Multi-Core Enhancement Causing Problems?

[Aikouka]

Ever since I upgraded to Haswell, I've been having intermittent BSODs with all of them being WHEA_UNCORRECTABLE_ERROR exceptions. I normally have no problem tackling BSODs, but this one was new territory for me, and what made it worse was that it happened so randomly that I could never try and reproduce it. In fact, it almost always happened when I wasn't even at the computer. It's happened about 5-6 times, and I've only been there for two of them. Only one occurred during an intensive application (a game).

So, I was messing around in the BIOS the other day, and I noticed that Multi-Core Enhancement was turned on. I had been using Sandy Bridge prior to Haswell, so this feature was new to me. Essentially, it was overclocking my CPU to 3.9GHz. I guess I'm wondering... could that have been causing my problems? I realize that 3.9GHz isn't exactly much of an overclock for an i7-4770k (stock is 3.5GHz). I've been talking with Intel about the error, and since I can't get the CPU to fail any tests (IBT, Prime95, Cinebench, Intel's diagnostic tool, etc.), I'm pretty sure they're going to give me the run around. :\

 

[inf64]

Try disabling that feature in BIOS and see how it goes.

 

[Ed1]

Ever since I upgraded to Haswell, I've been having intermittent BSODs with all of them being WHEA_UNCORRECTABLE_ERROR exceptions. I normally have no problem tackling BSODs, but this one was new territory for me, and what made it worse was that it happened so randomly that I could never try and reproduce it. In fact, it almost always happened when I wasn't even at the computer. It's happened about 5-6 times, and I've only been there for two of them. Only one occurred during an intensive application (a game).

So, I was messing around in the BIOS the other day, and I noticed that Multi-Core Enhancement was turned on. I had been using Sandy Bridge prior to Haswell, so this feature was new to me. Essentially, it was overclocking my CPU to 3.9GHz. I guess I'm wondering... could that have been causing my problems? I realize that 3.9GHz isn't exactly much of an overclock for an i7-4770k (stock is 3.5GHz). I've been talking with Intel about the error, and since I can't get the CPU to fail any tests (IBT, Prime95, Cinebench, Intel's diagnostic tool, etc.), I'm pretty sure they're going to give me the run around. :\

On , Asus P8Z77 if I remember right with Aitweaker set on auto even though the multi-core enhancement is enabled it only really works when you set XMP , gets works then as I had to disable it ( I like to set each core multiplier .

Are you sure it was really working ? run prime95 and you should see each core at 3.9 , w/o it enabled some cores would be 3.7 .
That is kind of odd so little OC would cause issue .

 

[cytg111]

have you inspected the dump file ?

 

[Aikouka]

Try disabling that feature in BIOS and see how it goes.

I did disable it, and it was working fine. However, this problem is so difficult to diagnose and test because it's so darned random. The time between the first and second BSOD was about three weeks. The shortest time between errors was about 5 days, so I've had a difficult time ensuring that any changes really help.

However, Intel's correspondence indicated that they wanted certain information, which they asserted was only on the CPU. So, I had to remove the CPU from my system, which given I have a custom water cooling loop, I don't enjoy going through the hassle of constantly removing it. So, I've temporarily placed an i5-4670k in the machine.

It will be interesting to see what happens because now I've run the following combinations:
i7-4770k + ASRock Z87 Extreme 4
i7-4770k + ASUS Maximus VI Formula
i5-4670k + ASUS Maximus VI Formula

I haven't had any issues with the latter yet (it's only been two days), but both i7 builds would BSOD.

On , Asus P8Z77 if I remember right with Aitweaker set on auto even though the multi-core enhancement is enabled it only really works when you set XMP , gets works then as I had to disable it ( I like to set each core multiplier .

Are you sure it was really working ? run prime95 and you should see each core at 3.9 , w/o it enabled some cores would be 3.7 .
That is kind of odd so little OC would cause issue .

I think it works (i.e. properly disables) on my Z87 regardless of XMP being on or off. I do have XMP turned on right now (DDR3-2400), and I just checked my PC, and HWiNFO reports the highest clock speed as ~3.7GHz. They're all currently at 3.6GHz.

