IMC failure, should I be worried?

T

tzl99

n00b
Joined
Oct 25, 2007
Messages
49
Hi again and sorry for the long post,
Recently I asked about advantages of Broadwell-E vs Haswell-E. Well I went for a Broadwell-E CPU and received my parts yesterday. Main parts are i7-6850K, 64 GB (Corsair Vengeance LPX 4x16GB @DDR4 2400), Asus X99-a II, evga 850W P2 PSU, and using an old GTX760 just for testing. Everything went fine with assembly. It's currently sitting on a rubber mat on my son's desk, not in a case and just one drive is attached for testing. Yesterday afternoon I ran memtest86 for 8+ hours with the memory set to 2133 (Default) and it passed one cycle (it takes a LONG time to test). Then I setup XMP to run the memory at 2400 and ran overnight. This morning it was 75% through the 2nd run and had no errors.

So I installed Win 10 to test the system at defaults (other than memory at 2400) and stress it.

I have two issues that I'm wondering about. First, Win 10 reports 63.9 GB free. I figure this must be reserved by some piece of hardware but I'm not sure what piece? With 0.1 GB out of 64 missing, I'm not that concerned and I figure someday I will find out what it may be. Though if someone has some advice on where to start looking, I'm all ears.

The reason I'm concerned is that I then downloaded and ran Intel Diagnostic utility (latest v3.0.0.25) and ran it. It fails at the IMC test! So I Google around and find some suggestions that I tried below:
1. Disable search for a sample (engineering), since my chip is retail - IMC test fails anyway.
2. Reset memory to default 2133 MHz - IMC test fails again.
3. Load all bios defaults - IMC fails.
4. Update bios - I'm already on bios 0601 that I flashed last night. Board came with bios 0401, but still fails.
5. I then remove all but 1 DIMM of memory and I sequentially try all 4 DIMMS separately using 2 different memory sockets - everything fails. And by the way, Win 10 reports 15.9 GB when 1 DIMM is in place.

So I Google some more and come across links to similar failures using Broadwell-E processors and Win 10. Here is one link:
IPDT 6850k IMC fail?

This is somewhat reassuring. I run about 45 minutes each of the Intel extreme tuning utility CPU stress and memory stress without issues. Win 10 is not, so far throwing up any errors. CPU-Z reports my hardware correctly, shows I have 64 GB and that it is running with quad channels.

As a final test, I actually installed Win 8 (not 8.1) on my system on a separate drive (I removed the drive with Win 10). Guess what? System properties in Win 8 reports 64.0 GB of RAM, not 63.9, but Task Manager still reports 63.9 GB? That is strange, but Taskmanager does say 89 MB is reserved, though doesn't tell me by what.

I then run Intel Diagnostic utility and in Win 8 it passes all the tests, including IMC.

SO WHY AM I WORRIED? Two reasons. One is that when I was testing my system with a single DIMM, I found the system would not boot when I had the DIMM in the non-recommended slots according to the manual. It would give me a Q-code error of 53 (memory error). When I moved it the other slots, it ran fine. If I run a single DIMM, why does it matter where it goes? Is my memory faulty? Or motherboard? My memory is not on the QVL that I could see, but G.Skill Aegis memory with the 64 GB kit is. I'm tempted to buy it and return the Corsair. I'm buying low profile memory as I'm using a Noctua NH-D15 cooler.

The second reason is when I was examining the board last night I think there is a partially bent pin on the socket. My eyesight is not the greatest but I used a magnifying glass and then took a few pics (it's hard takin g sharp pics since I cannot find my tripod). In the first pic, I'm zoomed out, 2nd pic is cropped and unfortunately blurry. But you can see that the pin at the center of the circled hex is slightly off where it should be.

qkjPu
qkjPu


I know bent pins aren't covered under warranty, and since the system is working, passes a couple of passes of memtest, Win 8 seems to think nothing is wrong, the Intel diagnostic utility is fine in Win 8, maybe the pin is making enough contact. What do you all think? Should I replace the memory? I don't know why some sockets don't recognize the memory when I tested with single DIMMs? Or maybe I need a new motherboard (damn) since I have a bent pin?

Louie.
 
I passed the test with the CPU OC'ed and Memory @ 3000 , but... I don't have 64GB's worth... nor is my BW-E a 6850K.

Try very carefully bending the pin back in line with the others?
 

Attachments

  • snip.PNG
    snip.PNG
    66.8 KB · Views: 137
Suprfire, I wish I could try to bend it back. I have non-conductive tweezers for electronics, but the pins are so small and so close together that I really think I would probably bend a nearby one worse than slightly bend that pin back. The space between the pins is just 1 mm, so the pin is off by a fraction of a mm. Since you have a 6800K and you passed the test, it could be my MB, memory or compatibility of the components beyond the bent pin. I may try to get the G.Skill Aegis memory early next week which is on the QVL and pay a restocking to send the Corsair back, if the Aegis works. Also, I still think it funny that the same hardware passes the same test but in Windows 8, not Win 10.

