Bad processor?

[limer]

System Information:

Windows 7 Professional 64-bit
Corsair 400W PSU
Phenom II X4 955BE
Gigabyte GA-870A-UD3 (rev 2.1)
6 GB DDR3 @ 1333 MHz RAM (2 x 2GB A-DATA and 1x 2GB Samsung)
Powercolor Radeon HD 4830
WD 640 GB HDD (WD6400AAKS-00A7B0)

Machine passes a full pass of Memtest86+ 4.00 (as of October 14, 2010). Prime95 ran for 15 minutes without error (and a very loud fan). Core temp showed around 60 degrees Celsius max at any time.

I've been having strange issues with this newer build all along, but I'll go over the recent past.

At one point, the md5 of a file would change on every calculation (be different from what it was seconds before).

I downloaded the newer virtualbox, did an md5 using hashcheck (http://code.kliu.org/) and found it to be a mismatch from virtualbox's listed md5. I downloaded virtualbox again and also found it to be a mismatch. I then created an md5 file for the second virtualbox exe I downloaded and would recalc the results getting a different md5 every time.

If created an md5 file from another file on the same drive and it would match every calculation and so I suspected the drive might be an issue. I began testing the md5 of more files and the system BSOD with a 0x0000003B. I do not believe this is graphics related.

After a reboot, I'm unable to reproduce the md5 mismatches. Today, I completed Windows 7 updates and after a reboot, the network adapter showed: enable, cable disconnected and then no network connectivity. It would cycle through this until I rebooted. After a few more reboots and a good memory test later, it's no longer an issue.

The problems come and go and I can't make heads or tails of it.

Update (more background information):

When I originally built the machine, I forgot to plug in the 4-pin CPU power connector and it wouldn't POST. That was on the original motherboard (same model). I also ripped the CPU out of the ZIF socket without the arm up (locked) after I unlatched the heatsink/fan. I didn't use much force although the CPU was glued to heatsink and I didn't realize it. I inspected the pins very carefully and none appeared bent. This is part of the reason I wonder if I didn't somehow damage the CPU or memory controller within the CPU. This happened on the original, RMAed motherboard (exact same model).

I also sometimes have the system hang at post, but usually for five to ten seconds and then it successully moves on and into booting Windows. Sometimes it will hang and never move on beyond detecting the IDE drives. I have to power the systed down and on again. Something I notice that I don't believe used to appear during POST is:

cpuid:00100F43 Patch ID:00B6

 

[BTRY B 529th FA BN]

Memtest HCI

 

[lxskllr]

1 pass of memtest isn't enough. Let it run overnight. Sometimes subtle errors don't show up right away. I've gone over 4 passes before errors started occurring.

 

[limer]

Memtest HCI immediately revealed errors and so I took it down to the Samsung stick of RAM. I ran Memtest86+ 4.00 against this, also without errors. I then put it through HCI Memtest:

Later, I'll try the other sticks of RAM, one by one, and post the results. This is the second set of A-Data RAM I've received due to a previous RMA. I'd like to believe that it all comes down to the RAM, but this machine has had enough past issues to make me second guess a lot. I'll keep you posted.

 

[limer]

All three sticks pass HCI Memtest when installed by themselves. I will continue to test the ram in various combinations. Are we still convinced it's solely a ram issue?

 

[lxskllr]

Yea, I'd say it's definitely ram. Try boosting the Vdimm with all the sticks in, and see if you can get it to pass. You might want to loosen the timings a smidge also.

 

[limer]

You know, iirc, the A-Data RAM is rated for 1600 but runs at 1333. When I looked into the BIOS, it wasn't possible to up the voltage to 1.6V. I believe the autodetection set it at 1.5V @ 1333 MHz. I also seem to remember not being able to go above 1.55V without the BIOS stopping me.

Considering the Samsung is designed for 1333 @ 1.5V . . . how much do you think I should increment the voltage? I don't know if I can choose individual sticks.

I'll look into it and verify this with screenshots much later today.