Although, one thing I did find awkward in my discussions with Intel. They fed me the canned line of how overclocking your memory can void your warranty. This just blows my mind. Why does using Intel's XMP void your warranty?

But I agree about how it seems weird that such a small OC could cause the problem.

EDIT:

have you inspected the dump file ?

Yeah. The dumps specifically point out the CPU cache as the culprit.

Here's an example of one...

WHEA_UNCORRECTABLE_ERROR (124)
A fatal hardware error has occurred. Parameter 1 identifies the type of error
source that reported the error. Parameter 2 holds the address of the
WHEA_ERROR_RECORD structure that describes the error conditon.
Arguments:
Arg1: 0000000000000000, Machine Check Exception
Arg2: fffffa800d555028, Address of the WHEA_ERROR_RECORD structure.
Arg3: 00000000be000000, High order 32-bits of the MCi_STATUS value.
Arg4: 000000000100110a, Low order 32-bits of the MCi_STATUS value.

Debugging Details:
------------------


BUGCHECK_STR:  0x124_GenuineIntel

CUSTOMER_CRASH_COUNT:  1

DEFAULT_BUCKET_ID:  WIN8_DRIVER_FAULT

PROCESS_NAME:  System

CURRENT_IRQL:  f

STACK_TEXT:  
fffff802`e78b2a28 fffff802`e7a36965 : 00000000`00000124 00000000`00000000 fffffa80`0d555028 00000000`be000000 : nt!KeBugCheckEx
fffff802`e78b2a30 fffff802`e7bd5ef9 : 00000000`00000001 fffffa80`0cefeaa0 00000000`00000000 fffffa80`0d555028 : hal!HalBugCheckSystem+0xf9
fffff802`e78b2a70 fffff802`e7a36703 : 00000000`00000728 00000000`00000000 fffff802`e78b2bd0 fffffa80`0cefeaa0 : nt!WheaReportHwError+0x249
fffff802`e78b2ad0 fffff802`e7a36020 : 00000000`00000010 fffffa80`0cefeaa0 fffff802`e78b2c88 fffffa80`0cefeaa0 : hal!HalpMcaReportError+0x53
fffff802`e78b2c30 fffff802`e7a35f1b : fffffa80`0caf0b40 00000000`00000001 00000000`00000000 00000000`00000000 : hal!HalpMceHandlerCore+0xd4
fffff802`e78b2c80 fffff802`e7a35d78 : 00000000`00000008 00000000`00000001 00000000`00000000 00000000`00000000 : hal!HalpMceHandler+0xe3
fffff802`e78b2cc0 fffff802`e7a36f0f : fffffa80`0caf0b40 fffff802`e78b2ef0 00000000`00000000 00000000`00000000 : hal!HalpMceHandlerWithRendezvous+0xd4
fffff802`e78b2cf0 fffff802`e7acd77b : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : hal!HalHandleMcheck+0x40
fffff802`e78b2d20 fffff802`e7acd52e : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : nt!KxMcheckAbort+0x7b
fffff802`e78b2e60 fffff880`04a043a2 : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : nt!KiMcheckAbort+0x16e
fffff802`e78a6888 00000000`00000000 : 00000000`00000000 00000000`00000000 00000000`00000000 00000000`00000000 : intelppm!C1Halt+0x2

STACK_COMMAND:  kb
FOLLOWUP_NAME:  MachineOwner
MODULE_NAME: GenuineIntel
IMAGE_NAME:  GenuineIntel
DEBUG_FLR_IMAGE_TIMESTAMP:  0
FAILURE_BUCKET_ID:  0x124_GenuineIntel_PROCESSOR_CACHE
BUCKET_ID:  0x124_GenuineIntel_PROCESSOR_CACHE
Followup: MachineOwner

Note that the process name and bottom of the stack trace are usually never the same. It also makes me wonder if it might just be bad Windows 8 drivers on Intel's part. However, if that was the case, I would suspect that I would be seeing far more complaints.

 

[Termie]

If you've had crashes while the system is idle, it isn't the overclock. Are you running default voltage? No offset, right?

I believe I've had the same error you experienced when setting a negative offset on my 4670K that took the voltage down too low at idle (didn't have crashes at load).