Also, I notice you also have 0.1 GB missing in task manager, do you know what is using the memory?

Thanks,
Louie.
 
It probably just rounds the number down to the nearest tenth of a GB
 

Attachments

  • perfmon_2016-06-25_18-33-35.png
    perfmon_2016-06-25_18-33-35.png
    12.5 KB · Views: 102
Thanks,
In Win 10, I have 89 MB reserved. I just noticed that my Win 7 box has 33 MB reserved. But what is odd is that the WIn 7 machine with 16 GB in System properties says 16.0 GB of RAM, though task manager is 16351 KB (out of 16384). In Win 8, the new machine says 64.0 GB in System properties though task manager says 63.9 GB, with 87 MB reserved. Win 10 task manager says 89 MB reserved, and says I have 63.9GB. But only Win 10 in System properties agrees with Task Manager, in Win 8/7 they report different memories.

In any case, I'm not concerned about 0.1 GB missing, I'm more concerned about the IMC failure. Thanks for your response.

Louie.
 
Ok,
I've figured out the 0.1GB part at least. I do know that sometimes when large portions of memory are "reserved" it may be a hardware error, or a 32-bit OS with over 4GB of memory installed. Or perhaps, incorrectly seated memory, faulty memory controllers etc. But in regards to just 0.1 GB used, it appears to be small segments of memory reserved by the Bios and other MB resources. In Win 10, the GTX 760 is reserving about 67 MB all on its own. I found this by bringing up Device Manager. Then clicking on View --->Resources By Connection, then opening up the "memory" list below. The GTX 760 has 4 entries, 2 for video that take up 16777216 and 33554431 bytes, then 2 settings for the Audio taking up 16777216 and 16383 bytes. Multiple other devices are reserving smaller amounts, some just taking up 10s of KB. So that solves that mystery. Though again, Win 10 System properties reports what Task Manager does, but Win 7 and Win 8 system properties just list total RAM installed and don't agree with their task maangers.

So, the only thing to figure out at this time is if I should get new memory to see if the IMC failure is due to the Corsair memory with Win 10, or wait and see if a bios update may fix that. The system has been stressed for about 6 hours so far (3 hours of OCCT, 3 hours of Intel extreme tuning). In the middle, while being stressed I also compressed a 2-hour long 1080P video to fit on a cellphone for kicks, and nothing crashed. Tonight I'll run Memtest again overnight.

Louie.

P.S. The Asus bios is overclocking my processor even though I am running it stock. I thought the max boost of the 6850K was 3.8GHz but I just noticed it has been running at 4GHz all along while running a stress test and compressing video at the same time. Max temp so far has been 59C, in a room at 24C, with the Noctua with one fan installed and set in the bios for silence. The system is out in the open though, not in a case. I don't think this chip will OC well though, CPU-Z is reporting 1.308 V used to run @4GHz. I shudder to think what 4.3-4.4GHz would have taken.
 
You're fine. Honestly, you're looking in to this a little too hard; is it working properly? Have you had any issues outside of stress testing / that one utility? What is this system actually intended for -- something mission critical where you can't afford one reboot a week? A little more background would be helpful.

As for the reported voltage on your chip... automatic OCs set up with motherboard defaults tend to be pretty heavy on the voltage by default. Just because it's set to 1.3 vcore now doesn't mean it needs anywhere near that to stay stable.
 
Silent-circuit,
I think you're right. The system is not for any mission critical work. Gaming, browsing, running 1-2 low intensity VMs and on the weekends I compress camcorder videos to post for relatives over-seas. I overbuilt this system basically because things have moved so slowly on the Intel side that I realized I"ll keep it 5-6 years minimum and I wanted a system I won't feel a need to upgrade for a while. I've never bought HEDT before because "normally" I upgrade every 2-3 years. Really did I really need the 6850 over the 6000? Did I need 64GB, no its ludicrous. In the past I would have bought a 6700K, slapped in 32GB and called it a day. I overdid it but it mildly upsets my wife, so it's worth it ;-)

I've been testing this system since yesterday right after work and it hasn't missed a beat. In fact, had I installed Win 8/8.1, instead of Win 10, I would have never thought anything is wrong as Win 8 reports 64.0 GB of system ram (outside of Task manager) and the intel diagnostic utility passes all tests. It was just that in Win 10 it fails the IMC test and it reports 63.9 GB installed but after looking into that, that is not an issue either.

So far it has run Memtest86 for 10-11 hours, 3-4 hours of Intel Extreme tuning utility and 4 hours of OCCT. I've run OCCT in the primary mode, that I think uses AVX and the Linpack mode with 90% memory usage (my system got up to 58 GB used LOL). The last 4 hours I've run it for fun at 4.2 GHz and it has been stressed continuously. Funny enough, about voltages, when it ran at 4.0 GHz with default bios settings it ran at 1.308V according to CPU-Z (don't know how accurate this is). So I set the multiplier in the BIOS to 42, didn't change any voltage settings, and guess what? It ran OCCT and Intel Extreme Tuning at 4.2 GHz also at 1.308V. With AVX (I think this is the primary setting for OCCT) because my temps went as high as 79-80C, whereas in IET it never got above 72C. I did install the 2nd fan for the Noctua and set it as "turbo" where it gets to max fan speed at 80C. For the first time today I finally heard the fans. As an aside, that Noctua is really silent! I think it is better than my Silver Arrow in my primary, now secondary, system.