 

[lxskllr]

I haven't really been up on the latest hardware tbh, but going from memory, DDR3 is more sensitive to overvolting than earlier standards were. I assume you set voltage control to manual before trying to change, right? If so, I'm a little surprised that it stopped you from overvolting. All the BIOSs I've used will happily allow you to burn up your machine using manual settings :^D

In any case, Google around for data on how much you can get away with overvolting DDR3. Overclocking sites would probably be a good place to search. For me just sitting here knowing nothing, I wouldn't go over .05V, so 1.55V total. You may be able to get away with more, but if nothing else, it'll be harder on your ram. Again, with me sitting here knowing nothing, I'd go to 1.55V, but reduce the speed, and maybe slacken the timings. Do one thing at a time, and test between attempts. If you change a bunch of things at once, you won't know what really worked, and it'll be harder to keep track of what you've done. It may come down to the ram just being incompatible with each other. If that's the case, I'd just use the 4gb until you got more money for new ram. Hell, 4gb might be enough for you. That's a pretty good amount for most uses now.

 

[allae]

Memtest HCI

 

[mfenn]

You know, iirc, the A-Data RAM is rated for 1600 but runs at 1333. When I looked into the BIOS, it wasn't possible to up the voltage to 1.6V. I believe the autodetection set it at 1.5V @ 1333 MHz. I also seem to remember not being able to go above 1.55V without the BIOS stopping me.

Considering the Samsung is designed for 1333 @ 1.5V . . . how much do you think I should increment the voltage? I don't know if I can choose individual sticks.

I'll look into it and verify this with screenshots much later today.

Some DDR3's rated voltage is 1.65V, so you should be safe at least to that number. Voltages above 1.5V are more of a concern for Nehalem's memory controller, and not for the memory itself.

 

[limer]

More information included in the OP for those interested. It's been a busy week at work but I'll do more testing today. Thanks for the responses.

 

[limer]

Both sticks of A-Data RAM at 1.6V and good so far. I'll keep the thread updated if I continue to have issues (which I suspect I might). I understand that the SPD tab of CPU-Z should reveal what the manufacturer stamps as conservative possibilities. It seems strange that both the A-Data RAM and Samsung RAM show exactly the same data and that it doesn't appear to be correct.

 

[limer]

Booted the machine this morning to a number of applications crashing and then taking the OS with it. A few reboots didn't fix this issue. Testing with both Memtest86+ 4.10 and 4.00, I see a large number of errors around 2300 to 2560 MB range. This is @ 1.6V at both 1333 MHz and 1600 MHz (which the ram is rated for).

I tried another Memtest86+ 4.10 this afternoon and had the same results around the same range. For the hell of it, I swapped the two sticks into each other's slots and now have no errors. I'm off to work and will be happy to try suggestions and other tests tomorrow.

 

[lxskllr]

If you have more problems, try 1.65V and see if that straightens it out.

 

[mfenn]

You could be seeing the slow death of your RAM. I personally wouldn't go above 1.65V for the memory controller's sake.

 

[limer]

I booted the machine with the RAM. 1333 MHz @ 1.66V (.02 increments) and timings at auto. Failed memtest around the same 2450 MB range. I then tried without any other drives or unnecessary devices and passed. Eventually got all the devices back to where they were and it continued to pass. Ran something like five tests where the last four all passed.

Thanks for the heads up mfenn. I don't plan on going above this voltage although the RAM is rated from 1.55V to 1.75V iirc.

I just ordered another 4 GB or Mushkin Blackline RAM from newegg and have a new Sempron 140 waiting to go in later if I need it for testing. Growing tired of the inconsistent issues but appreciate the help.

 

[degibson]

All three sticks pass HCI Memtest when installed by themselves. I will continue to test the ram in various combinations. Are we still convinced it's solely a ram issue?

Thread is tl;dr, but you could be looking at a bad slot rather than a bad DIMM. Try a known good stick by itself in each slot.

 

[limer]

Thread is tl;dr, but you could be looking at a bad slot rather than a bad DIMM. Try a known good stick by itself in each slot.

Part of the problem is that I don't know which stick is "good". Wouldn't a bad slot be consistently bad?

tl;dr?