Stabilizing i7 920 @ 4.2 with 12GB 1600 RAM

[evilrad]

First, the system specs:

i7 920 DO @ 4.2Ghz
Asus P6X58D Premium
Patriot Viper Xtreme 12GB (3 x 4GB) DRAM DDR3 1600 (7-9-7-21)
XFX Radeon 5870
Corsair 750TX Power Supply

The system that I thought was stable, actually wasn't. I did a 24 hour run with Prime95 on both small FFT and blend test without any problems. I also did an 8 hour run on Memtest86+ without errors. It was only after I started gaming that the system would lock-up. And one time the system BSOD durning a video (mkv file).
After some searching, I found out that 12GB of 1600 RAM is too much of a strain for the IMC (should have done my homework before I bought the RAM). Instead of having to run a massive amount of volts for QPI/DRAM, I decided to go to the lowest RAM divider and run the memory at 1200. Here are my settings that I'm currently testing with:

Ai Overclock Tuner.....................[Manual]
CPU Ratio Setting......................[21.0]
Intel(r) SpeedStep(tm) Tech............[Disabled]
Xtreme Phase Full Power Mode...........[Enabled]
BCLK Frequency.........................[200]
PCIE Frequency.........................[100]
DRAM Frequency.........................[DDR3-1205MHz]
UCLK Frequency.........................[2410MHz]
QPI Link Data Rate.....................[Auto]

CPU Voltage Control....................[Manual]
CPU Voltage............................[1.375]
CPU PLL Voltage........................[1.80]
QPI/DRAM Core Voltage..................[1.30]
IOH Voltage............................[1.20]
IOH PCIE Voltage.......................[1.50]
ICH Voltage............................[1.20]
ICH PCIE Voltage.......................[1.50]
DRAM Bus Voltage.......................[1.60]
DRAM DATA REF Voltage on CHA...........[Auto]
DRAM CTRL REF Voltage on CHA...........[Auto]
DRAM DATA REF Voltage on CHB...........[Auto]
DRAM CTRL REF Voltage on CHB...........[Auto]
DRAM DATA REF Voltage on CHC...........[Auto]
DRAM CTRL REF Voltage on CHC...........[Auto]

Load-Line Calibration..................[Enabled]
CPU Differential Amplitude.............[800mV]
CPU Clock Skew.........................[Delay 300ps]
CPU Spread Spectrum....................[Disabled]
IOH Clock Skew.........................[Auto]
PCIE Spread Spectrum...................[Disabled]

C1E Support............................[Disabled]
Hardware Prefetcher....................[Enabled]
Adjacent Cache Line Prefetch...........[Enabled]
Intel(r) Virtualization Tech...........[Disabled]
CPU TM Function........................[Enabled]
Execute Disable Bit....................[Enabled]
Intel(r) HT Technology.................[Disabled]
Active Processor Cores.................[All]
A20M...................................[Disabled]
Intel(r) SpeedStep(tm) Tech............[Disabled]
Intel(r) C-STATE Tech..................[Disabled]

I currently have the RAM running at 7-7-7-21 and I'm running 3DMark 11 for many hours to see if it crashes.
Like I said, I'm currently testing the system as I type (posting from my laptop). These settings could be it. Even so, I'd like some feedback on my BIOS settings if you see anything that should be different. Also, is there a faster way to test than running multiple hours of 3DMark? I've even managed to do a 9 hour run of OCCT with different settings, only to have 3DMark crash after 2 or 3 hours. Video card is not overclocked and I can run 3DMark all day with stock settings and/or overclocked with only one stick of RAM running at 1600, so the video card and/or drivers aren't the issue. Heat is also not an issue as I'm on custom water cooling.
Any advise would be greatly appreciated in getting my system 100% stable. Thanks

 

[redduc900]

With the DRAM multiplier previously set to x8 at a 200 BCLK = 800MHz DRAM frequency, what was the VTT (QPI/DRAM Core Voltage) set to before lowering it to 1.30V? And w/ the RAM at 800MHz, did you try either loosening the timings or increasing the DRAM voltage?

 

[evilrad]

With the DRAM multiplier previously set to x8 at a 200 BCLK = 800MHz DRAM frequency, what was the VTT (QPI/DRAM Core Voltage) set to before lowering it to 1.30V? And w/ the RAM at 800MHz, did you try either loosening the timings or increasing the DRAM voltage?

