Any good memory timing/troubleshooting guides?

M

musky

[H]ard|DCer of the Year 2012
Joined
Dec 14, 2009
Messages
3,154
I did quite a bit of searching last night for a guide on how to set up memory after beating my head against a wall for hours trying to get one of my machines stabilized. I did not have any luck to speak of, so I am hoping someone has found something that is helpful. I hear about memory problems fairly frequently around here, so I thnk something like this would be useful when you are overclocking about anything. Things like:

- Where should I start and where should I go for stability with my tRRD, tRTP, tWR, tWTR, and tRFC setting if my memory does not have an XMP profile for the rated memory speed (**cough OCZ cough**)?

- What is the best testing program for memory, both in and out of Windows (Memtest86+ ain't it btw)?

- How much voltage is too much (did you know there is a safety corelation between Uncore voltage and memory voltage? I didn't)?

- Is there any way to isolation test individual memory sticks in a dual or triple channel setup without removing memory?

- What do various symptoms (hang, reboot, program crash, etc.) mean in terms of possible memory issues?

I know several of you run your memory on the bleeding edge, so any experience any of you have that could help others is much appreciated. I am actually just trying to get an OCZ Gold PC1600 kit to run at its rated speed (and am failing miserably), so my goals aren't nearly as extreme.
 
CPU-z great a great toll for getting the base memory timmings. my ocz doesnt have a profile for 1600mhz either. standard timming for mine are 8-8-8-24-88-2t@1600 7-7-7-20-72-1t@1333
 
I was at 9-9-9-24-88-6-12-6-6 2T, and it still wasn't folding-stable at 1080. It passed Memtest86+ and LinX at up to 1520 or better. This is why I am confused. I switched to some OCZ Platinum I happen to have, and am folding at 1688 with the same timings. I don't follow the 88-6-12-6-6 part and if that has much bearing on stability. Both sets default to 59-4-8-4-4, but that is for 7-7-7-18 @ 1066, so I suspect it is wrong. 88-6-12-6-6 was based on the XMP settings for some other brands of PC1600 memory.
 
Don't bother with XMP.
/thread :D

**I don't use XMP settings on any systems and I have no problems oc'ing the memory. I also don't go nuts with the timings, I just do the usual first 4. (ie 8-8-8-24, 9-9-9-27)
**I use Memtest86 off a bootable USB thumb drive. 1 complete pass of all tests is pretty much a sign of stability, you can also do a bunch of passes (say 5 or 10, it goes quick) of test 5. [general consensus I've found = test 5 is pretty stressful on ram, it helps point out bad oc]
**Only way to isolate is test 1 stick at a time (unfortunately)

Best way to test ram/test your oc accordingly is to follow the 3 step guide, separate your bclk first, make sure that's stable, then test your ram and make sure it's stable at your stable bclk speed, and then move onto raising the multiplier. (assuming you're talking about i7 of course). He does leave out some things like changing uncore frequency/UCLK to 16x and making sure QPI is set at 18 (or 36 depending on what the bios setting is). I thought my one 920 was a dud and wouldn't do higher multi with higher bclk until finding out I needed to change the UCLK and take it off auto (stupid Gigabyte overcompensating it :eek:).

I've got my 6gb of 1600 Corsair running at 1680, didn't have to loosen the timings (still 8-8-8-24), didn't have to raise the juice (still 1.66vdimm). I'm wondering if messing around with XMP is what's causing your problems....I'd leave the timings alone, again other than the first 4.

The safety thing you're talking about is pretty simple, you have to keep Vtt (or QPI voltage) within .5v of the vdimm voltage. IE on stock settings my 920 Vtt is 1.175, while the vdimm is 1.66 putting them within .5v of each other. Danger would be leaving Vtt at 1.175 and running vdimm at 1.9. It's something to think about, but not necessarily worry about - the chances of you approaching those boundaries are pretty slim.
 