If you don't have an offset, it could also be power saving. While that should of course work, you could try disabling it to see if the crashes go away.

 

[cytg111]

yea, wait and see with the i5-4670k .. right now i'd put my money on the 4770k being the culprit

 

[Ed1]

Yes, that looks like it might be low voltage thing .

Maybe try , reflashing bios to newest version .
load defaults with XMP only , make sure voltage on auto .

I think as far as Intel anything above 1866 is OC as far as memory goes , they probably even think multi-core enhancement is OC, which tech it is .

 

[Dave3000]

It could have been the multicore enhancement that could have caused the problem. Maybe you got a crappy 4770k and it can't handle running all cores at full turbo boost. Stock setting for 3.9GHz turbo on the 4770k is only when 2 cores are awake and active. Running all cores at 3.9GHz turbo is outside of spec for that CPU. Overclocking is the luck of the draw and even on the i5- 4670k that you just bought might not be able to handle all cores running at full turbo boost. On a note, I don't consider the stock turbo boost settings overclocking because, well, they are still stock settings.

 

[Aikouka]

If you've had crashes while the system is idle, it isn't the overclock. Are you running default voltage? No offset, right?

It's occurred during idle, under light use (browsing), and under heavier use (gaming).

I'm a pretty big noob when it comes to overclocking, so I've been a bit timid to touch anything other than XMP. I'll admit that I even had trouble finding XMP on my Maximus as it's under some weirdly-named option! So, I guess what I'm trying to say is that my voltages should be normal.

Here's an image showing all of the CPU-related information for as long as the i5-4670k has been active: http://i.imgur.com/UYMxAW9.png

Maybe try , reflashing bios to newest version .

I think I technically did that once when I flashed the latest BIOS on my ASUS board only after realizing that I already had it.

I think as far as Intel anything above 1866 is OC as far as memory goes , they probably even think multi-core enhancement is OC, which tech it is .

I definitely understand why Multi-Core Enhancement is considered bad because the motherboard is technically cheating by applying the turbo boost to every core instead of the proper subset. Although, if Intel doesn't want people using XMP, then why create it?

 

[BallaTheFeared]

Its 124, low vcore dead to rights.

101 and straight shut downs are almost always low uncore voltage.

Either bump the voltage or lower the clock speed.

 

[Dave3000]

It's occurred during idle, under light use (browsing), and under heavier use (gaming).

I'm a pretty big noob when it comes to overclocking, so I've been a bit timid to touch anything other than XMP. I'll admit that I even had trouble finding XMP on my Maximus as it's under some weirdly-named option! So, I guess what I'm trying to say is that my voltages should be normal.

Here's an image showing all of the CPU-related information for as long as the i5-4670k has been active: http://i.imgur.com/UYMxAW9.png




I think I technically did that once when I flashed the latest BIOS on my ASUS board only after realizing that I already had it.



I definitely understand why Multi-Core Enhancement is considered bad because the motherboard is technically cheating by applying the turbo boost to every core instead of the proper subset. Although, if Intel doesn't want people using XMP, then why create it?

XMP and Multicore Enhancement are 2 different things.

 

[Aikouka]

Its 124, low vcore dead to rights.

101 and straight shut downs are almost always low uncore voltage.

Either bump the voltage or lower the clock speed.

Are you talking about with or without Multi-Core Enhancement enabled? I turned it off a few days ago, but since these BSODs have happened up to weeks apart, I can't be terribly sure when I can say things are safe.

However, if we're talking about me running this CPU at what should be stock settings, I'm not sure why I should have to touch the voltages. Intel does test these CPUs a the proper voltages, right?

Anyway, I don't think the voltages in HWiNFO64 are even correct. I've got ASUS's AI Suite open, and while HWiNFO always reports values above 1V, AI Suite is showing me values going as low as .752V. Of course, that's when my cores appear to be doing just about nothing.

XMP and Multicore Enhancement are 2 different things.

I know. I just mean that one (XMP) is an Intel specification, so I don't see why the people who created the specification would have it void their own warranty.

 

[ShintaiDK]

Are you talking about with or without Multi-Core Enhancement enabled? I turned it off a few days ago, but since these BSODs have happened up to weeks apart, I can't be terribly sure when I can say things are safe.