Actually, when I originally tested the system I had the QPI/DRAM set to 1.25. I then loosened the timing on the RAM to 9-9-9-24 at 1600 and eventually made it up to 1.475 volts on the QPI/DRAM. RAM voltage was left at it's rated voltage of 1.65.
I think it was one of your replies that I saw on another post that said that 1.5 - 1.55 volts on the QPI/DRAM was probably needed for that much RAM running at 1600+. I wasn't sure about going that high on the voltage, so I decided to try to run the RAM at a lower speed with tighter timings.
As of right now, 3DMark 11 has been looping through the test for 4.5 hours. If it makes it to 6 hours, I'll probably call it good. Funny how the system can be Prime95 stable for 24+ hours, yet crash in a game or while playing a video file.

 

[evilrad]

So I successfully ran 3DMark11 for 6 hours. So I thought I was good to go. Then today while playing Killing Floor, I got the once famous "Grey Screen of Death." I thought the GSOD issue was fixed long ago?
I have the XFX 587X-ZNFV. I downloaded and flashed the latest BIOS from XFX's website, only to noticed while it was flashing that I already had that version. So what do you think, purely a video card problem or is the system still not stable? What should I try next?

 

[redduc900]

I would probably temporarily remove one module, leaving the remaining two (8GB) in a dual-channel config. Then again change the DRAM multiplier to x8 w/ timings set to their rated 7-9-7-21 at 1.65V, w/ VTT set to ~1.35-1.40V, then after running Memtest86+ through a couple of passes try playing Killing Floor again.

 

[evilrad]

I would probably temporarily remove one module, leaving the remaining two (8GB) in a dual-channel config. Then again change the DRAM multiplier to x8 w/ timings set to their rated 7-9-7-21 at 1.65V, w/ VTT set to ~1.35-1.40V, then after running Memtest86+ through a couple of passes try playing Killing Floor again.

Running memtest86+ is useless for me. I can run it over 9 hours with all 3 sticks in at the rated speed and come out error free.
I upped the QPI/DRAM voltage to 1.35 and the DRAM voltage to 1.64 and I'm letting 3DMark11 run again. Only time will tell if these settings are good. I appreciate the suggestions and I'll post my results (hopefully good results). Thanks

 

[psionic98]

GSOD is almost always gfx related. Did you OC your 5870? If so you might not have a stable oc on the card. Dropping it down 1-2 notches on both mem/core could fix it.. or turning up the volts!

 

[evilrad]

GSOD is almost always gfx related. Did you OC your 5870? If so you might not have a stable oc on the card. Dropping it down 1-2 notches on both mem/core could fix it.. or turning up the volts!

Nope, card is at stock. I let 3DMark11 run last night while I was asleep at my previous settings and I work up this morning to the desktop. According to the system up time, it looks like the computer restarted after about 4 hours.
So I either get a GSOD (it's happened before while running 3DMark11) or a restart. Actually, it could be that the system is restarting after setting on the GSOD for a while. I really don't know because I'm not constantly watching the test.

 

[evilrad]

Just got another GSOD
I've put the ram timings at 9-9-9-24 and upped the QPI/DRAM voltage to 1.4, so I'll see what these settings result in. If this fails, I'll go back to running one stick at a time and see if I can reproduced the GSOD and even test at pure stock settings. I really don't want it to be the video card, since that would mean me having to break down my water cooling loop in order to RMA the card.

 

[evilrad]

Well, just short of 8 hours I got the GSOD once again

So now I'm left with testing with one stick at a time for an extended period. If that passes, I'll test with 2 sticks. If it passes with two sticks I'll be extremely pissed. I can't be the only one in the world running 12GB on an overclocked 920!
It seems that increasing the QPI/DRAM to 1.4 volts allowed the test to run longer (although it could be purely coincidental). So I'll pose this question again: How many volts (normally) on the QPI/DRAM is required for running 12GB? What is the max safe voltage I can run in it if need be? Looking at my settings in the first post, is there anything else that I need to change?

 

[redduc900]

According to Intel, 1.35V is the absolute maximum VTT for the i7-900 series. Not only does the amount of RAM affect the required VTT, but also the number of DIMM slots populated, the BCLK (in your case 200), and the effective DRAM frequency. Asus for whatever reason normally requires more VTT than other manufacturer's boards. Try enabling the DDR3-1600 XMP profile, and then check to see what the BIOS auto-sets the VTT to.

 

[evilrad]

Alright, it's looking more like it's the video card. I ran 3DMark11 last night with only one stick (4GB) at 1200 with 9-9-9-24 timings, and I woke up to the GSOD.
I'm now running on stock settings with the same stick to see if I can get the GSOD. If I do, I'll swap out sticks just to make sure it's not a bad stick (which I don't think it is because Memtest passes). I'm not looking forward to pulling the card. I'll keep you posted.

