Help with possible mem/board incompatbility GA-P55A-UD3 & Patriot

kghastie · Jan 26, 2011

- Gigabyte GA-P55A-UD3 (rev. 1) (BIOS v. F11)
- Patriot G Series ‘Sector 5’ Edition 4GB (2 x 2GB) 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10666)
- Intel i7-860 2.8GHz

I am on the most up to date BIOS and all drivers that I know of are updated (not that that matters for MemTest).

So this is my first build, and as it turns out I've been having a lot of instability, and am now failing MemTest regularly. I already RMA'ed one of my 2GB sticks ("Stick B") and I'm still having issues. So that is 3 sticks that each fail in both slots 1 & 3, so it seems there is something else going on.

I've tried MemTest with the Optimized and Failsafe BIOS settings, and it went like this:

Optimized BIOS - Stick C in Slot 1 & Stick A in Slot 3: FAIL
Failsafe BIOS - Stick C in Slot 1 & Stick A in Slot 3: FAIL
Optimized BIOS - Stick C in Slot 1: FAIL
Failsafe BIOS - Stick C in Slot 1: PASS for 25+ hrs

Here are CPU-Z results for 3 of the 4 configs:

CPU-Z Failsafe - 1 stick
CPU-Z Optimized - 1 stick
CPU-Z Failsafe - 2 sticks

I should note that all I want is a stable machine that runs at spec (no interest in overclocking at this point in time).

When I put BOTH sticks in, a lot of the settings change. For example:

2 sticks vs 1 stick
Mem frequency: 1066MHz/1333MHz
CPU Freq: 1999 MHz/2800MHz
CPU Clock Drive: 700Mv/900Mv
System Memory Multiplier (SPD): 8.0/10.0 (well, it's set to auto but I think that's what it gets set to)

And while with 1 stick I had timings showing 9 9 9 24 (which is how the RAM is labeled), with 2 in I get this:
Channel A: 8 8 8 20 4 19 48 1
Channel B: 7 7 7 20 4 19 60 1

Is this an indication of an issue? I have no idea what the implications of this are. Am I supposed to change those to 9 9 9 24 etc...? Why would the BIOS opt not to do that then, and have less stable settings in Failsafe mode?

I was told by (I think a Patriot employee) to do the following (but he is no longer responding to my questions):

The timings should be pretty straightforward. Use the native bclk, set only 1333, cas 9, trcd 9, trp 9, tras 24, trfc 72, and dram voltage 1.65-1.66v, qpi voltage 1.25-1.35v, cpu voltage 1.30v.

I can't see where to set QPI, and I'm a little afraid to guess at how to change the other ones. I think I see where to do it, but as a total noob to timings, etc..., I'm afraid I'll hose the entire thing.

Any ideas based on the info above, or where I should go from here? I'm not sure what is relevant and what is not, but I tried to pick out things that seemed like they would be relevant. I have BIOS screenshots and a few diffs of the above CPU-Z result files, if that helps.

Here are some of the questions I'm trying to answer:

1. Can I safely make the changes suggested above with both sticks in? Or should I try it with one stick first?

2. Why would the BIOS drop all the timings and voltages on me? Is it Gigabyte power-saving features or something?

3. Are there other settings that I need to modify to support the suggested changes?

4. I can't find a place to modify QPI timings - any approach there?

5. Will I be better off swapping the PGV34G1333ELK for memory that is more compatible with my board (is it "incompatible"?) I have no desires to overclock or anything - I just want stability without performance below what is expected from typical configs

6. Should I try installing the Gigabyte eXtreme software - will that change BIOS settings more safely for me (safely as in, more difficult for me to screw up)?

I could really just use a few steps to go from here and some verification that I can make these changes without worrying that I will screw something up bigtime.

Thanks in advance.

jaquith · Jan 26, 2011

Well according to the P/N PSD32G13332 is 1.5v and not 1.65v, and CAS 9. Also, according to CPUz DIMM 2 is 1333EL Series, and XMP.

Therefore, this is a case of mix-matched RAM ; http://dl.dropbox.com/u/17530430/post/cpu-z/cpuz-2011-01-25-Failsafe-C-in-slot-1-A-in-slot-3.txt So I would NOT place them in the same channel.

You can 'try' the RAM in slots DDR3_1 & DDR_2 which are the first 2 slots, and NOT a typical installation. Otherwise purchase a Matched Set, and I would recommend Corsair or G.SKILL per you MOBO per the Configuration Tools {RAM Mfg}, or Certified/QVL {MOBO Mfg}.

kghastie · Jan 26, 2011

jaquith :

Hm, so this is definitely confusing. I think the sticks are not being recognized correctly (by the BIOS?). They are supposed to be identical sticks, and have the same model number. The labels on both sticks say:

PGV34G1333ELK / 1333 / 1.65V / 9 9 9 24

Here's the newegg link: http://www.newegg.com/Product/Product.aspx?Item=N82E16820220435

Physically they look identical as well, and exactly like the newegg image (except one has a white on black sticker and one a black on white sticker). They definitely do not look at all like the PSD32G13332 pictured here.

