Prime95: Worker #5 Stopped.

[2is]

Memtest is pure shit. I've had bad sticks pass well over 24 hours in memtest while prime blend errored in less than 5. Memtest may spit at error if the stick is jacked up enough that windows won't even load without a BSOD.

 

[2is]

Go into your bios, go to advanced mode, AI Tweaker tab. Change the Memory Frequency to 1600. Then, in the same tab, go down to DRAM Timing Control. Focus on the Primary Timings. From top to bottom they should read 9-10-9-27-2

This is slightly less aggressive that the XMP profile but its a good starting point to start testing at.

Secondary timings should be left at Auto

 

[Synomenon]

You should go tell Corsair that about memtest. They won't proceed with RAM rmas unless you get bad results with Memtest86+.

 

[2is]

They probably know and thats why they use it to minimize their RMA's lol. Anyway, try the settings I outlined in POST 27 and see if you have better luck.

Interestingly enough... The bad stick I had that would pass memtest for hours on end but failed Prime was a Corsair stick. I didn't bother with RMA at the time though. It was an old stick and I had overclocked/over volted it for years.

 

[Idontcare]

I'm convinced the problem is not with my RAM. What's the best way to test my CPU w/o actually damaging it if is not what's causing the Prime95 blend test failures?

Haven't read the rest of your thread yet, so apologies in advance if the following suggestion is irrelevant, but usually when you are trying to track down hardware faults you start intentionally underclocking the relevant stuff to see if the problem goes away. That way nothing is put at risk of damage during the failure-isolation phase of your debug.

So in this case, underclock your cpu and see if it improves. Keep doing more and more underclocking in steps (reducing the multiplier by 1 or 2 at a time).

If that doesn't make the error go away then go back and set it to stock and then work on the ram. Reduce the ram clocks one step at a time and see each time if it passes your stress test.

 

[Idontcare]

You should go tell Corsair that about memtest. They won't proceed with RAM rmas unless you get bad results with Memtest86+.

Glad I had GSkill then when I had a memory error that was not detectable with memtest86+ but was detectable with HCI memtest.

Ironically enough the way I found out my ram was suspect was because I couldn't get my rig to pass LinX when the CPU was at stock. Thought I had a bad cpu or mobo, but then I underclocked my ram and it suddenly worked.

So then I went looking for memory testers, because memtest86+ wasn't finding the ram issue, and came across HCI Memtest. HCI found my issue correctly and GSkill accepted the RMA on that basis. But who is to say that HCI finds all errors?

 

[2is]

You'll also want to do small FFT to test the CPU instead of the ram. Persoanlly I'd start with RAM trouble shooting for the simple fact that RAM is far more likely to be bad than a CPU, and the fact that the test you are getting errors in typically get errors when there's a faulty stick.

 

[sangyup81]

If you do not get errors with small FFT but get it with blend, your IMC is what is failing and I suggest some added voltage to that part of your CPU

 

[wsaenotsock]

Prime is more demanding that just memtest, so just add a tiny bit of voltage and see if that does it.

 

[Synomenon]

You'll also want to do small FFT to test the CPU instead of the ram. Persoanlly I'd start with RAM trouble shooting for the simple fact that RAM is far more likely to be bad than a CPU, and the fact that the test you are getting errors in typically get errors when there's a faulty stick.

If you do not get errors with small FFT but get it with blend, your IMC is what is failing and I suggest some added voltage to that part of your CPU

Ok, the small FFT test has been running for 10 hours now with no errors. I'll let it run for 2 more hours to see what happens. If there are still no errors when I stop the test, what should I do?

How do I raise voltage to the IMC?

 

[2is]

Ok, the small FFT test has been running for 10 hours now with no errors. I'll let it run for 2 more hours to see what happens. If there are still no errors when I stop the test, what should I do?

How do I raise voltage to the IMC?

You don't, I'm not sure what that post is talking about, nor does erroring out in blend mean your IMC is to blame. Personally I think the small FFT is a waste of time, doubt its your CPU. Once youre done though i think you should Make the changes to your memory speed/timings I outlined earlier. Ill stress this again, don't avoid troubleshooting. I know you're more than willing to run simple tests here and there, but you also need to actually make changes that can help correct the problem.

 

[Synomenon]

Ok, just changed the settings in the BIOS to what you said in post 27. Going to start blend testing again.

 

[Synomenon]

So the blend test has been running now for 12 hours and no errors!

Is the RAM bad since it can't run at the 1866MHz speed it is supposed to be able to run (without failing Prime95's blend test)? Or is this a problem with the motherboard not being able to handle two 8GB sticks running at 1866MHz?

My motherboard is an ASUS P8Z77I-Deluxe/WD.

Again, this is my RAM:
CMD16GX3M2A1866C9
Dominator Platinum - 1.5V 16GB (2x 8GB)
http://www.corsair.com/us/memory-by...hannel-ddr3-memory-kit-cmd16gx3m2a1866c9.html


What should I do now?

 

[2is]

I'm sure the board is fine. The memory may be ok too, more testing is needed though. I've run across a few sticks that needed a voltage bump when running at their most aggressive XMP settings, not ideal but not exactly uncommon either. Next step, once you're confidant enough that you aren't getting errors is to bump the speed back up to 1866 but leave the timings at were you left them earlier and retest. If no errors, I'd say you can consider the matter closed and start enjoying your new system. The timings selected are only slightly less aggressive than the XMP profile sets them to, but nothing that will make any real world difference at all.

