New rig: Pre-OC'ing stability test errors

[Scshadow]

Haf X
Asus Rampage III Formula
Core I7-950 @ stock
Cooler Master V8 CPU Cooler
3 kits(24GB) of G.SKILL Ripjaws Series 8GB (2 x 4GB) DDR3 1600 (PC3 12800) Desktop Memory Model F3-12800CL9D-8GBRL running in triple channel.
(2x) Palit 1GB GTX 460 in SLI
Corsair 950TX 950 watt PSU
Windows 7 Pro 64 bit

Well first off, I'm on my second build and I intend to OC. My first build was nothing but making sure the Ram was clocked and timed according to specs and unlocking my BE 555's extra 2 cores. On this build, all I've been running is basic tasks and not much gaming on stock settings for several weeks. I want to get a moderate overclock. 4ghz on my i7-950 would be nice for starters.

The overclocking guide I found said to make sure my system was stable at stock. I'm actually a little bit surprised I hadn't ran prime 95 already. When I realized I hadn't even downloaded prime 95, I decided to do so and run it. And to my surprise, I get an error not 10 mins into the blend test.

"FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file."

I've never actually gotten an error from prime 95 because my first build wasn't built for OC'ing and pushing the hardware to its limits. I'm not actually quite sure where to go from here. Any advice? The blend test of prime 95 says it tests the CPU and some Ram, should I start to prepare myself mentally for RMA'ing at least one component?

 

[klepto12]

this could be a problem with your ram why you need 24GB is crazy to much of a chance of one of the sticks being bad i would try them all separate and see if you still get the error.

 

[Scshadow]

this could be a problem with your ram why you need 24GB is crazy to much of a chance of one of the sticks being bad i would try them all separate and see if you still get the error.

Well the 2 stick set was cheap enough on black friday sale. My X58 chipset northbridge has triple channel memory slots. I thought about buying 2 sets and leaving 1 stick out and just decided to go ahead and get the 3rd kit.

Anyway, I was just looking into Memtest86+ 4.10. It appears to support all 24 GB of my Ram so I guess I'll boot into it and run it overnight. Or is it preferred to prime 95 1 stick at a time?

 

[klepto12]

just try mem test first

 

[Scshadow]

Memtest86+ ran one pass of the tests with no errors. I don't really know if its sufficient so I'm going to go out and leave it running, try and give it a couple more passes. Any suggestions for the next test?

 

[Leeky]

I think you should aim for at least 5-7 passes without error before assuming the RAM is OK. Literally run it for 8-10 hours solid overnight and see how it stands in the morning.

Since your using an Asus motherboard, which are notorious for being iffy with RAM timings, can you confirm that your listed RAM timings, as well as the listed voltage are being correctly recorded in BIOS please. Usually more often than not its the BIOS not setting the RAM timings or voltage correctly and the RAM throwing fits now and then, or when pushed.

 

[Scshadow]

I think you should aim for at least 5-7 passes without error before assuming the RAM is OK. Literally run it for 8-10 hours solid overnight and see how it stands in the morning.

Since your using an Asus motherboard, which are notorious for being iffy with RAM timings, can you confirm that your listed RAM timings, as well as the listed voltage are being correctly recorded in BIOS please. Usually more often than not its the BIOS not setting the RAM timings or voltage correctly and the RAM throwing fits now and then, or when pushed.

Its on its 8th pass with no errors now so I'm going to go ahead and stop it.

My ram had to be manually clocked at 1600 mhz it was rated for(seems pretty standard across all boards) and then it auto-configured the timings to spec 9-9-9-24-2n. Now I think I got so excited to have my computer boot up that I neglected to check the voltage. The rated voltage is 1.5v. Looks like Dram core voltage is running about 1.35 on auto. When I select 1.5v instead of Auto, it boots up showing 1.46v in bios but windows hangs and doesn't boot up. No blue screen, it just hangs.

 

[LinkedKube]

Rampage extreme 2 and 3 boards show ..05 more v's than what is actually applied. Put 1.55 in and you should be okay.

 

[Scshadow]

Rampage extreme 2 and 3 boards show ..05 more v's than what is actually applied. Put 1.55 in and you should be okay.

Well, I don't have the extreme board, just formula. But since it is reading 1.46set to 1.5, it makes some sense to me. So I went ahead and raised it to 1.55(actual voltage shows up about 1.51) and to my nooby surprise, it boots again. I figure if it boots at 1.35 but not at 1.45, I thought that'd be a step in the wrong direction.

Now to my surprise, Prime95 is failing faster and on the first round of tests in less then a couple mins. Same rounding errors as my first post. 7/8 workers failed before I went ahead and stopped the 8th. While awaiting a response, I went ahead and started the memory test over again now that its set to 1.55v.

Thanks for the help so far guys. I hope to pass the favor on when I figure this stuff out a bit more.


Edit: Well, Memtest is failing big this time around. Over 8000 errors before it seemingly has quit. I guess I'll start pulling the modules and see if it can be isolated to certain modules. But... how can I have errors now while it made 7 complete passes at undervoltage?

 

[LinkedKube]

I run my board at 1.7v for stability. Has been running that way for two years now. Ran stability test up the wazoo. Still fine. 1.7 = 1.65 actual voltage.


