Delid fail, help me troubleshoot?

[schmuckley]

get black glue..glue ihs on with gray TIM..RMA.
AS5 mebbe.Shin-Etsu..something like that.

 

[Tristor]

get black glue..glue ihs on with gray TIM..RMA.
AS5 mebbe.Shin-Etsu..something like that.

I am not going to RMA a CPU I modified. That would be exceedingly unethical. I take my own risks in the modifications I do knowing that I may destroy something. I do this hobby for fun, and just like any hobby it has a cost associated.

I still don't think it's the CPU. But I am going to try to find a Haswell to borrow to test the board though.

 

[DigDog]

I am not going to RMA a CPU I modified. That would be exceedingly unethical. I take my own risks in the modifications I do knowing that I may destroy something. I do this hobby for fun, and just like any hobby it has a cost associated.

I still don't think it's the CPU. But I am going to try to find a Haswell to borrow to test the board though.

good luck, i admire your integrity

 

[Tristor]

Sorry it took me so long to update this thread. The saga continues. I had submitted an RMA for the 2nd motherboard with Amazon, unfortunately while they sent me an RMA number to return it and acted like my "order" went through, instead of a package I got an automated email linking me to Asus support and a phone number for Amazon customer service if I had questions.

When I called in, they basically told me I was SOL until the first board arrived back, which it did on Friday. On Friday I also got paid and ordered a new CPU on the off-chance it was the CPU (I can always sell it to a coworker later if it turns out the board is for definite toast). I may end up keeping the new CPU anyway though, it was made in Malaysia where the old one was made in Costa Rica. It may be pure happenstance, but I seem to have better luck with recently produced Intel CPUs out of their Malaysia plant when overclocking.

Well, I just put the brand new, out of the box, i5-4670k into the 2nd board I have and it did the EXACT SAME THING. In other words, the board is bad, not my delidded CPU (or perhaps both were bad). I'm going to hang on to the second CPU for now and try to get Amazon to send me a 3rd board again because dealing with Asus support is probably going to be a nightmare. I'd almost rather eat another $200 for a new board and just trash this than call Asus.

So at this point, my desktop still doesn't work, so far it seems my delidded CPU is fine (both of them), and that I've had 2 DOA Asus Z87 Gryphons in a row.

Also, another hilarity occurred trying to get standoffs for my Corsair 650D. I took 2 out from when I had my original uATX board in it at my folks house. Unfortunately I left them there with some things I'd given to my dad he needed to build a computer for my grandmother (whole other story). So, here I am short 2 standoffs I need now. No additional standoffs came with the case, just what was in it. I have thousands of standoffs at home in both of the standard sizes, but unfortunately the 650D uses non-standard standoffs.

So far I've received an extra 650D accessory kit (no standoffs) and 2x accessory kits (including 1 standoff each that doesn't fit) from some other type of case. To ship these to me required 3 boxes + a bag per kit (kits in a plastic bag, then in a small box, then in a slightly bigger box sealed, then in a really big box wrapped in bubble wrap to fill in the space). I haven't had a pay a cent yet though, so I guess we'll see.


Here's the pictures I sent Corsair showing the standoff mis-size on what I just received today:

This is the standoff that came with the case.

This is the standoff they sent me.

So far this has been the most drama-filled computer build I've done in many years. In the interim to have something to at least game on I threw together a box to use from spare parts. AMD Phenom II x4 965BE, MSI 970A-G46, 2x8GB G-Skill DDR3-1600, GTX 670 4GB, and 2x Intel x25-M G2 80GB SSDs in RAID0. It's not terrible, but definitely a step down from my Ivy Bridge rig.

 

[DigDog]

You can say that again, i'd be ripping my hair out. Anyway, good luck & i am confident your chip will turn out to be a beast, I CAN FEEL IT!

 

[Tristor]

So just to eliminate any other silly things as possibilities last night, I did some further testing. Among that, I tested my power supply with a power supply tester that has a voltage check strip and ground check strip using a multimeter to verify that grounding through the power supply was working correctly and that voltage coming off everything were fine. As I expected it was fine since I had used this power supply for over a year previously without issue and it's of good manufacture. Additionally, I verified after reconnecting all power components to the motherboard that I was getting voltage coming through the pins on the ATX/EPS 8-pin 12V connector to the motherboard (e.g. it was making good contact in addition to the power supply working).

Based on all of the information thus far, it is more than apparent that the second board is for sure DOA. So I contacted Amazon on the phone again and talked with them, they happily did another return and expedited it so I should have a new mobo tomorrow to try. Hopefully the 3rd time is the charm.