However, if we're talking about me running this CPU at what should be stock settings, I'm not sure why I should have to touch the voltages. Intel does test these CPUs a the proper voltages, right?

Anyway, I don't think the voltages in HWiNFO64 are even correct. I've got ASUS's AI Suite open, and while HWiNFO always reports values above 1V, AI Suite is showing me values going as low as .752V. Of course, that's when my cores appear to be doing just about nothing.



I know. I just mean that one (XMP) is an Intel specification, so I don't see why the people who created the specification would have it void their own warranty.

You havent considered its your mobo/BIOS/memory thats the problem?

Your BSOD seems related to the Microsoft OS driver for Intel CPUs when it changed to C1 state.

 

[Dave3000]

I noticed that you are using DDR3-2400. That is way above the supported spec for Haswell. Haswell only supports up to DDR3-1600. It could be the the memory controller in your 4770k can't handle DDR3-2400.

 

[Aikouka]

You havent considered its your mobo/BIOS/memory thats the problem?

Your BSOD seems related to the Microsoft OS driver for Intel CPUs when it changed to C1 state.

I've changed out motherboards, memory modules, upgraded BIOSes and now changed CPUs. Except the newer Samsung 840 Pro SSD, there is literally nothing new in this PC that is left to change... well, and the second video card, but the first BSOD happened before I installed that. Also, there's Windows 8.

That's just one of them. It's never the same thing. Here's a post where I have all of them listed. To note, the Cinebench one was not with Cinebench actively benchmarking. I left the software open, and that occurred while idle. The only two that happened with me at the machine were the Firefox and Steam crashes. I was playing a game during the latter one.

I noticed that you are using DDR3-2400. That is way above the supported spec for Haswell. Haswell only supports up to DDR3-1600. It could be the the memory controller in your 4770k can't handle DDR3-2400.

It's happened whether I'm at base DDR3-1333 or DDR3-2400 speeds. I've tested that with the same DDR3-2400 memory at DDR3-1333 and normal DDR3-1333 memory that I ran for quite awhile in my Sandy Bridge system.

 

[BrightCandle]

Steam is a brilliant stability test, specifically left on the store looking at the widget at the top that goes through the games. Its presumably a flash program and what it does is make your CPU ramp up and down in a periodic fashion. There is every chance that firefox was doing the same thing depending on what you were looking at in it.

Given that there is probably a speed and voltage combination in there somewhere that isn't sufficient. I also suggest you drop the RAM to the stable supported settings while you determine CPU stability, you have to turn off all overclocking to verify components, you have gone to Intel too early and with an overclocked machine and they won't be able to help you if you run the components out of spec.

 

[blackened23]

I have used Multicore enhancement on many systems (with asus boards) without ever having issue, not once. Are you sure it isn't another issue such as memory, etc? Haswell is very very picky with memory speeds. Keep this in mind. Also ensure that your voltages are set correctly as others mentioned.

Multicore enh + stock settings should not crash.

 

[Aikouka]

Steam is a brilliant stability test, specifically left on the store looking at the widget at the top that goes through the games. Its presumably a flash program and what it does is make your CPU ramp up and down in a periodic fashion. There is every chance that firefox was doing the same thing depending on what you were looking at in it.

I can't remember what I was doing in Firefox, but I think I remember what happened with Steam. I wasn't actually looking at Steam, but I do remember that I was trying to access the Steam overlay while in game. For some reason, I couldn't get it to come up. Although, I might've accidentally hit CTRL+SHIFT+TAB instead of SHIFT+TAB. Well, it was pretty much right after that where it BSOD'd.

In a similar multimedia situation, I recall watching a recorded TV show via VLC, and suddenly, all I could do was move the mouse. Nothing would respond to clicks and I had the same tiny bit of audio feeding through constantly. It sat there for a bit but it never BSOD'd.

Given that there is probably a speed and voltage combination in there somewhere that isn't sufficient.

This is what I don't understand. Why should there be an issue when I should be at acceptable speeds with the proper voltages. I haven't changed any CPU settings other than disabling multi-core enhancement.