Sent from my DROIDX using Tapatalk

 

[Scshadow]

Alrighty, I've been through a bit of a time testing my ram. Every stick individually passed Memtest86+. So I decided to test one at a time with prime 95. The first stick I choose to do that with seems to have been the bad one as it wouldn't boot the OS. My first thought wasn't to say that stick was bad so I kept trying to tweak the voltage. When I finally moved on to the next stick of RAM, I see I'm able to boot and I'm running prime95. Any idea on how long to let prime 95 go on each stick? I figure I'll go to sets of 3(since my 6th is bad) once I test all of them, I'll let that test go for 24hours. Does an hour per individual stick seem sufficient?

Also I'm seeing 55-60c on my CPU. I read my processor runs pretty hot like that. Although, looks like my computer is auto setting voltage a little high for stock clock. I'm reading 1.15 from forums online while I'm getting 1.23 on my monitoring software. Does this sound about right for Non-OC load temps?

 

[Ritwik7]

Are these non OC load temps when running Prime 95? If so, then it's ok. But if these are the idle temps, then I guess it would be a good idea to reapply the TIM.

 

[Scshadow]

Are these non OC load temps when running Prime 95? If so, then it's ok. But if these are the idle temps, then I guess it would be a good idea to reapply the TIM.

Non OC prime95 load temps, thats correct. Under v8 cooler with as5 applied via the dot method as recommended by AS5's website for my specific processor. And well um, I've still got my case open and laid down to swap ram modules so I don't know how much that makes a difference.

 

[Ritwik7]

Prime95 stresses the CPU a lot. So temps pushing 55-60c is not a problem at all.

I'm just a little confused as to the exact problem. Stress tests usually do detect a lot of errors that might not occur in a real life scenario. I suggest you OC the CPU in steps and try playing games on the system and see whether you're having any issues/glitches. Crysis is quite sensitive to unstable overclocks.

 

[dividebyzero]

First off...I wouldn't start using the machine for data storage until you get the stability issue sorted out. Dodgy memory usually goes hand in hand with data corruption. The first you're likely to notice is that the OS will likely end up corrupted and you'll need to reinstall.

Second...Populating 6 DIMM's with 4Gb sticks is problematic for a number of boards -including Asus. It requires a hefty jump in MCH voltage as a general rule (I think XMP profiles will default the QPI/DRAM v to ~1.25v. You may need to go as high as 1.35v). In addition the Ripjaw (RL) modules are compatible (as far as Asus' QVL goes) in 2 or 3 DIMM setup only. Whether this is because the other configurations haven't been tested, who knows?

Have you set the XMP profile in the AI Tweeker section of the BIOS?
Are you running the RAM at 1.65v (actual) as Supersmash suggested ?
If you're in any doubt about the BIOS settings (including the subtimings which I assume you've left at "Auto") I suggest you load CPU-Tweaker 1.5 just to confirm that the boards settings are commensurate with the embedded memory profiles.

The only sure way to check out the problem is to systematically test each module seperately, then once you have two ok modules (since they are dual channel kits), run Memtest86+ again. Repeat for each of the next two kits.
Even a pass for all three kits is no assurance that the RAM will play ball with the board since your RAM is not a matched 24Gb (or even 2 x 12Gb kits). You may well need to relax the timing to DDR3-1333 - even 12Gb RAM on some boards has trouble at DDR3-1600. 24Gb might be asking a lot.

 

[Scshadow]

First off...I wouldn't start using the machine for data storage until you get the stability issue sorted out. Dodgy memory usually goes hand in hand with data corruption. The first you're likely to notice is that the OS will likely end up corrupted and you'll need to reinstall.

Second...Populating 6 DIMM's with 4Gb sticks is problematic for a number of boards -including Asus. It requires a hefty jump in MCH voltage as a general rule (I think XMP profiles will default the QPI/DRAM v to ~1.25v. You may need to go as high as 1.35v). In addition the Ripjaw (RL) modules are compatible (as far as Asus' QVL goes) in 2 or 3 DIMM setup only. Whether this is because the other configurations haven't been tested, who knows?

Have you set the XMP profile in the AI Tweeker section of the BIOS?
Are you running the RAM at 1.65v (actual) as Supersmash suggested ?
If you're in any doubt about the BIOS settings (including the subtimings which I assume you've left at "Auto") I suggest you load CPU-Tweaker 1.5 just to confirm that the boards settings are commensurate with the embedded memory profiles.

The only sure way to check out the problem is to systematically test each module seperately, then once you have two ok modules (since they are dual channel kits), run Memtest86+ again. Repeat for each of the next two kits.
Even a pass for all three kits is no assurance that the RAM will play ball with the board since your RAM is not a matched 24Gb (or even 2 x 12Gb kits). You may well need to relax the timing to DDR3-1333 - even 12Gb RAM on some boards has trouble at DDR3-1600. 24Gb might be asking a lot.

Well, unfortunately, I hadn't ran any testing(which I thought I did when I built it). Still I don't have much data on this thing yet. If my OS goes corrupt, oh well.

The more and more I read, the more I realized I'm asking to get fairly lucky to run all my ram at once. Officially, it seems the x58 chipset specification doesn't support more then 1 stick of 1600 mhz(and presumably higher) per channel. I think intel and asus are basically saying, thats a fair amount of heat with modules just slightly not touching. Run at your own risk. Anyway, I found XMP profile late last night. I kept hearing that term tossed around but I hadn't actually found it.I was reading my mobo manual and I thought I had looked at every setting I could set in the bios. Oh well, I found it. And its all a learning process I suppose. I'm not quite sure why XMP is doing better then when I was manually setting it but if it works, I'm happy.