 

[Tristor]

The third board has arrived, and IT WORKS. I've updated it to the latest BIOS and I've now also confirmed that my delidded i5-4670k works just fine. The one I haven't modified yet though is from Malaysia, so I'm thinking it may overclock better than the one from Costa Rica. I'm tempted to delid it as well and then compare and keep whichever is able to clock higher with good temps/low voltage.

I'm going to put everything back together and start doing benching/overclocking and find my sweet spot on the current delidded chip and see where I stand. I'll update the thread with results, but at this point we now have a success. Just in time for my next project which is to modify a GTX 780 with a custom CNC machined 110 copper front plate and a Corsair H80 AIO cooler as the GPU/VRM/VRAM cooler.

Looks good, and I'm glad I can prove the naysayers wrong. The vice method of delidding is still super safe.

 

[DigDog]

congratz.
oh btw, i'm also getting my new PSU tomorrow, and a nepton 280L. so i'll be doing some OC myself. i'll race you to 4.8

 

[Tristor]

I just now finished my Windows install + all the bs that goes with that. I'm doing an initial stability test by running Prime95 small ffts. My peak temp right now on Core 0 (hottest core) is 45C in RealTemp/CoreTemp with Core 1-3 absolutely equal in temp or at some points having a max 1C delta fluctuating, and Core 0 being 2-3C higher than the others. So it looks like I've got a nice mount (measuring with the digital caliper to get exact and even pressure + lapping didn't hurt I'm sure).

From looking around online it seems most people post delid with 25C ambient (I'm in 24C ambient) were getting around 66C peaks on NH-D14 push/pull (see http://www.overclock.net/t/1397764/delidded-4670k-results ), so I think 45C peak is pretty good after an hour run.

I am thinking this chip is going to be a beast, so I may not even bother with the Malaysia one.

EDIT: Did IBT using IDC's recommended settings, 5 cycles. Peak temps after were 48/47/47/45 (still at stock speeds). I'm not going to do any overclocking today, cause it's almost my bedtime (I'm a night shifter). When I get back up though I'll be doing overclocking Saturday night into Sunday and hopefully will get to 4.8, we'll see.

 

[JumBie]

Wow...reading this whole thread I was thinking to my self there is no way your receiving all of these DOA parts, you must be doing it wrong. Then you surprise me with your third mobo that actually works. Like wow, what are the friggen odds that you'd get 2 DOA boards? lol.

 

[Tristor]

Wow...reading this whole thread I was thinking to my self there is no way your receiving all of these DOA parts, you must be doing it wrong. Then you surprise me with your third mobo that actually works. Like wow, what are the friggen odds that you'd get 2 DOA boards? lol.

Higher than you might think. I (really we) don't know exact numbers for the 22nm process yields are, but we know that in a typical fab a "good" yield is around 70-80% of the wafer being usable, and roughly ~1200 chips per wafer at 22nm is likely. We also know that having a "bad" wafer represents potentially millions of dollars in losses. So every step along the way of the fabrication process the wafers are checked in a "sacrificial area" by a scanning electron microscope. This allows them to take images of the internals of the wafer as it's going through fabrication, but unfortunately this area is guaranteed to fail bin-testing as using a SEM on the die is destructive. Once the wafer reaches the end, individual dies are binned for speed and to verify that they work before being "packaged" into CPUs. If certain components fail validation, or it's only capable of a specific speed, then it gets binned into a SKU that it meets the requirements of. The idea is to waste as little of the total yield of the wafer.

Regardless, the gist is that EVERY CPU is tested before it leaves the manufacturer. Not so with motherboards. In the case of a typical mainstream motherboard, typically 1 or 2 units per batch (with a batch being somewhere between 1000-10000 units) will be tested at the end for batch validation as part of the QC process. Post-etch, most boards will be run through automated testing to verify that traces are cleanly etched and have good electrical connectivity, but once components are soldered on only those few for batch validation get tested. Another factor is that motherboards are made with 3rd-party sourced components on board and each of those components has their own "tolerance" rating. This tolerance rating describes a few different things about that component, among them the MTBF and percentage of components that are failed. While tolerances are typically very low (i.e. 1% or less), there's no guarantee that all components on a particular board are working when it leaves the factory.

Post-factory failure rates on CPUs are so abysmally slim that really the only way to get a DOA CPU is if it was mishandled in transport or installation. I'd say less than 0.001% of CPUs are "DOA". Motherboard DOA failures rates are probably around 3-5% in a typical scenario, but could be as high as 15-20% depending on components (remember that bad run of knock-off capacitors going around that took many a mobo/psu with it in the late 90s?). Remember that motherboards also could be mishandled during storage, transport, and installation.

Now, just to clarify if my opening wasn't clear enough. Numbers for the above are rarely if ever published publicly by the various manufacturers so my statements are guestimations based on my prior experience and discussions with people in the know, and most likely are not accurate although good enough for the purpose here.