I also suggest you drop the RAM to the stable supported settings while you determine CPU stability

It's already crashed at stock DDR3-1333 speeds.

you have to turn off all overclocking to verify components, you have gone to Intel too early and with an overclocked machine and they won't be able to help you if you run the components out of spec.

Wouldn't I be getting memory errors if my CPU couldn't withstand the memory speeds? I've run MemTest86+ for multiple passes and it showed no problems.

Are you sure it isn't another issue such as memory, etc?

I doubt it. I've run with my newer memory and my older memory that was used for about a year in a rock-solid i5-2500k system. Both have tested fine in memory tests yet the system has BSOD'd with either installed and regardless of the memory speed (newer memory has been run at DDR3-1333 and DDR3-2400).

Haswell is very very picky with memory speeds. Keep this in mind.

I didn't get that impression from Anand's Haswell review:

Haswell got an updated memory controller that’s supposed to do a great job of running at very high frequencies. Corsair was kind enough to send over some of its Vengeance Pro memory with factory DDR3-2400 XMP profiles. I have to say, the experience was quite possibly the simplest memory overclocking I’ve ever encountered. Ivy Bridge was pretty decent at higher speeds, but Haswell is a different beast entirely.

That's actually one of the things that made me decide to try out faster memory speeds as I've wanted to do a bit more encoding.

Also ensure that your voltages are set correctly as others mentioned.

To reiterate, I'm rather confused as to why my voltages would be wrong in the first place. I'm pretty certain that they're all set to Auto in the BIOS on this ASUS board just as they were on the ASRock board. I would assume that if this was a software issue (BIOS setting bad voltages, driver failing, etc.) that there would be more people complaining about it. ...unless no one else is using Windows 8.

 

[Termie]

Can you confirm whether or not the system has crashed since you turned off MCE? I may have missed the answer in this thread.

 

[Aikouka]

Can you confirm whether or not the system has crashed since you turned off MCE? I may have missed the answer in this thread.

It hasn't yet. I think I turned it on about 5-6 days ago, but this past Monday, I swapped out the i7-4770k for the i5-4670k. So, the machine has only been active with the new processor for nearly two days now. It's hard to know whether I'm crash-free because of the wildly varying time between crashes.

 

[blackened23]

Anand doesn't paint a clear picture with regard to high speed DDR3 and the Haswell chip. It definitely is picky with RAM speeds once you hit turbo clocks or begin to overclock; JJ from Asus has a Z87 overclocking guide and he spent a good 10 minutes discussing this. And he has tested over 500 4770K chips along with varying types of memory.

-Basically, at stock speeds Haswell is good with 4 DIMM high speed DDR3.

-At turbo and overclock speeds, 4 DIMM high speed DDR3 will lower your OC or turbo potential. Think of it this way. If you have a high CPU overclock, you can't use high speed 4 DIMM DDR3, unless you get lucky. If you use high speed 4 DIMM DDR3, you can do so if you are fully running stock speeds (and I understand this is what you are doing.).

ON AVERAGE, there is a negative correlation between high speed DDR3 and Turbo/overclocked Haswell speeds. When one goes up, the potential for the other goes down. If you're aiming for a super high 4770K OC, don't aim for more than DDR3-1600. If you are using stock, then you can purchase DDR3-2400 and should be fine. That's the TL'DR of what JJ said in his presentation, and he (and the guys at asus) has tested probably 1000 4770K chips with tons of different memory brands over a course of many months.

That said, you mentioned that you are running stock. You also mentioned a crash at DDR3-1333 speeds. Maybe I missed it, but is your entire system running at stock? That means, GPU, RAM, everything? Also, is it a fresh Windows installation? Did you install all of the applicable Z87 drivers from the Asus CD? Are you using AI Suite? I've installed new motherboards and crashed continually unless I did a fresh Windows installation. I don't know what the cause is, but there are sometimes Windows is finicky with regard to being installed with old chipset drivers and will crash as a result. I've had that happen more than once over many years. Installing all of the chipset and motherboard drivers are a must, if you have not done this. A fresh windows install is preferable with a new motherboard as well.

 

[Aikouka]

I'm going to list my PC stats here because I think it might make things a bit easier...

