?
When you mention voltages like; .100, .150, .200, etc. are you refering to entering offsets?
Where do these voltage calculations come from and do they take into account different memory setups?
Whats the best software to use for memory stablity tests? I have been using OCCT Linpack 64bit, all cores, AVX off. It never detected a problem.

I want to make the memory stable at XMP first before I start to OC. My target OC is a modest 4.2 GHz but it would be fun to see what it could do. I will drop the memory to 1333 MHz when I start OCing though. I also would like to use adaptive mode (Turbo mode?) to reduce heat at load and frequency idle.

Any help is appreciated since I'm quite new to this.

Thanks,

Pete

ASUS Maximus VI Extreme
Intel i7-4770K
G-Skill 32-GB 2133 MHz, C11
Samsung 840 Pro 256GB - X2
WD 1TB Black
EVGA GTX-780 Ti SC w/ACX
EVGA SuperNOVA 1000 P2 PS
Corsair H110 cooler
Corsair 650D case
Windows 7 Pro 64

Do any of these voltages scale with load or frequency? no. i helped a friend oc and his only wanted 0.050 sa for 2800 mem. you just have to fiddle until you find the right frequency.
When you mention voltages like; .100, .150, .200, etc. are you refering to entering offsets? yes + off set in my case i did encounter 1 chip that stock was .950 and needed -.050 to be stable
Where do these voltage calculations come from and do they take into account different memory setups? they come from the gods of the interwebs....joke. no its a case of this setup wants this...so if you have 32gb but take out half your setup may no longer work.
Whats the best software to use for memory stablity tests? I have been using OCCT Linpack 64bit, all cores, AVX off. It never detected a problem. i use a 10gb ramdisk so honestly meh.....when i run my encoding from handbrake or avidemux if it has any instability it shows there.

on a side note imo .300 is really high even though its only 1.15.

search all posts by me and you can find some useful info about adaptive vs offset

info from raja from an old thread of mine.

Raja@ASUS wrote:
The voltage to the System agent does not impact performance in this way, so you're looking at the wrong thing here Coldstatic. The voltage helps with stability as it supplies the IO transceivers - higher voltage can help reach required slew rates (acceleration) for faster switching frequencies.

The PCH communicates with the processor via the DMI bus - likely DC coupled with the PCH side being supplied from the PCH core voltage and the DMI transceivers on the CPU side tied to one of the rails (SA, IO or VTT etc). Either way, performance should not really change to any appreciable degree simply by manipulating the IO rail voltages.

I could post some suppositions about error margins and the buffers (crudely related to required slew rates having a small change in a benchmark score) - but I'd rather not go down that route on a forum as it would not mean a thing in terms of accurate information (I'm not even sure such fail-safes are part of the design).


The performance hit you're experiencing is simply due to you lowering the cache ratio and possibly DRAM divider (if IRST drivers are used as they reserve memory as cache) to reach a higher CPU clock. The DMI bus will communicate with the CPU via the ring ratio - so lowering the ring ratio speed will impact the DMI perf to some degree.

On the M6E there's a nifty trick to help with BCLK that can impact DMI, but that only ties in to BCLK ranges and not the type of OCing you're eluding to above.

Raja@ASUS wrote:
Memory training is performed during pre-POST. As memory stability is reliant on the right level of VCCSA and IO-D, any instability during the training process will extend POST time because the IMC will re-attempt to train the DIMMs if they are not fully stable (if they continue to be unstable the board will halt with a memory training or IMC related error code).


Friendly note: May I suggest you break up long posts a bit more with paragraphs, as it makes reading them difficult 😛


PS: If your UEFI version has "source clock tuner" try using different values and comparing PCH performance.

http://rog.asus.com/forum/showthread.php?40712-System-Agent-IOA-IOD-hd-input-output-stability

[Nate152]

OK Pete, This is good a bump in the system agent voltage made code 15 flee. You can try xmp mode, if you are still unstable, I would just set it up manually as we will either have to add more voltage or relax timings and we don't want to do either. I would not drop the ram speed to 1333 MHz. For best performance we want to have your ram running at its rated speed. I see you have the H110 listed which is a very good cooler. I see you getting a little more than 4.2 GHz. It is easy to overclock your cpu and most settings can be left on auto. Temperature controls how far you will be able to overclock. You want to see 80c or below when stress testing. Enable speedstep and set the windows power plan to balanced to throttle down your cpu when not under load. If you would like help overclocking to 4.2 and beyond just say so. Install cpu-z, Hwmonitor or real temp, and prime 95.