 

[evilrad]

So I'm 99.9% sure it is the video card now. I ran at stock settings with only one stick of ram and the system locked up during one of the 3DMark11 test. I then swapped out to another stick and got a GSOD while running the test.
To be 100% sure, I plan on formatting and installing the OS with the bare minimal drivers needed and run the test again.

 

[evilrad]

So I decided to resurrect this thread to let you people know that, after almost a year, my system is finally stable. And no, it didn't take me almost a year of testing to get it stable. It was that I put it off for almost a year until the BSODs (during gaming) were really starting to annoy me. It's not that they increased in frequency. I could go weeks without have a BSOD and then get a couple during a weekend of gaming. I guess I just snapped one day and decided to put in the time and effort to make it stable.

So, what was the problem you ask? Turns out that my chip has a very good IMC. With 8GB (2x4GB) of 1600 RAM with 7-9-7-21 timings the stock QPI/DRAM voltage (1.20) is just fine with a BCLK of 200. In order for me to use 12GB (3x4GB) at that speed and those timings, it only needed 1.225 volts.

Every guide that I've read and everyone on any forum has said that you'll need a whole lot more volts in the QPI/DRAM (1.5-1.6) if you're running a lot of RAM at high speeds. Turns out that advice is what was killing me. My chip's IMC HATES a lot of volts. If I use anything above 1.45 volts, then the BSOD start to rain down. What's also sad is that if I use the XMP for the RAM, it sets the QPI/DRAM voltage to 1.4!

Of course, that was only part of the problem. Even though my chip has an excellent IMC, the chip itself is not so great when it comes to vcore. To get a stable 4.2Ghz (my definition of stable = 100 pass of IBT using 10GB of RAM & 24 hours of Prime95) required 1.425 volts in the Vcore.


So what have I learned?
1. I now know to base system stability on more than one stress test. What I thought to be stable almost a year ago, was only stable using Prime but would fail IBT. My friend's 2500K was just the opposite. His system would fly through IBT but almost instantly fail Prime.
2. No matter what people or guides say, always start with the stock voltage. It will save you a lot of time testing if you chip happens to be like mine (doesn't like a lot of volts in the IMC).


Side note: One day I need to update my sig. Maybe I'll update it on the next build.

 

[Shadowmonk]

My Hyperx mem on the 920 build i did with 12gb needed 1.6 QPI, and it never hurt anything, but would GSOD without the QPI voltage jacked up

 

[Woomack]

ASUS boards need higher QPI/VTT than most other brands ... especially P6X58D.
My brother had 12GB ( 6x2GB ) 1600 8-8-8-24 with 1.4V QPI/VTT, cpu @3.6GHz on P6X58D-E and in mysterious way both, board and i7 920 just died after a year. Board couldn't find any memory and cpu was acting like IMC just didn't work.
ASUS boards always had problems with memory or problems with bios. If you check other brands then it's hard to find other producer with memok buttons and other "unique" technologies that make memory work.
If you can't set memory stable @1600 using 6 slots and 1.4V QPI/VTT then first check if higher IOH at about 1.2-1.25V makes any difference. It's working for me on both P6X58D and R3E with 3 different cpus.

 

[evilrad]

ASUS boards need higher QPI/VTT than most other brands ... especially P6X58D.
My brother had 12GB ( 6x2GB ) 1600 8-8-8-24 with 1.4V QPI/VTT, cpu @3.6GHz on P6X58D-E and in mysterious way both, board and i7 920 just died after a year. Board couldn't find any memory and cpu was acting like IMC just didn't work.
ASUS boards always had problems with memory or problems with bios. If you check other brands then it's hard to find other producer with memok buttons and other "unique" technologies that make memory work.
If you can't set memory stable @1600 using 6 slots and 1.4V QPI/VTT then first check if higher IOH at about 1.2-1.25V makes any difference. It's working for me on both P6X58D and R3E with 3 different cpus.

It is "assumptions" like that which cause problems. Everyone says that Asus boards need more volts in the IMC then other brands. It's assumed to be true based on past user experiences, but it's not the rule. My case is the perfect example. I have an Asus board with a lot of RAM which only needs (and will only work correctly) with very little volts above stock.

The reason I brought the thread back to life was to let people know how "rules of the thumb" can turn out to be only assumptions. The only "rule" that continues to be true is that every CPU and motherboard is different. What works for one person may not work for the next.