CPU: i7-4700K
MBO: ASUS Maximus VI Formula
RAM: 16GB (2x8GB) G.Skill Trident X DDR3-2400 ( F3-2400C10D-16GTX )
GPU: 2x EVGA GeForce GTX 680 2GB (02G-P4-2680-KR)
PSU: Corsair AX860
SSD: Samsung 840 Pro 256GB
SSD: Crucial M4 256GB
HDD: 2x Hitachi 2TB (HDS723020BLA642)
ODD: Pioneer BD-RW (BDR-206)

-Basically, at stock speeds Haswell is good with 4 DIMM high speed DDR3.

-At turbo and overclock speeds, 4 DIMM high speed DDR3 will lower your OC or turbo potential. Think of it this way. If you have a high CPU overclock, you can't use high speed 4 DIMM DDR3, unless you get lucky. If you use high speed 4 DIMM DDR3, you can do so if you are fully running stock speeds (and I understand this is what you are doing.).

ON AVERAGE, there is a negative correlation between high speed DDR3 and Turbo/overclocked Haswell speeds. When one goes up, the potential for the other goes down. If you're aiming for a super high 4770K OC, don't aim for more than DDR3-1600. If you are using stock, then you can purchase DDR3-2400 and should be fine. That's the TL'DR of what JJ said in his presentation, and he (and the guys at asus) has tested probably 1000 4770K chips with tons of different memory brands over a course of many months.

I've heard things in the past about CPU overclocking hurting memory overclockability. I've also heard that filling all the DIMM slots can have a similar effect, and that's one of the reasons why I went with two DIMMs instead of four. It also lets me consider going up to the somewhat ludicrous 32GB of memory!

That said, you mentioned that you are running stock. You also mentioned a crash at DDR3-1333 speeds. Maybe I missed it, but is your entire system running at stock? That means, GPU, RAM, everything?

As of right now, everything is stock but the RAM, which has XMP enabled to set it to DDR3-2400 speeds. Thinking about it, it might be better to just turn that off to see if things are solid at the most basic levels before tinkering with an overclock -- automated (e.g. XMP, MCE, etc.) or manual.

Also, is it a fresh Windows installation?

Technically, no. It was fresh back with my first Z87 board (ASRock Z87 Extreme4) as I went from Windows 7 to Windows 8 on a new SSD. However, it was somewhere around 5 AM when I finally finished the build with the ASUS motherboard, and I was too tired to deal with Windows. However, I did get the fun of calling up the Windows registration hotline because it failed validation (as did Office 2013).

Did you install all of the applicable Z87 drivers from the Asus CD?

I grabbed the latest drivers from ASUS's site. As a side note, the one good thing about their site? The downloads are fast. It was painful trying to use ASRock's website with their 100-300KB/s downloads!

I did try installing the latest chipset drivers from Intel with the ASUS board, but it didn't seem like it actually updated anything.

Are you using AI Suite?

Yes, I have it installed, but it's normally not running.

I've installed new motherboards and crashed continually unless I did a fresh Windows installation.

As I mentioned, the original ASRock board was setup with a fresh copy of Windows, and it crashed.

A fresh windows install is preferable with a new motherboard as well.

I've been thinking about it and possibly just going back to Windows 7.

 

[Ferzerp]

The error he posted tells exactly what the culprit is.....

The processor cache had an uncorrectable ECC error. If running out of spec, run it in spec. If running in spec, and no microcode or bios updates are available, warranty the proc.

 

[Termie]

The error he posted tells exactly what the culprit is.....

The processor cache had an uncorrectable ECC error. If running out of spec, run it in spec. If running in spec, and no microcode or bios updates are available, warranty the proc.

And just to add to that, you can run the cache at a different speed than the core. My Asus board uses MCE and sets both the cache and core on my 4670K at 3.8. I wouldn't be surprised if cache at 3.8 (or 3.9 for the 4770K) would be unstable at stock voltages in a small percentage of samples. The OP might try keeping the cores at 3.9 but setting the cache at 3.5, which is stock for a 4770K, I believe.

I think that the Haswell refresh will have motherboard manufacturers rethinking MCE all together. Haswell is running closer to its limits and thus has less headroom than any line of processors since the Pentium IV.