Memtest86 is a good memory testing tool.

[LaserPete]

Hello Nate and Coldstatic,

The lastest, I've decided to up grade the memory to G-Skill TridentX 2400 C9 16GB (4X4). I ordered yesterday.
I think I probably have a weak IMC on the CPU so maybe 16GB instead of 32GB will fair better.
I lowered the System Agent to 1.05v from 1.15v (now +.203v) and it boots into OS fine. The next thing is to test with MemTest86 and lower it until it fails than raise it by . 025v. Does this sound about right? Also leaving Digital and Analog I/O's on Auto.
On cold board starts it still hangs up at "15" when trying to retrain, except if I enter the memory settings manually. So that issue has not FLED.
The MRC Fast is still enabled, should it be?
I will be tuning this memory until the other arrives. Not sure what to do with it, keep it or try to return it for a loss. Probably just keep it.

What issues might I encounter with this new memory kit? I'm guessing the same because of a possible weak IMC but for all of the tweaks I have to be doing I'd might as well be doing it with the faster memory. Any special consideration for using the memorys 1.65v over the current 1.5v?

Nate, I really do like the H110. I ran OCCT CPU test for 15 minutes and got a peak of 61C with 20C ambient at 3900 MHz, Vcore of 1.213 on Auto adaptive/Turbo. By the was, it ran 5C hotter if the cooler fans were running over 1000 RPM. So I used the Standard fan profile. I have the fans pushing the case air up and out through the radiatior. The two case fans are pulling air into the case and the metal mesh guards for the case fans have been cut out. The Corsair 650D case fans are left on low using their 3-position switch.

I will be getting AIDA64 soon. I do have HWMonitor and CPUz and only use the temps in the Intel section of HWMonitor.

Coldstatic, Very interesting reading. I'm sure I could tweak this 32MB memory kit just fine but I'm beginning to not trust it. I really don't need 32GB, for now anyways. Do you have any tips, thoughts or pointers on setting up the TridentX 2400 C9 16GB ?

Thank you both for the feedback!

Pete

[COLDSTATIC]

Ok i need to correct my earlier statement of .300 to high. I was wrong. I thought .300 was in the red but its not. Also i decided to play around with my settings a bit and found some interesting results. I will post up more details in a bit.

[COLDSTATIC]

Ok so i went back and re-read your initial post. I completely missed the fact that its xmp settings were for 1.5. Thats weird. In my head all mem is 1.65 in xmp. The funny part is that the original idea with ddr3 was to get the volts lower. They originally planned to get down to like 1.2 but with people overclocking memory so much it became popular and then the standard.

Are you running the extra ram timings? If so they might be causing part of your problem. Secondly with your memory warmed up vs cold boot issue....there is a setting (cant remember what its called) that disables some of the mem training if the system is already warmed up b/c the memory will train differently war or cold.

Now i am gunna tell you how i dialed in my volts for SA ioa and iod. Hopefully it will give you an idea on how to do yours. A few days ago i decided to manually for the xmp settings and not let it sit on auto. I also enabled the extra ram training. I immediately started getting crashes when encoding. Games and everything else was fine but under 100% load after 20-40 min it bsod'ed on my twice. Also my system has a nifty little trick to tell me its not right. After the system is warmed up and i restart if the me is off when windows initially starts the icons on my desktop dont load for like 20 seconds (and overall the system just feels every so slightly less snappy when i initially do a task).

If i took mem down from 2400 to 2000 it was fine on its current volts. I cant completely remember if it was .100 or .150 for sa and .1 for ioa and auto for iod. The extra volt would bring ioa up to 1.115 iod on auto was at 1.115 and sa was 1.015.

I then moved mem back to 2400 and sa to .200. Still same windows slow start issue. Then i moved sa up by .025 and restarted the went on a few times. As it gets closer the slow start would lessen a bit and ended at .250. The i started bumping ioa by .015 and the same thing it would slowly get better then go bad again as i went past the sweet spot i went all the way up to .150 or .165 which was now putting the ioa at 1.145-1.160.

I then decided that my iod on auto was still at 1.115 and i wanted it closer to my ioa (this is not always the case....before i messed with timings the ioa was .100 lower than iod). So i bumped iod by .045. After the restart with the +offset of .045 my iod went from 1.115 to 1.065. So i had to bump it up with .150 and now it was close to ioa again. Once again the system is getting slightly better each reboot.