However, a little background:

Originally I bought a 2x2 kit (sticks A & B). I had MemTest failures with those, but stick A ran successfully in slot 1 for 8 hrs, and in my inexperience I thought that was enough time to determine it was not faulty. That led me to believe that stick B was faulty, so I RMA'ed only stick B, and had it replaced by Patriot with something that is supposed to be identical.

What are the possibilities here? I would say Patriot could have made a mistake and sent a mismatch, but if it has the same model number and stats printed on the sticker, shouldn't that mean it's compatible an identical? It seems to me that the BIOS is identifying the RAM incorrectly (although I don't even know that that is possible). That may be why I've seen some posts about this RAM-type being incompatible with my board.

Once I get in front of the machine again, I'll try just stick A and see if the BIOS treats it the same way as it did with stick C alone, or if it indeed thinks these are two different types of memory.

So what seems most likely to blame? Incompatible mobo/RAM combo? Patriot mistake? Faulty memory? Faulty board? Bad BIOS settings?

jaquith · Jan 26, 2011

Problem in a nutshell, you ALWAYS need to RMA both sticks to keep them "paired." The sets are BIN Sorted into matching Sets of 2 or 4 in your case. Patriot simply sent you RAM with different and incompatible ICs.

No BIOS setting is going to address this problem.

Solution, send back both sticks and clearly state you want a replacement Match Set as you originally purchased.

Good Luck!

kghastie · Jan 27, 2011

OK, so even though it is the same model number with the same timings and voltage listed on the stick, they might not actually be a match, and this might cause them to show up in CPU-Z as a totally different model?

So there are differences within model numbers? Or are you saying it must be mislabeled? I mean it is pretty clearly not actually PSD32G13332, even though that is what CPU-Z is reporting. Can that mis-reporting be caused by mismatched memory? Just trying to get a little understanding on what is going on...

Thanks

clarkjd · Jan 27, 2011

The two sticks were obviously made at different times(different lot numbers)and could therefore not be EXACTLY the same design, even though the specs are the same.

jaquith · Jan 27, 2011

Take another look at the CPUz report that I linked above, and BIN sorting isn't as simple as selling sets of 2. Each stick is tested and even right off the assembly line there are variations. So you don't want to purchase sticks individually and put the in the same channel. Wiki BIN Sorting.

kghastie · Jan 28, 2011

Thanks, guys. I will be RMA'ing both sticks and am currently trying to convince Patriot to send me the replacements first.

In the meantime, I've been testing the one correct stick and it's been unstable with every setting I've tried except for fully Failsafe[strike] or Optimized plus DRAM at 1.65V and QPI at 1.31V[/strike], with all power-saving features turned off. At 1.24V QPI it was failing. [strike]I'm now on hour 15 of Memtest and hour 3 of Prime95 (large-FFT Torture Test) with QPI at 1.31V.[/strike] After about 20 hrs at QPI 1.31V, I got ONE red line in MemTest - that's all it takes to be deemed a failure, right? Now I am trying at DRAM 1.66 and QPI 1.41. It's gone for a few hours but obviously I need to run for a day at least before I can tell.

So Patriot claims the memory is recommended for that board but they recommend I run it at - DRAM 1.65V and QPI 1.4-1.5V, both of which are in the Intel warning zone and show up red in my BIOS. Is this safe? And will it continue to be ok when I have 4 sticks of this memory? Or am I better off just ebaying it?

I'm posting my CPU-Z results and my HWMonitor, both snapshotted during a P95 run at DRAM 1.64V and QPI 1.31V. Would anyone mind checking the results and letting me know if everything looks to within a normal range? Or should I post them somewhere else?

And what the heck do I do about this memory? I feel like I am pushing to the brink (since I am not an overclocker and just want to run stock stably), and I am not comfortable here. Is that just me being nervous or is this RAM a bigger headache than it should be?

Thanks to you both for your help and advice.

jaquith · Jan 30, 2011

Safe voltages, {recommend I run it at - DRAM 1.65V and QPI 1.4-1.5V} If you RAM is indeed 1.65v then I don't recommend the QPI/VTT/DRAM/IMC Voltage much above 1.35v and 'maybe' SUPER-MAX 1.40v, and 1.5v is too high.

Again, some people don't 'get' the Matching until I explain a little more: every stick coming off the assembly line has some slight behavior variation, CAS e.g. 6 means 6/1,000,000,000 of a second, the IC frequencies (Sine Wave) works best if it synchronized to each other verses 'node' http://en.wikipedia.org/wiki/Node_(physics) , or an analogy is tying 2 engines together that run at the same speed/torque/timings vs being different.

Them 'working' poorly together if at all make little sense when you're dealing with Atomic Clock speeds. Corsair and G.SKILL have the best testing and matching.