 

[fixbsod]

Wondering if this is a Corsair thing...my RAM is very similar, Corsair 1866 but 4GB sticks. As prev noted I've had to run the RAM at 1600. It could very well be that the RAM wants more than 1.5v for 1866. I never had any issues with Memtest (tho never tried the 1 stick at a time) but am using 1600 with a little tighter timings than the 9-10-9-27-2 not sure exactly.

I havent had any issues / bsods since doing the 1600. I do have the cpu only slightly oc'd to do 3.9 on all cores (i7-2700k), otherwise normal -- uses speedstep and all that so not 3.9 all the time.

edit: actually looking back over past posts (joined to fix my bsod issues) and i think the 1866 -> 1600 fixed my hardlocks but the bsod was likely fixed updating realtek onboard lan and other drivers. however i did get some worker stops with 1866 as well.

no issues since 1600 and updated drivers and not gonna bother going to 1866

 

[tweakboy]

GO default stock clocks first. Let make sure nothing is wrong with your hardware.

Then you can OC and test with Prime. quick answer,, raise the CPU vcore voltage to get more stability. gl

 

[Synomenon]

I'm sure the board is fine. The memory may be ok too, more testing is needed though. I've run across a few sticks that needed a voltage bump when running at their most aggressive XMP settings, not ideal but not exactly uncommon either. Next step, once you're confidant enough that you aren't getting errors is to bump the speed back up to 1866 but leave the timings at were you left them earlier and retest. If no errors, I'd say you can consider the matter closed and start enjoying your new system. The timings selected are only slightly less aggressive than the XMP profile sets them to, but nothing that will make any real world difference at all.

Thanks 2is. I did let the test run an additional 5 hours so 17 hours total and no errors. I'll change the RAM back to 1866 and leave timings at 9-10-9-27-2, then run the blend test again.

GO default stock clocks first. Let make sure nothing is wrong with your hardware.

Then you can OC and test with Prime. quick answer,, raise the CPU vcore voltage to get more stability. gl

Is that a response to fixbsod or to me?

 

[2is]

GO default stock clocks first. Let make sure nothing is wrong with your hardware.

Then you can OC and test with Prime. quick answer,, raise the CPU vcore voltage to get more stability. gl

If by quick answer you mean wrong answer...

 

[BonzaiDuck]

Prime95 Blend test makes the memory work. I have seen situations when small FFT will pass "forever," but large FFT will fail after so many hours.

I thought someone told him to try looser timings. He could simply punch in the SPD or spec timings, or lower the speed. No need to fret over "XMP." He apparently has two 8GB or (2x8GB = 16GB) sticks.

At some known spec set of latency timings, he might bump up the RAM voltage a tad to see if it increases stability -- in increments.

I'd have to look up the processor (Intel) because I'm not familiar with i7-3770t . . . well, I just found the Intel spec page . . .

16GB either as 4x4 or 2x8 would take about five days to get 1000% coverage with HCI Memtest. But the RAM should work at stock latencies and voltage, without adjusting the IMC-related VCCIO.

You can try kicking up the VCCIO, but it shouldn't need more than 1.1 and 1.2V is about as high as you should probably take it.

Thing is -- the CPU, mobo and RAM should pass the RAM at 1000% using stock settings. If it isn't and some minor adjustment won't fix it, you may want to try some different RAM or get an RMA.

 

[Synomenon]

Ok, you guys saw my posts above asking how to change timings, etc.. What you're talking about now just goes right over my head.

Are you saying that the 17 hours of blend testing I ran wasn't necessary?

 

[BonzaiDuck]

Ok, you guys saw my posts above asking how to change timings, etc.. What you're talking about now just goes right over my head.

Are you saying that the 17 hours of blend testing I ran wasn't necessary?

I'm saying that you couldn't prove RAM stable without at least half the "HCI" thorough testing. That's because you have 16GB of RAM and it takes longer.

Do you have a spare RAM kit -- known reliable -- that you can test with? Have you tried testing one at a time?

You can do the Prime95 Blend testing as well, but I usually certify the RAM with a DOS-level Memtest like HCI, then test my overclocks with both smallFFT and largeFFT. In addition to HCI "thorough," I might be satisfied if these FFT tests will go ~12 to 14 hours each.

Then I'd run LinX at controlled "Maximum RAM" settings to see if the GFLOPS fell short of a known maximum (for my SB 2600K it's close to 130). I fine-tune the CPU voltage until I see least variation.

Did you say you'd tried two different Z77 boards? Kinda points you toward a RAM problem, doesn't it?

 

[Synomenon]

The problem with the first board turned out to be unrelated to RAM / memory problems. It was due to a corrupted Windows install.

I don't have spare RAM, but I did test each stick separately with Memtest86+ for at least 24 hours and then with a bootable version of HCI Memtest to 1000% coverage.

I'm not interested in overclocking at all. I just want a stable system that will run its components at the specs. their manufacturers say they should run.

If the system passed 17 hours of blend testing in Prime 95, isn't it good enough to run at stock settings and speeds for at least 17 hours a day?

 

[Synomenon]

Just passed the seven hour mark on this round of blend testing with the RAM at 1866MHz and timings at 9-10-9-27-2. No errors.

 

[2is]

Awesome, I'd run it for a few more hours just to be sure, if you pass you pass. You could, if you want, re-load the XMP profile and bump your dram voltage up, but I personally wouldn't. The slightly more aggressive timings with the XMP profile yield in zero tangible performance gains so not really worth running if you need a higher voltage to do so.

 

[Synomenon]

Actually, I started this blend test at 9:37am today (Sunday, December 16, 2012). It's now 11:45pm and still no errors. That's 14 hours. I think I'll stop the test now. That's good enough for me.

Thanks 2is and everyone else for guiding me through the troubleshooting!