OK, here is the problem. This particular memory is Memtest86 stable at 1600, 8-8-8-24 (which is what it is rated for.) The board at current settings is stable to at least 211 bclk, and the board + cpu is stable (and folding away happily right now) at 211x19. However, the system with this memory at about any possible setting is not at all folding stable. It will pass LinX at 200x19 with memory at 1520, but F@H will crash in about 10 minutes. Two different systems have done the exact same thing with this memory. So, I have to conclude that the memory is bad. What I don't know is if it is all bad or just one stick, and how to find out.

I don't see anything in you post that I haven't done (although everything there is very good advice.) I didn't load the XMP settings since OCZ memory does have them. I was just using XMP settings as a place to start for manually setting the timing beyond the 8-8-8-24-2T part since I don't have an "Auto" setting on this particular board. My memory voltage has been between 1.64 - 1.68 while trying different things.
 
Sorry, my reading skills are lacking today. :eek:

You have to test 1 dimm at a time, memtest won't allow you to isolate testing on a specific dimm if more than 1 is installed.
 
Memtest passed with all three sticks installed up to 1600MHz (I didn't try any higher), as did LinX. The only thing that fails at 1600 is F@H.
 
When you say it "fails", what exactly happens...client problems? System crash? BSOD? etc.
Are you sure your cpu oc is stable? When you say it passes LinX, what's your testing methodology...are you using the default set in LinX of 10,000 on problem size, or are you having it do larger problems (therefore testing more of the memory)?
How are you testing in LinX, what's the other info...multi, bclk, volts, divider, QPI, UCLK etc. It sounds like a cpu oc issue to me since the ram passed memtest.
 
Right now, I have the same CPU, same board, higher overclock, same memory timings, same every other setting but bclk, but different memory. I literally pulled the old memory, put in different memory, changed the blck to 211, and let it run. It has been folding for 3 hours now without issue. With the old memory, this same rig was unstable folding at 19 x 200, 20 x 190, and 21 x 180, as well as several other combinations. With this new memory, it is stable at 190 x 211 and 20 x 200 that I know of. The memory is the only difference, so I highly doubt it is a general cpu overclock issue

"Fail" is normally the fah client crashing, not the machine. I uninstalled/reinstalled it probably a half dozen times. What is running right now was the last install that was unstable with the other memory.

I am running LinX with whatever the default settings are - 20 passes, takes about 6 minutes to run.

Stable setting with current memory
19 x 211, 1.296 Vcore, 1.336 VTT, 1.66 memory, memory at 1688 9-9-9-24-2T, QPI and UCLK are whatever the default is.

Unstable setting with other memory:
19 x 200, memory at 1600, everything else the same
20 x 190, memory at 1520, everything else the same
21 x 180, memory at 1440, everything else the same
21 x 180, memory at 1080, everything else the same
I have tried changing Vcore up to + 0.05, VTT+/- 0.02, and memory voltage +/- 0.02
 
sounds like maybe a bad mem chip in the mix? and now that you mention it i have been through the same problem. qpi should be the lowest setting
 
After another several hours yesterday, I am still no further. The system is unstable at any speed with this memory. I isolation tested each chip, and they all caused F@H to crash. They also all pass Memtest86+ and LinX both individually and together in triple channel. The exact same settings with different memory are running perfectly right now. Also worth noting is that F@H crashes, not the system. It never rebooted or hung or anything unless I did something silly with a setting while trying to get it stable. I'd get a pop-up in Windows saying fah6.exe has stopped working - do you want to close the program? If I clicked yes, then hurried up and ctrl-c'ed the cmd window, I would not cause any issue with the current unit.

As far as the two settings I had on Auto, here is what they actually are:

QPI frequency - I have "Full-speed" and "Slow-speed" as options. Full is 7.2 or so at BLCK 211. Slow is 4.8 at BCLK 200. I have it on slow now. The stable memory I have in the system now works with either setting at BCLK 211.

Uncore frequency - Auto is 2x memory, so it is 3200 with BCLK 200/memory 1600. I just left it on auto since from what I could tell, that is correct.

I am really stumped on this one. I'm going to either need to figure this out or be down an I7 system after I get my E5620 this week to fire up my SR-2.
 
You must log in or register to reply here.
Back
Top