My Dad was hospitalized with CHF so I was away; he's okay now.

kghastie · Feb 8, 2011

Well, after much adjustment of the voltage as per your recommendations, I could get close but could never quite get consistent passes to MemTest.(

So I bought some Crucial memory, and am still having MemTest failures. But what I have noticed is that I pass and fail MemTest in very specific scenarios. (I have only just now noticed this pattern of behavior since I have been using the two sticks of Crucial, but that of course doesn't mean that it wasn't happening before with the other memory as well.)

So the interesting thing is that if I run Windows, use the computer for browsing/gaming, whatever, and I restart, I fail MemTest almost immediately (within a minute or two, maybe less).

If I then restart from MemTest and boot back into MemTest, it seems that I can run MemTest without failing indefinitely (I have done this for 12+ hrs several times).

I am not sure what other variables might be changing as well, but it does seem to be consistently reproducible in this way. I have not yet played around with the failure state to see if, for example, it only occurs if I run Windows for more than X minutes, or if I shut it down for an hour and then start it. I'm working on narrowing those down.

Does this point toward a specific config or piece of hardware that might be having the problem? Does this trigger anything for anyone?

(Hope your father is doing better, jaquith)

Thanks

kghastie · Feb 11, 2011

So I have done some further testing and have confirmed that if I am booted up (like into Windows) and shutdown for a few hours even, then boot into MemTest it will fail within a minute every time. If I restart from within MemTest I can't get it to fail at all. That has to be a clue, right?

kghastie · Feb 15, 2011

Anyone? I'm really at a loss for what to try next.

kghastie · Feb 27, 2011

So here are some more sequences that I've tested several times. Enough to be satisfied that I can cause Memtest to pass or fail consistently with any of the below below sequences. (Edit: I put the result first in each sequence for scanability, although I realize that's kind of confusing. So "FAIL" can be read as: "Memtest will FAIL if I do the following:")

1. FAIL: Boot into Windows. Shutdown. Boot into Memtest.
2. FAIL: Boot into Windows. Shutdown and leave over night. Boot into Memtest.
3. PASS: Boot into Windows. Shutdown. Shut off PSU. Boot into Memtest
4. PASS: Boot into Memtest. Restart or shutdown. Boot into Memtest.
5. PASS: Boot into Ubuntu off LiveCD. Restart. Boot into Memtest.
6. FAIL: Boot into Windows. Shutdown. Disconnect Hard Drive. Boot into Memtest.

So basically, any time I reboot FROM Memtest or Ubuntu INTO Memtest, I pass. Any time I reboot FROM Windows INTO Memtest, I fail, unless I have switched off the PSU.

Weird results to me but they are so consistent that I can't believe no one has any clue what I could even begin to look at. Or what I can rule out?

kghastie · Feb 28, 2011

It was not the PSU. Bought a quality Antec 550W and had the same behavior. Any suggestions besides RMA'ing the motherboard? Or is that the only thing left at this point...?

No one has anything?

And I expect to get hammered for asking this question, but is there any chance that these Memtest results are ignorable and will have no impact on actual performance of my machine? Like some kind of false positive?

clarkjd · Mar 1, 2011

kghastie :

Based on results of scenarios 2 & 3, I think there is some kind of residual charge from the motherboard that is causing the Ram to still "remember" what it was doing before trying memtest. Scenario 3 removes ANY residual charge, therefore nothing to remember. Scenario 2 the motherboard still has power to it even if it is shutdown, therefore residual charge. At this point, I would consider the motherboard as the chief suspect. Windows uses more memory to boot up than memtest or Ubuntu, Therefore more likely to fail going into memtest. At any rate really weird results :pt1cable:

PassMark · Nov 27, 2013

kghastie :

I know this is an super old post. But by chance we might have stumbled upon a solution to this today.

I have been trying to help out a MemTest86 user with the same problem as described above. Cold boot into MemTest86 is error free, but a warm reboot from Windows into MemTest86 causes RAM errors in high memory addresses.

The problem might be the NEC/Renesas USB3 controller. I am speculating that booting into Windows seems to activate the controller, probably with memory mapped I/O and then once activated it causes problems until the next power down.

So if anyone has this exact motherboard, with MemTest86 errors, can you try and turn off the on board USB3 controller in BIOS to see if that fixes the problem. Either post back here or in the other thread we have about the same Gigabyte P55A-UD3 problem.

kghastie · Nov 27, 2013

My God, sir, I got the email 30 seconds ago and I'm rebooting as soon as I hit send. Just gotta find that Memtest disk....

kghastie · Feb 9, 2015

Sorry to say that I did try turning off the USB3 controllers on the board in the BIOS, and it didn't seem to fix my problem.

[apologies for necro'ing this, but I didn't want someone to stumble upon this and think there was an easy solution]

Help with possible mem/board incompatbility GA-P55A-UD3 & Patriot

Distinguished

Glorious

Distinguished

Glorious

Distinguished

Splendid

Glorious

Distinguished

Glorious

Distinguished

Distinguished

Distinguished

Distinguished

Distinguished

Splendid

Distinguished

Distinguished

Distinguished

Share this page