Tonight I'll run Memtest for another 12 hours, if it passes, then I'm installing the parts in the case tomorrow and install my apps. I think I'm done. The IMC failure just shocked me and I was looking for trouble where there might be none.

Thanks for the reassurance!
Louie.
 
Hi,
I did find those documents and another called "core-i7-lga2011-3-tmsdg.pdf" that has the numbering scheme used to identify the pins. However, I'm having a hard time with the blurry photo finding the exact pin. The best I can come up with is the F34 pin, and it is a memory channel. It is listed as DDR3_DQ[12]. So I'm assuming it is involved in data channel 3? (from 0-3). I'm not sure how to isolate that.

I tried installing a single DIMM in all 8 sockets and I find that the system boots only if sockets x1, not x2 are used first. That is I can boot up with a single DIMM in sockets A1, B1, C1 or D1, but not A2, B2, C2 or D2. I just get a memory error (code 53) with the x2 sockets. With a single DIMM in each of the 4 sockets, Win 10 shows 15.9 GB free (though I now know where the 89 MB is), and the IMC test fails in Win 10. Again, in Windows 8 the IMC test passes. A poster above me with a 6800K but different memory and a different motherboard does pass in Win 10. But in the Win 10 community forums others with Broadwell-E processors fail the IMC test in Win 10 and an Intel rep is saying to wait for a software update. Looking more on Google, I've seen people with i5-6600K and i7-6700K CPUs that reported IMC failures on Asus motherboards last year, then corrected with a bios update.

In the meantime, the system has now passed just under 24 hours of memtest86 without an error and spent nearly 8 hours stressed yesterday. Today I'm installing Win 10 officially, with activation. The pin is off by a fraction of a mm and maybe the landing pad is wide enough to still make contact? Perhaps that's why it is working. I guess if the pin were completely bent or broken, that would be another story. It was nice of you to look that up though, thanks.

Louie.
 
Im not positive you pin is even bent honestly...But people have recommended using mechanical pencils to straighten them out when needed ....Anyways this is a "tool" i would use if i had bent pins. Granted i guess the task is harder for you guys with the higher pin counts.
 
One of the intel guys said this is an issue with broadwell and their current version of the intel cpu test. As long as you aren't getting memory errors you are fine.
 
Yeah, I read that soon after I built the system and was looking for reasons for the failure. The system has been up and running for exactly 2 weeks now and it's been fine. I've let memtest run overnight for 3 nights and there were no problems. So I think it is ok. Not much of an overclocker though. I'm mainly running at stock (boosts to 4GHz) for now and for another 2-3 weeks to be sure there are no issues. I did run it at 4.2GHz for 2 days but it took about 1.308V. I'm pretty sure I'll hit a wall at either 4.3 or 4.4Ghz (I'm aiming for 1.35V). Still it compresses video so quickly. My i7-2600K @4.7 was no slouch but I think generational improvements and the two extra cores have cut my compression times in half. Thanks for the response.
Louie.
 
Sounds spotty, especially if you're passing everything else. Amazing system by the way.
 
tzl99 said:
Yeah, I read that soon after I built the system and was looking for reasons for the failure. The system has been up and running for exactly 2 weeks now and it's been fine. I've let memtest run overnight for 3 nights and there were no problems. So I think it is ok. Not much of an overclocker though. I'm mainly running at stock (boosts to 4GHz) for now and for another 2-3 weeks to be sure there are no issues. I did run it at 4.2GHz for 2 days but it took about 1.308V. I'm pretty sure I'll hit a wall at either 4.3 or 4.4Ghz (I'm aiming for 1.35V). Still it compresses video so quickly. My i7-2600K @4.7 was no slouch but I think generational improvements and the two extra cores have cut my compression times in half. Thanks for the response.
Louie.
Click to expand...

memtest86 will only find manufacture problems (faulty ram) with your sticks (which is very rare). HCI memtest or google stresstestapp (in linux) will stress the memory to show if there is a problem with IMC and memory overclock/timings/etc. Note though with HCI memtest it will also stress your cache if it is overclocked so even if IMC and memory is fine it could show up as errors due to your CPU uncore instability.
 
Lee,
Thanks for the info. It will come in handy if I get any errors or crashes. So far I'm not overclocking (other than the default 4GHz the Asus bios sets) as I want to run my system stock for a month or more before delving in. I only tested 4.2 GHz for a day or two to check temps with my HSF. I think the IMC error is a software issue with Broadwell-E and Windows 10. When I came across this error, I temporarily installed Windows 8 on a 2nd drive and ran the Intel diagnostic utility and the computer passed all tests including IMC in Windows 8 with the same hardware.

Louie.
 
You must log in or register to reply here.
Back
Top