Now i continue to follow the same process as before but moving in .005 increments. I started with SA and moved down to .215. Then iod to .140. After ioa to the same. But things were not improving again. So then i would bump sa +.005 then on next restart -.005 iod. then another -iod and so on and so forth until i found the sweet spot.

I ended at sa .230 ioa .130 or .140 same for iod. the end result is ioa 1.160 iod the same and SA 1.045. Dram volt is 1.6700

[LaserPete]

Hello Coldstartic,

Yes, the memory is spec'ed at 1.5v XMP. This mem kit is also on the ASUS QVL. I thought it might help out the CPU with temps and of course theres the Intel's warning of < 1.65v.
No more tipy toeing around it, I'm jumping in with both feet.
My quess it that I have a weak IMC in the CPU and with 4 sticks at 32GB total it needs help. BTW I expect the same with the new mem thats on the way, also 4 sticks but 16 GB.
This mem at 2133 MHz XMP has really loose timings of 11-11-11-31, which I thought would be easier to OC but the overall mem speed is about equal to 1600 MHz with tight timings.
I have loaded XMP timings and have manually entered the primary timings of 11-11-11-31-2N. I haven't messed with the secondary timings. I think this is what you mean by extra timings. The secondary timing values are qathered during Memory Training, correct? In my case memory training gets hung up at "15" on a cold power on, first bootup after turning on the power supply.

Now on to what I've changed,
I lowered SA from +.300v to +.200v to give me 1.05v SA, auto was at 0.848v. This crashed once at login but not always.
I then raised IO Digital +.100v to 1.125v, auto was at 1.025v. So far I've had no crashes at logins.
I ran both MemTest86 and OCCT Linpack for hours, before these changes and they detected no problems.
Next I might drop SA to +.150v and if necessary raise IO Analog by +.050v to 1.071v, auto is at 1.021v.

The real test for me is getting passed logging in. This is when an amazing number of services are all starting at once.

From what I understand, the MRC Fast enabled (under MISC in BIOS) allows memory training to be disabled on warm restarts but I'm not sure if this should be turned off.

The new memory will be better, just kidding. I expect that I will be going through this again.

Also, I am using Kaspersky AV that came with the board software. Can anyone recommend anything better. I do have a subscription licence to install Norton 360 Premier on another computer, just not sure if I want to install it.

Thanks,

Pete

ASUS Maximus VI Extreme
Intel i7-4770K
G-Skill 32-GB 2133 MHz, C11
Samsung 840 Pro 256GB - X2
WD 1TB Black
EVGA GTX-780 Ti SC w/ACX
EVGA SuperNOVA 1000 P2 PS
Corsair H110 cooler
Corsair 650D case
Windows 7 Pro 64

[LaserPete]

Update,

Windows crashed at login so I upped the Analog IO +.050v to 1.071v from auto of 1.021v.
I compared the ASUS SPD Info BIOS page with the memory timings Second and Third setting page.
Many seem to be different than they should be.
Since I'm having troubles with lockups during memory training at code "15", I'm not surprised.
But I'm really not sure.
Next I will disable MRC Fast BOOT and see what happens.

Here are some screen shots to show what I mean and for those that like pic's.



Thanks,

Pete

[LaserPete]

Sorry, duplicate post!

[COLDSTATIC]

Its my understanding you want mrc ON so that when you restart it will not fully train ram again. My understanding is that warm trained would have better timings but might not start correctly.

Instead of making .050 jumps maybe try .015 to get closer to what it wants. Also give 1.52-1.55 a try for the ram. BTW that's where the IMC pulled parts of it power from in nehalem.

[LaserPete]

Hi Coldstatic,

Well, I adding +.050v to the Analog IO which made it worse. So I set both Analog and Digital IO's back to Auto. Then raised the SA up to +.215v so now its at 1.065v. Also raised memory voltage from 1.50v to 1.52v. Three good restarts so far. Maybe the big thing here is that I also entered the memory settings manually using the XMP profile (2133 GHz, 11-11-11-31-2n). I'm also wondering if it would be more stable if the CPU was running at something higher than 3900 MHz?

I might even start dropping the SA if it looks to be too stable, to see where that point is.

There is a small town here in Oregon named Nehalem, I wonder if that where Intel got the name from?

I will leave MRC Fast BOOT on enabled.

Thanks,

Pete