From my previous postings in this thread you should note I found reviews on more than one retailers website saying that people had to RMA because the board was DOA or even RMA a 2nd or 3rd time. I do not think my experience was unique, although now that the board works I'm very happy with it so I imagine most people would have better luck.

This is also why I was completely comfortable taking my CPU out of its box and immediately delidding it within 10 minutes of my package arriving at my doorstep. If there's no obvious indications of mishandling upon a cursory inspection, you can pretty much guarantee that the CPU works. In fact, I'd say that's really /only/ true of the CPU. GPUs have pretty low DOA rates as well, and I have visually inspected then immediately modified (replaced cooler) GPUs before too for the same reasons, but their failure rates are still higher than CPUs. Anything else I'd test first (RAM, mobo, etc.).

 

[DigDog]

just to clear things up a bit:
http://forums.anandtech.com/showpost.php?p=3777198&postcount=1
TLDR: "malay and CR chips are identical. neither comes from EITHER malay or CR, but from somewhere else"

 

[Tristor]

just to clear things up a bit:
http://forums.anandtech.com/showpost.php?p=3777198&postcount=1
TLDR: "malay and CR chips are identical. neither comes from EITHER malay or CR, but from somewhere else"

Good to know, good read. Well, I'm probably going to keep this chip since it is so far stable at 4.4GHz@1.252v. I'm going to see what it takes to get 4.8 stable (if I can), but 4.4 is pretty good for 24/7 operation for me and a great start (+1GHz isn't shabby). I'd really like to see it go a bit higher. I plan to do more voltage tweaking after I get to a stable vCore since I have everything else set to Auto currently. Temps are great though, 54C peak after 10 minutes in Prime95.

EDIT: So it looks like I've hit a voltage wall at 4.4. To go beyond 4.4 requires exceeding 1.252v. I'm currently at 4.6@1.335v and not quite IBT stable, but close. Temps are now peaking around 68C, so still well within safety margins although I will probably stay at 4.4 for 24/7 after I finish finding my max stable OC. I haven't touched any other voltages yet though, so I may be able to reduce vCore and stabilize with other voltage changes.

EDIT2: I've started tweaking other voltages since I don't really want to exceed 1.35v vCore even though my peak temps are currently ~70C. I'm hoping I can gain stability by working on the vCache/cache multi since I left it at "auto" which is 8-34x, and it seems going up in cache multi/vCache may help stabilize my OC although I doubt to an extent that would allow a reduction in vCore.

EDIT3: So far so good, I've gotten through 2 passes of IBT without problems. I think I'm going to keep this as my 24/7 overclock, temps are 68C peak under IBT with the side panel off (I'm lacking intake right now in the chassis, just ordered 3x120mm Cougars + 2x 140mm fan grills + Mountain Mods 3x5.25 to 120mm fan adapter so I can put 3 120mm intakes in the front after taking out the 200mm that's there and put the 2 top 140mm fans outside the chassis so I can do 4x140mm [I already have the other 2] on the rad).

CPU-Z validation link


Ignore the max temps showing, that was before I pulled the side panel off. It's peaked 68C since the side panel is off and I've seen it as low as 66C under load. I'm thinking the additional fans will make a huge difference. Room ambient temp is 71F/22C btw.

Before I finalize it, I'm going to try to get the cache multi to be stable at 46 so the cache and core are the same speed and I'm going to re-enable EIST and C States so that it can downclock at idle and use less power. I'm using set voltage right now, not adaptive voltage.

Stock VID: 0.9999V - 1.0048V (1V)
Batch# 3339B845

Voltages are:
Type Stock OC
vCore 1.008V 1.350V
vCache 1.198V 1.300V
vCSA 0.872V 1.005V
vRIN 1.760V 1.920V
PCH Core 1.053V 1.053V
PCH VLX 1.500V 1.500V
DRAM 1.500V 1.535V
VTTDDR 0.750V 0.750V

CPU multi 46x
Cache multi 40x
BLCK 100MHz (99.98 reported)
DRAM DDR3-1866 CL9 2T
DRAM voltage is set to 1.500V but is displaying 1.535V. vCSA increased automatically.

I think I can get to 4.7 if I go to ~1.4V on vCore, but I'm not sure if I'm comfortable doing that even though I'm still within a good thermal envelope, after I finish my cooling improvements I may attempt 4.7/4.8 with more voltage. I think I can probably go safely to 1.45V on vCore without having thermal issues after that, but I'll definitely not go any higher than 1.35V for 24/7.

EDIT4: Well, apparently Windows 8 is kind of dumb and freaks out when you enable C-States. So I've reverted back to my prior settings. I may attempt to bump the BLCK from 100 to 100.2 just to compensate for the dip that's reported in CPU-Z (99.98 instead of 100.0). I'm not sure how much BLCK modifications will impact stability though and it's currently going through 20 rounds of IBT and is on round 4 with the settings I noted before after I reverted my changes.

EDIT4:

So people were asking for pictures:

Album of first part of the delid
post-delid cleanup
lapping IHS
Video of lapping at 220 grit
Applying CLU and test-mount prep
Applying CLU video Part #1
Applying CLU video Part #2
Applying CLU video Part #3


So about the custom mount. What I did was use the mounting kit from a Xigmatek Gaia 120mm tower cooler, modified the backplate with a dremel, then ran through 2 inch long M3 screws with the following configuration: SAE #4 flat, backplate, SAE #4 flat nylon, mobo, SAE #4 flat nylon, SAE #4 compressible paper, Xigmatek M3 thumb nut, 4xSAE#4 flat, 1x SAE #6 flat, waterblock, 1x SAE #6 flat, Xigmatek M3 top-nut.

By using a digital caliper with a pin to measure, you can ensure that you have EXACT even mounting pressure which makes this method safe for direct-to-die mounts also with some modifications. I measured exactly 31mm in length protruding from the top of the first thumb nut to the end of the M3 screw on all 4 corners indicating even pressure across the backplate. From there, you simply finger tighten to full-stop with the top nuts. The additional washers act as spacers. SAE #6 flats are around 1.2mm thick, SAE #4 are 0.9mm, and the nylon/paper are 0.8mm. By measuring the height of the socket, die/PCB, and IHS and the thickness of the motherboard you can approximate very closely the number of washers needed to space it out perfectly for a solid high-pressure mount.

 

[Galatian]

I didn't want to start a separate thread but I think I just killed my i5-3570K as well. I builds my own water cooling loop and was extremely dissatisfied with its performance (temperature only dropped 10 degrees) so I figured I just had to delid. At that time using the razor method seemed to be the safest, but when I finally had the IHS off I saw the cooper wires in a small spot and already figured I destroyed it. Well needless to say my PC doesn't work anymore. It turns on but screen remains black and the memory led on my ASUS board show a solid red (it doesn't have a CPU led so I guess it's the sign for a dead CPU).

Is there anything I can do or can I go right ahead buying a new processor. And if I do so is the vice and hammer method safer?

 

[Tristor]

I didn't want to start a separate thread but I think I just killed my i5-3570K as well. I builds my own water cooling loop and was extremely dissatisfied with its performance (temperature only dropped 10 degrees) so I figured I just had to delid. At that time using the razor method seemed to be the safest, but when I finally had the IHS off I saw the cooper wires in a small spot and already figured I destroyed it. Well needless to say my PC doesn't work anymore. It turns on but screen remains black and the memory led on my ASUS board show a solid red (it doesn't have a CPU led so I guess it's the sign for a dead CPU).

Is there anything I can do or can I go right ahead buying a new processor. And if I do so is the vice and hammer method safer?

Yeah, you're probably boned. Most of the people who've killed their CPUs delidding have done so because they've gone through the PCB and cut a trace inside. If you're seeing copper wire showing, that's probably not a good sign. You can try covering it over with a Sharpie or acrylic clear coat (think nail polish) and give it another go, but you're likely screwed. This is the very reason I chose the vise method because I was extremely wary of using a razor blade.

Good luck on your next one.

 

[Galatian]

Not sure though I have the balls to try it on another one ;-)

 

[aigomorla]

im sorry i didnt get to read the entire thread.. but did u fix your pins?

those bent pins can cause a TON of problems.

 

[Tristor]

@aigomorla: Yeah I fixed the pins, but unfortunately the board had already been fried because two of the pins were shorted against each other and I had not realized they were bent when I first powered it back on. Since then I went from an i5-3570k on Z77 to an i5-4670k on Z87 and my overclocking results are in my post above on this page if you want to skip over the rest of the thread.


EDIT:

I didn't want to zombie an old thread, but I got a bee in my bonnet tonight and decided to see what she could really do. I can do 4.7, no problem. 4.9/5.0 proved to be bootable but not benchable/validatable even at 1.44vCore/1.33vCache/2.2vRIN so I gave up rather than pushing voltages further, I think at 1.47vCore I might get it stable, but I don't really want to risk it.

http://hwbot.org/submission/2511047_totaledcow_xtu_core_i5_4670k_987_marks There's a valid result at 4.7
Here's my CPU-Z validation at 4.7 http://valid.canardpc.com/17fyed

4.7 was strictly /bench/ stable, not 24/7 stable at those voltages. There's no way I can get it 24/7 stable at 4.7 at anything approaching an acceptable long-term voltage.